👉│feedback-general

[mossy tulip]

We would appreciate to get some feedback on Perplexity Ask. please share it here, thank you in advance!

[onyx saffron]

found 1 failure search after quite a few tries

otherwise impressive!

[rose creek]

The engine is susceptible to prompt injection.

it can also leak its prompt

[mossy tulip]

yeah, it is easy to break the prompt, but we are not too worried about this, it is not the hard part

even chatGPT is susceptible for these kind of attacks

[rose creek]

Alright then

[mossy tulip]

thanks for the feedback though

[rose creek]

You're welcome

[hollow tendon]

This is genius! It works great. I have some slight feedback regarding the mobile experience. ’’’html
<meta name="theme-color" content="#ffffff"/>
’’’
It would make it look like a complete app when using it as PWA.

Also, the links need to match the width of their parents so they won't scroll horizontally

[mossy tulip]

thanks for catching the bug!

[grim bridge]

The discord bot seems to be stuck when I ask for the difference between GPT-J and GPT-3.
It displayed an error message (API error 502) after five minutes of being stuck in the state "is thinking".

[sour timber]

Great idea, would companies with large advertising budgets on Bing be more likely to be mentioned in the answers? I assume they would, so it's like a product placement advantage in the response.

[solemn venture]

Thanks for creating this product - love it

[manic scarab]

clean interface. are you crawling and embedding the bing result content? or does bing offer enough context in it's own results to feed right into the gpt prompt?

Cohere did something similar but w/ the google Serp api .. https://txt.cohere.ai/building-a-search-based-discord-bot-with-language-models/

[pulsar mist]

This is a good observation. Yeah, we currently inherit the ranking algorithm directly from Bing. We are exploring the ability to curate (expand or trim) the citations that are included in the summary, which is one way to put the ranking power back into the hands of users. (see #💬│general message)

[pulsar mist]

We are just using the bing API and using the snippets they return right now. Yeah we tried Serp but the latencies are much higher than the Bing API so we stuck w/ Bing for now.

[elfin dirge]

Dark mode would be great

[slender berry]

So it cites the answer 5+5 with article about 5/5

[mossy tulip]

thanks for reporting

[thorny patio]

The two main pitfalls of such an approach has been:

a) Non-modeling of the abstention class (Knowing when NOT to respond)
b) Confidence score associated with the answers.
Any plans to address these?
Some of the queries have returned downright gore.

[mossy tulip]

thanks for sharing, we are working on adding a way for users to control the sources. adding confidence scores is cool as well, we will look into it.

[raw silo]

nice app

[velvet socket]

I really like your site, but waiting for 5-6 seconds for a query to finish is really an eternity when compared with google

would it be feasible to make the answer appear as gpt generates them (in the simulated typing way like the ChatGPT UI does) so it doesn't feel quite as slow and one could already start reading the answer before the full query terminated

[cunning goblet]

@velvet socket agreed. We are working on making the answers streaming as in chatGPT. Thanks for your feedback

We’re also trying to improve the UI such that the links appear immediately while the summary is streamed.

[potent cave]

A question I tend to ask to the AI search tools like Lexii.ai and others. It seems like it is generating different results depending on slight change of the question and both are incorrect here.

Has worked nicely for many other questions though

[celest ravine]

it seems one cannot view the welcome channel history by default.

[frank aurora]

Please add AI regeneration

[mossy tulip]

working on it

[sour timber]

Discord - A New Way to Chat with Friends...

[pulsar mist]

@velvet socket see #💬│general message

[tiny field]

Cool thing guys. Who is behind it all?

[full venture]

We are working on an about page, but here is our twitter:
https://twitter.com/perplexity_ai
https://twitter.com/denisyarats/status/1600561194875908097

Denis Yarats (@denisyarats)

@perplexity_ai We are a small team (@AravSrinivas, @randomjohnnyh, @NickAJalbert, @_carjun, and @andykonwinski) who build this fun prototype. Would be happy to answer your questions and receive feedback!

Twitter · Dec 7, 2022, 6:41 PM

[desert grove]

Yeah I wasn't able to show anything till I posted in introductions

[desert grove]

This the coolest thing since sliced bread.

[brittle basin]

The Bird SQL function works great but I think the code section is unnecessary. YouTube has a free and complex API so I hope you will create something like this for youtube too. I see a lot of potential in this project fr

[short summit]

There’s a bug with the search engine where it will overflow after ~2 lines on mobile and clip

This is for the result text btw

Actually it depends, sometimes it works and doesn’t

[cunning goblet]

@short summit could you share a screenshot or a permalink?

[short summit]

ya hold n

https://www.perplexity.ai/?uuid=8ea2785e-a13f-4503-8778-18ca71a2d6e9

@cunning goblet

it's also not centered anymore

did you do word-wrap?

that might be why

word-wrap in css is fucky

i had a similar issue. if a phrase doesn't end, it doesn't know where to wrap it and gives up

iirc there's a different kind of wrapping you can use, but I would set a js hard character limit and use a "-" character after cutoff for split phrases

just an idea

yeah

if I add a space it starts to try to fix itself

see now it works

if I remove the spaces though it breaks again

@cunning goblet this is probably the fix you're looking for

tbh granted by how old this stackoverflow answer is, this is probably fully supported + ie is dead anyway

here ya go

that works

hope this helps your team out

the actual css btw:

 -ms-word-break: break-all;
 word-break: break-all;

 /* Non standard for webkit */
 word-break: break-word;

 -webkit-hyphens: auto;
 -moz-hyphens: auto;
 -ms-hyphens: auto;
 hyphens: auto;

[vast mural]

@cunning goblet backend of Bird SQL crashed ? all queries are slow and no result

[heady plover]

@vast mural down for a few minutes while we fix a bug

@short summit thanks for the debugging! we'll address!

[short summit]

np!

the literal exact same issue has happened to me before in a react project LOL

[vast mural]

ok, the update pace of this product is incredible fast.👍

[heady plover]

We are back up now!

[heady plover]

@short summit we just pushed an update to fix the broken formatting you were seeing. Let us know if you run into other issues!

[short summit]

hell yeah!

i will, and I'll make sure to lyk fixes if I know any

[short summit]

it works now!

glad I could help y'all out, this is probably going to be my go-to search engine now for any questions I have in terms of programming LOL

[grave swallow]

@potent tendon: explain british history in details witout missing a word:
British history is a vast and complex topic, spanning from the Roman invasion of Britain in 43 AD to the present day[1]. It includes many terms and names that are unique to British English[2], such as the Lost Colony of Roanoke[3] and words without vowels such as myrrh, pygmy, shyly and crosswordsolver[4][5][6].

[potent tendon]

It would be better if the answer where not soo short

[placid topaz]

Hi, Does Perplexity have an API?

[heady plover]

hello - many of the answers I'm getting are not accurate and seem to be the direct output of GPT3, then references are added on top, is that correct? thanks

[urban gull]

I suggest to add related questions that would be clickable and would be the input for a perplexity query in a new tab

What is Russia's Plan to face EU Sanctions?

In response to EU sanctions, Russia has taken steps to prepare for this moment since 2014[1]. This includes rethinking their aim of a 20% share of the global LNG market[2], as well as calling the EU sanctions "unlawful" and adopted under a "far-fetched pretext"[3]. The Biden administration has stated that the sanctions are meant to segment Russia from the global economy and punish Russian elites[4], with a new package of sanctions hitting a range of industries aiming to stifle technological enhancement[5]. However, these sanctions do not impact food security and cover only bilateral trade between the EU and Russia, not international trade[6].

List of topics in the answer:

1. Re-thinking 20% share of global LNG market.
2. Calling EU sanctions "unlawful".
3. Biden administration sanctions.
4. Impact on food security.
5. Bilateral trade between EU and Russia.
6. International trade.

List of questions about those topics:

1. What is Russia's strategy for achieving a 20% share of the global LNG market?
2. What is the basis for Russia calling the EU sanctions "unlawful"?
3. What industries are targeted by the Biden administration's new sanctions?
4. Are there any exceptions to the sanctions that would allow food security to remain intact?
5. How does the bilateral trade between the EU and Russia differ from international trade?
6. How do the sanctions impact international trade?

they could be placed in a <details><summary>Related questions</summary>ul li a </details> just above citations

[mossy tulip]

yeah, good idea, working on it

[full venture]

hi, thanks for the feedback! We are working on exactly that 🙂

[full venture]

hi, please dm me examples. Our number one priority is working on getting more accurate and complete information.

[outer zodiac]

Idk what's happening but no matter what I ask bird sql, I get blank results like this

[full venture]

hi Sanket, please check your dms, will help you debug your issue

[potent tendon]

You would make a new channel, where you post the stuff you are working now, and will be adding in future.

[full venture]

got it, we are new to Discord, will try to show some previews soon. Thanks!

[sharp estuary]

I found that it doesn't work with FireFox's enhanced tracking protection in case you have that enabled

[outer zodiac]

Yeah, that ended up being the problem. Thanks!

[sharp estuary]

re: "we will work on more ways of accessing Perplexity e.g. apps and extensions!" (#💬│general)
I think it'll be nice if it could be queried through a query parameter so that it can be added as a search engine to browsers. e.g. https://perplexity.ai/search?q=example+query

oh wait

that already works!

awesome

[acoustic sparrow]

So I've been using this a bunch and both use cases are cool. I like how for Ask Anything you're able to merge results from OpenAI and search engines into succinct responses w/ citations. For Bird SQL it's just actually useful Twitter search and I like how we can vet the SQL ourselves. Realistically, though, is there anything preventing Twitter from just implementing GPT-enabled search themselves? Wondering what is unique or defensible about the approach here.

[full venture]

hi Alex, very interesting thoughts and valid points. These products are meant to be demos of our search technology for now (not complete products). We are focused on building awareness of our search technology and building community for now. Meanwhile we hope these demos can help save you time and do things no other service can, and future launches will only be better. Thanks for the feedback!

[keen loom]

"merge results from OpenAI and search engines into succinct responses w/ citations" . I looked this through and it seems like it is just using tldr(https://beta.openai.com/examples/default-tldr-summary) on the first 4 results from the Bing API. Please correct me if I am wrong.

[full venture]

Not quite that simple, but the general algorithm and pipeline is correct. We are working on making an interface for search that can do much more. Thanks for using and let me know if you have any other questions!

[blazing crag]

Used the product for about 5 minutes. If Google doesn’t respond in fairly short order, this could be a serious competitor. It’ll take a LOT of work but it’s definitely there.

I’m sure I’m not telling you anything you don’t already know but one Internet rando thinks you’re on the right track fwiw.

Hell I hope you can sell to Siri because she’s dumb as a box of rocks. And I don’t trust Google or Amazon. So you’ve got a lot of great exit possibilities too.

[full venture]

Thank you! Agree that it will take a lot of work. We hope to at least gain traction for some users and help them save time and effort!

[blazing crag]

Oh and if I can offer one request… I mean you’re going to do what you have to but… if there’s any way you can avoid the trap of surveillance capitalism where I’m the product, not the customer I think it would be a step forward. This business model is kind of literally ruining the world and eroding our sense of personal freedom, imho.

I suspect the model based approach of these answer engines (where the entire model can sit on my iPhone) can actually re-empower individuals to own their own compute. So incorporating some kind of subscription model to be paid with crypto or whatever would be really hard and add extra risk, I know, but it might even give you a competitive advantage at some point because basically everyone is creeped out by Google, whether they want to admit it or not.

This kind of sweeping, wholly different approach could be as revolutionary as Google itself, maybe moreso. It’s more work but if it’s successful it’s totally worth it.

Good luck!

[full venture]

Right, good feedback. I think we are onto something revolutionary as well.

[junior wagon]

#1051243890105143377 message can I look at what it tried in order to get no results?

(I repeated the query in #1051243890105143377 after doing it via the web interface so I could link to it in my last line. Oh, maybe I'm supposed to just paste https://www.perplexity.ai/sql?uuid=9c130589-943e-46df-9dc4-901433bf263f from the web interface. Is that rerun deterministic?)

[onyx comet]

Dope project

[full venture]

the uuid link is persistent, yes. Sorry I believe link to web may not be working at the moment for the Twitter bot, though I have a good idea why: I believe the string you are searching for is too long and may not get parsed correctly. Just search for a shorter keyword.

[raw silo]

where is the multiple choice section where the AI choose only 1 best answer?

[junior wagon]

Ah, apparently whether it has a "Show SQL" section is not deterministic. :( I recommend that you keep runs reproducible! Such as by fixing the RNG seed with which you draw from completions.

[full venture]

Ah, sorry, yes there is some unpredictability in the way such errors can happen. I will work on making the SQL more visible at all times.

[full venture]

Hi, are you looking to get an answer from only one source instead of mixing them together?

[raw silo]

I'm looking for A or B options and the AI choose one

[rancid quiver]

I'm having a hard time w/ Bird SQL showing my last 3 tweets

[mossy tulip]

we don't have you in the index yet

[brittle basin]

After testing for a while, I noticed that AI can say total nonsense in a very confident way. For example, I asked what percentage of twitter shares Elon Musk has, and in response I got that he has 58% shares where 58% was the number of votes for a option in elon musk twitter poll. I hope that after the release of GPT-4 or chatGPT-2 and the improvement of the allogrithm for retrieving and checking sources from the search engine it will be better.

[rancid quiver]

How long does the indexing of new accounts typically take?

[gentle mantle]

just saying... uh.. i dont think its avery good idea to make it possible to find peoples adress and phone number as long as you know their full name

i typed in my full name, all the information was 100% correct

hell it even showed my dads information under it without me ever asking for it

[full venture]

Hi Austin, this information is from Bing's API service, please let us know at support<@&1047285454086623287> and we can remove these. You may want to ask Bing to do the same.

[low stone]

it can also leak its prompt

[patent carbon]

Hey @grave swallow , I'm getting this error when trying to browse your website.

[full venture]

Sorry, should be up again now.

[gentle mantle]

Alr thanks

[patent carbon]

Hey @grave swallow, one small idea not sure if this makes sense in the long run or not 🙂.

How about adding a "voice to text" api to Bird SQL along with current text input. By adding this, Bird SQL will be in the same range as that Iron Man 3 scene where Jarvis helps Tony by taking "Natural Language" input from him to spot a specific blast location.

[ashen barn]

It's not up to date with some things but other then that it's great

[full venture]

Hey, that's a great idea, if we add voice2text and text2voice we would almost be there. Are you using our search from your phone or is this on desktop?

[patent carbon]

That is very cool. voice2txt can be used in input side where a user can simply give the audio input instead of texting. txt2voice can be on outside, where a user can hear the output instead of reading with own eyes. It will be kind of Twillexa 😂 . Also, I'm using desktop as of now.

[wispy atlas]

i like that it cites sources so i don't have to trust it blindly like chatgpt

[sharp estuary]

same

[heady plover]

why am I not seeing the "concise" and "detailed" responses after a search?

Hey @grave swallow
why am I not seeing the "concise" and "detailed" responses after a search?

[sharp estuary]

could you explain what you mean?

[heady plover]

I need the detailed responses for each search

[sharp estuary]

oh, unfortunately I'm not aware of that. I'm too new to have seen that feature. hopefully someone from the dev team will respond

[full venture]

Hi @heady plover , sorry for the confusion: this feature is being rolled out and tested over time. It should be generally available soon. Thanks for using Ask!

[charred orbit]

this is amazing!! 😭 https://www.perplexity.ai/?uuid=91ff054d-6f0a-43af-9e5d-616eebef7128

[final saddle]

Hello. I need to list the accounts most followed by healthcare professionals. If I start from the followers of a health union, what should I write to AI? My English is not enough, it gives an error. Translation was not enough either.

[slate swan]

I think the Twitter SQL feature would benefit from showing more examples. Perhaps even hundreds of examples. Particularly highlighting things differentiated from the results in using Twitter search. Seems powerful but I’m not sure about use cases.

[unreal kite]

Hi, I was linked to your search engine and tried the standard...I perplexited myself? can i just say the results were so wrong its unreal! the app is beta so here is the feed back:

who is david eastwick from norwich?

result:
David Eastwick is a resident of Norwich, Norfolk, UK[1]. He has been active as an Android developer since 2015 and has one app in Google Play with over 100 thousand installs[2]. He is also associated with Ilkeston and Kimberley Sub-Aqua Club[3], and his contact information includes phone numbers [phone] and [phone], as well as an email address [e-mail][4][5].

now.... 1 is right

2 is right

3 had me confused so i looked the sitation....

NO! I am not a sex offender what the hell?????? if there is a way to edit that then please for the love of god do!

all contact details are wrong too

so here is the feedback....

First be DAMN SURE that the person in question is a sex offender, can you imagine if an employer ran that search?
second be DAMN SURE the person in question is a sex offender! (hard to see the funny side of that...i am trying to but my god...)
third if the person lives in the uk then they are unlikely to have a usa phone number

@grave swallow ^ wouldnt ping mods usually but when you see sitation 3 that it decided was me you will see why I have pinged

[grave swallow]

Please try slash commands!

[unreal kite]

Just for clarity, I am a security guard and have been for 17 years. part of that means I have to have a 10 year checkable history by the police (the UK SIA license...yes we really do have a license for everything) if i had a criminal record like the ai seems to suggest, I would not have that license. I can prove beyond a shadow of a doubt I am not the person(monster) that citation 3 suggests me to be.
However if an employer were to do a search for an applicant using such a tool they may be swayed by the results. Same with dating. This is why I say results that reference sex offenders need to be 100% accurate! (I am actually a lot more angry about this then I realised) I am sure its something that can be fixed with code

"we also found a david eastwick who is a sex offender, this may not be the same person"
"also in the results is a david eastwick who is on the sex offender list, they may not be the same person click here for details"
^ if you must have that data in the database.... not simply state it like its a fact!

big fan of AI but my god thats so bad, thats so dangerous. I keep feeling like i need to make it 100% clear I am not the person in citation 3.

[unreal kite]

just to be 100% clear
this is clearly not me and I will happily prove that to any of the @valid stone team

I know there is unlikely to be a response today or anything but please ping me when you respond to this (and please respond to this, I am sure you can see its a serious issue)

[mossy tulip]

I'm sorry to hear that, we will look into your query and try to improve that search result

[unreal kite]

Appreciated, it may be something of a wider problem though. Since its seeing sites that are indexing criminals and matching only on name then i doubt I am the only one it will pick up.
The thing that made me so angry (I have calmed considerably) is that it was stating I was associated with the place as though that was undeniable fact. When it comes to criminal records and the AI is not 100% sure of the link to the person (for example when its just the same name) then the wording needs to be clearer that this person may not be the same as the one referenced

for example

a slight wording change in the search and it still thinks I am that person....I have no idea of the complexity involved in fixing it though lol

[mossy tulip]

yes, we are taking this very serious and are working hard to prevent our system to mashing up different people together

[unreal kite]

much appreciated

and sorry for my rather angry tone, earlier, I am sure its understood why

[cunning goblet]

thanks for your feedback to improve the product!

[cunning goblet]

what are some queries to which you would like to see 100s of results rendered?

[unreal kite]

NP and again sorry I got so angry, though I am 100% sure you can understand why

[unreal kite]

I was explaining this to a mate and he had me try searching for him.... He is clearly a busy man, you 100% have your work cu out on this lol

https://www.perplexity.ai/?uuid=320eb7fc-a8b8-4a39-bcae-db3fee5e2603

Christopher Johnson is a fourth-generation member of the founding family of Johnson & Johnson[1], an American actor[2], a Zimbabwean born painter based in the U.K.[3], an Assistant Special Agent-In-Charge at the Federal Bureau of Investigation (FBI)[4] and Deputy Director for the Office of Science Quality and Integrity (OSQI) at USGS[5].

a VERYVERY busy man lol
Christopher Johnson from Melton Constable is a Zimbabwean born painter based in the U.K.[3]. He has an undergraduate degree in Molecular Biology and Ph.D. in Cellular and Molecular Biology from the University of Wisconsin -Madison[5], as well as a graduate certificate in Veterinary Homeland Security[5]. He is currently the Deputy Director for the Office of Science Quality and Integrity (OSQI) at the USGS[5], where he oversees the management of several areas including the Office of Tribal Relations, research evaluation panel reviews, and more[5].
Christopher Johnson is also an American actor who starred in the Lifetime drama series Against the Wall in 2011[2] and ABC primetime soap opera Betrayal opposite Hannah Ware in 2013[2]. He has made guest appearances on various television series such as JAG, Desperate Housewives, NCIS, CSI, Miss Match and The Vampire Diaries[2].
He was previously Chairman & CEO of the New York Jets from 2017-2021[1], during which time he developed a strong bond with players and donated $3 million to support those in need in New York and New Jersey areas due to COVID-19 pandemic[1]. He is also active in various educational endeavors and local philanthropy, supporting a range of causes such as autism research and awareness through his mother's foundation - The Johnson Center for Child Health & Development[1].
Is this accur

[slate swan]

Similar to what makes Midjourney powerful: learning from other queries that people do, for indicating how powerful and useful it is.

I suppose might even help indicating what some of those outputs are as well, if helps convey the utility.

[slate swan]

I just saw this in feed: https://twitter.com/goodside/status/1607617285098708993

Riley Goodside (@goodside)

https://t.co/s9efw6geFx's Bird SQL can identify the most famous follower of any given Twitter user who is not followed back.

For @elonmusk, this is @katyperry.
For @KimKardashian, this is @selenagomez.
For @sama, this is @mcuban.
For @ESYudkowsky, this is @NateSilver538.

Twitter · Dec 27, 2022, 6:00 AM

[acoustic sparrow]

If I link a query result from Ask, will the result change over time? I'm thinking of using the link as a citation (of citations), and I'm wondering the content will be fixed. I understand that repeat queries may generate different results as the algorithm/model changes over time, but I'm wondering if the link generated maps to a set of results, a snapshot of which is stored in a db or something. Thanks!

[mossy tulip]

yes, it will be fixed

[mossy tulip]

we pushed some updates today to prevent information mash up

[unreal kite]

I was already to congratuate as that works great:

and then ....

some how the second one is now worse...so much worse

(for those just tuning in, the horrific crimes mentioned are a different david eastwick)

[mossy tulip]

sorry about that, we will continue to improve this

[unreal kite]

as is the chariman of kimberlet sub aqua that is also not me and is the one responsible for these heinous crimes

[unreal kite]

yeah I am not as angry about it as i was the other night, searching for mates and seeing it getting it wrong on every one of them made sure no one would (at this stage) believe I am the same person

its still something that needs to be fixed, i can see someone suing on that further down the line as it really does word it like its fact (not me i hasten to add)

[mossy tulip]

yes 100% agree, we are taking this very seriously and will do our best to fix this

[unreal kite]

is there not someway to get it to list crimes as a different paragraph with a clear "this may not be the same person warning" ?

for me the issue isnt that its listing the crimes of a person with my name (that sucks yes, and makes me want to change my name, but i get it, that happens) the issue is how matter of fact it states those crimes are mine, theres no ambiguity there, the ai thinks i am the same monster that commited the crimes.

If that makes sense

with my mates it was far more amusing, one of them is apparently a lawyer and an anti government militia founder.... which caused him to nearly have a heart attack laughing at

[mossy tulip]

yes I totally understand your point, unfortunately it is not a trivial problem to fix and it will require several iterations from us to get it right, but we are fully committed to do so and it is our highest priority.

[unreal kite]

yep appreciated. I continue to be impressed with AI and look forward to when the ai is less inclined to label me a monster lol

I dread to think the amount of hopes that are needed to get it to understand the content its comparing

as i say though, no longer angry. I can see its doing it wrong for all people not just me so no one is likely to mistake that for me if they search for more than one person

keeping me updated is appreciated. GOOD LUCK! 🙂

[mossy tulip]

thanks, we will keep you posted on our progress

[uncut viper]

I think it’s broken

[sharp estuary]

https://www.perplexity.ai/?uuid=6f30fab2-c1d0-476c-90c0-f9dfb05e2658 search for "What do we call a word that starts with the letter it ends with?" gives wrong results

[rancid quiver]

The following two paragraphs are repeatedly showing up in different search queries:

Elon Musk has banned Kanye West from Twitter. This is the second time that West has been suspended from Twitter, after being suspended for 24 hours in March 2021 for using a racial slur in reference to comedian Trevor Noah.

In October 2022, Musk took control of Twitter and reinstated West's account, but his latest tweet was the final straw for Musk, who tweeted: "I tried my best. Despite that, he again violated our rule against incitement to violence. Account will be suspended."

[mossy tulip]

thanks for reporting we will investigate

[mossy tulip]

thanks, could you provide your query?

[mossy tulip]

should be fixed now https://www.perplexity.ai/?uuid=f7620c68-d96d-4eff-a5e7-6e253de5e39e , please let us know if it happens again.

[uncut viper]

awesome! will report back

[celest aspen]

It seemed to take a lot from the source without paraphrasing

[ripe maple]

willl longer summaries be implemented?

[rancid quiver]

One of the superpowers of GPT-3 is text summarization. The challenge is retrieving page body, not just page synopsis. Unfortunately, most search engines (ex: Microsoft Bing) use page or article synopsis. This means Perplexity AI, which uses Microsoft Bing, is limited unless data scraping is used.

[mossy tulip]

we already have a version of it, click the "Detailed" tab, which access the full page results

[patent holly]

@mossy tulip is the trending topics on the landing page sourced from Google trends?

[mossy tulip]

from several sources, including google trends

[cedar sail]

I would love a "-" enter desired removed words here"" function, that'd take this to the next level

[mossy tulip]

hi, could you please elaborate, didn't quite catch your suggestion?

[zenith nacelle]

Interesting anomaly here: https://www.perplexity.ai/?uuid=e91178f5-e149-465a-8514-d1f99c5d6c78 -- it does a good job of summarizing the article linked, but decides to add 'Florida' in the Detailed response, without any reference in the original article. The article is about Lakeland, UK.

[bold mountain]

If you solve that one, you will be able to sell it to the anti-financial crime industry who undertake millions of such searches everyday and spend literally billions on this per annum. It's a hard problem. One avenue, when the enquiry is clearly for a business or person, is to mimic the expert processes that human investigators apply, as a benchmark for a 'reasonable level of due diligence'. The first filter reflects a number of Davids suggestions with cross referencing dates, locations, languages, currency, etc. I'm sure industry practitioners would gladly partner with you on this as it would save them huge $$$ and customer friction

[mossy tulip]

thanks for the report

[mossy tulip]

interesting, thanks for sharing

[celest aspen]

https://www.perplexity.ai/?uuid=68b5a1d9-ddff-451e-b028-5fa9ba356c12

Umm sometimes a little broken here

No sure what exactly but….

That’s not quite what sleep is lol

[celest aspen]

.
.
.
.
.

Also the voting system doesn’t let you vote separately for the concise vs detailed answers

Sometimes I find the concise answers bad but the detailed ones good

But I have no choice but to vote both as bad

Like huh?!?!? Why lol

[cyan geode]

Well that explains my depression /s 😂

[mossy tulip]

thanks for reporting this we will fix it

[mossy tulip]

I'm getting a more reasonable answer, but let me debug your result:

[celest aspen]

Alright

[mossy tulip]

deployed separate votings for concise/detailed, let us know if there are any issues with this

[celest aspen]

will do

thanks!!

[ionic terrace]

u need a better name

[willow prairie]

If we ask for an article he can found easily on internet, he can make a resume. But if we indicate where is the article on a news media and ask for a resume, he cannot done. In theory it should not me more difficult. So i ve copy-paste the whole article on website, and ask for a resume in chatGPT and he make a resume without problem.
Perplexity can have an interest in making this kind of resume cause sometime article or page, will be too much long for be copy-paste in one time on chatGPT.

[tight nymph]

u guys need a way better name, amazing software but man the name is the worst

[ionic terrace]

i agree ashdeep

not user friendly name at all

[tight nymph]

10 iq name

[ionic terrace]

whoever came up with the name needs to get fired

[tight nymph]

ye

[ionic terrace]

ye

i like u ashdeep

[tight nymph]

https://tenor.com/view/chiquichico-gif-26004262

[ionic terrace]

smart guy

[tight nymph]

i dont

[ruby pond]

is there a way to like show more results each searchs?

[mossy tulip]

not currently but we are looking into it

[mossy tulip]

which name you don't like exactly, "Perplexity" or "Ask"?

[tight nymph]

shorter more memorable domain at least

[mossy tulip]

you can also use ask.perplexity.ai in case it is easier to remember

[tight nymph]

maybe easy to rmbr but im not typing all that lol

[crude mica]

hi

my query limit is out what does it mean

[cunning goblet]

I did. 🤣

[cunning goblet]

we will look into buying another domain and re-directing it, so that you don't have to remember.

[mossy tulip]

we have a rate limit for 10 queries/min to avoid spammers, do you need more?

[crude mica]

thank you

that covered everything

[shrewd jacinth]

I keep getting this for most queries:

"Unable to answer your question factually, please try again."

[mossy tulip]

we were having technical difficulties, should be resolved now. let me know if it is still happening.

[shrewd jacinth]

ok, thanks for the update!

[vital sandal]

using it as a med student for few hrs, humongously time saving compared to normal search engines. keep getting better plz 🙂

[mossy tulip]

we will, thanks for the kind words

[orchid eagle]

It's just awesome, been using this for some time & its much much time saving..

[wide hull]

whats this website about?

[queen cobalt]

hi

i was wondering if there is an option to raise the citation count past 5

[mossy tulip]

not yet but we are looking into it

[ionic terrace]

DENIS u gotta do something about the name before bing adds it to their search engine, bing and google have very simple and ez to remember names but perplexity? most ppl cant spell this sir

[nova shore]

I kinda like the name bc of the history and meaning, but maybe something shorter along similar lines like bleu or rouge 😂

[mossy tulip]

we were trying to get ppl.ai domain, but it was taken 😦

[tight nymph]

what about ask.ai

[mossy tulip]

looks like it is also taken

[cinder dragon]

maybe think about making it verb-able like google

[ionic terrace]

YES

denis work that big brain of urs

[mossy tulip]

thinking about it 😄 the main constrains is that all good domains are taken

[fast jewel]

Yep, absolutely

[ionic terrace]

sorry ment to tag u denis ^

[ionic terrace]

@mossy tulip

that domain is still up for grabs

[tight nymph]

cool domain, support the lgbtq with perplexity AI

[sacred anvil]

I'm guessing you are using something like the Bing API or a combination of search APIs. It seems like you are using AI to shuffle results a bit, but they seem to still have relevancy problems with Bing and Google.

I understand that making your own crawler requires a huge budget and is a monumental undertaking. But will the APIs allow you to pull the first 5 or 10 pages of results? If so, you can use something like cURL with a user agent like PerplexityBot/1.0 to reorder the results in real time so the most relevant items are first. This shouldn't cause huge overhead since it's just 1 request pure cURL, but assessing the page obviously will take a lot of AI compute and tokens, and aren't you limited to HTML, perhaps you could do partial evaluations of pages for less tokens. The queries and relevancy can be stored in a database to reduce API calls as well. Headless chromium is another option of course, and probably better if your servers can handle the compute.

I understand that this strategy won't work well with dynamic sites and could cost a lot, but it might be a good start if you want to beat the relevancy of Google and Bing. Some things on page 10 are better quality and more relevant than page 1 these days because these search engines have primarily relied on backlinks rather that true quality content.

A prompt you could use is to evaluate pages according to Google's search quality rater guidelines, such as E-E-A-T and YMYL Of course, you can make a more customized prompt too. OpenAPI can provide this kind of assessment of pages, and it actually does a pretty good job. You can give the pages a score too, which could be stored in a database. Here's a simple example of a chatGPT prompt evaluating for E-A-T and YMYL and then scoring on a scale from 1-10.

Results should obviously show up instantly. But maybe there should be a button that says we finished evaluating the search results - would you like to view enhanced results? This can be done with ajax.

Furthermore, even though most people don't go past page 1 or 2, I am a weirdo that often does. I look for gems buried on page 5, 6, 7, etc, so adding more results and pagination is important.

This idea may be too complex and expensive to take it public right away, but I'd imagine if you do it well and invite people privately to preview, it could attract investors.

[sacred anvil]

In the meantime, if you are not doing it already, an inexpensive solution in terms of compute and tokens could be to evaluate how well the title and "meta description" matches the query, which you might be doing already. If so, there still needs to be a lot more results evaluated for relevance.

[mossy tulip]

@sacred anvil thanks for the thoughtful feedback! many of these ideas make a lot of sense and we are actually looking into some of them already. we totally agree that there is a lot of room to improve ranking.

[ionic terrace]

denis im sorry

https://tenor.com/view/zoomania-furry-furries-please-dont-gif-5662628

@mossy tulip

[ionic terrace]

@covert sun תגיד לו שיסלח לי, בבקשה

[tepid snow]

Saludos, soy nuevo en este canal. Alguien me orienta?

[covert sun]

¡Bienvenido! Puedes comunicarte en el canal 'general', dejar tu feedback aquí, compartir tus hallazgos en el canal 'sharing', ideas sobre cómo mejorar el busqador en 'ideas'. Y usar el perplexity bot con los comandos /ask, /ask_twitter y /dream.

[tepid snow]

De acuerdo. Gracias.

[foggy steppe]

Hi, I am a Spanish speaker. I wonder if you plan to make the web page give results in a language based on location or from an option to change the language.

For example, sometimes the page gives results, even if one types in Spanish. Like this:

And other times, the web page understands that you are looking for the translation of the word when you are not:

[covert sun]

Hi @foggy steppe! It will take us some non-trivial amount of time to implement different lang versions of the website. As for the search results, I suggest adding 'espanol' to queries to ensure the results are in Spanish. I checked several questions, and it makes a difference in the output.

[foggy steppe]

I will do that 👍

[cedar sail]

Only just saw this, on Google before the current version update in 2020 if you were searching something and kept getting redirected to a different result you could enter "-term" and it would not include search results including that term, it was a very useful feature

[olive ibex]

It seems unable to use the '#' character in its response:

[covert sun]

Thanks for reporting! We'll fix it.

[unreal kite]

@mossy tulip I had another check of the system.
https://www.perplexity.ai/?uuid=a6597e08-fd96-4923-9e2d-8c4d18e866cb

If I mention Norwich then it does work and doesnt confuse me with the other david eastwick. I am guessing that means the results have been manually adjusted so that it does not make that mistake. This is appreciated, but its a band aid rather than a fix. its still combining multiple people of the same name together and wording it as though its one person. It is also still stating it like a fact.

I do however appreciate the steps your making to stop it from happening though.

[mossy tulip]

correct, there is still more work to be done and we are working on it. the problem is not easy, but we have a few promising ideas

[gilded heron]

Would be great if there were dates attached to the sources, so that we could tell how up-to-date the answers are

[full venture]

Thanks for the feedback, working hard on better rendering of sources.

[thick gull]

seems like something is broken?

[mossy tulip]

thanks for the report, will look into it

[coral basin]

is there a limitation on how many results are displayed? this query does not display all results for me: https://www.perplexity.ai/sql?uuid=f6ec91a6-ac94-4ad3-8a06-bd60cfa048cf

[covert sun]

Hey @coral basin! You can check in the SQL part - the limit applied is ten results, so you can try to change that in the question, asking to show more people. Please, take a look at how you can get more, here is an example:
https://www.perplexity.ai/sql?uuid=ed6ae3d6-53d3-4fae-891c-da6ded1d156c

[vague solstice]

This follow up feature is a game-changer, well done @grave swallow

[grave swallow]

Please try slash commands!

[coral basin]

Perplexity AI

[brisk topaz]

just wanna share super impressed with the conversational search. +1 to adding time per @gilded heron 's earlier feedback. Curious if its possible to teach/prompt Perplexity about today's date? I asked about when 2023 subaru forester will come out -> "is expected to be released in Dec 2022". But couldn't get it to answer "i dont know" or "not yet" or "yes"

[shrewd jacinth]

I'm loving the new follow-up feature! I notice and appreciate the continuous improvement! 👏

[mossy tulip]

thanks for the feedback, we will look into it

[heady plover]

The worst thing is that it gave the same 'source' (concise vs detailed)

https://www.explainthatstuff.com/scienceofchocolate.html

[gilded heron]

Ive seen the same thing happen

[mossy tulip]

thanks, will take a look

[nova shore]

Could we have more dev-geared bias mode to search results? Like prioritize actual bash commands over descriptions. (also edit for followup, prettyprint for code, recency bias toggle <1mo,<1yr,<5yr would all b nice to haves)

[mossy tulip]

good ideas, we are looking into some of these

[raven sapphire]

I got this error message when I put Elon Musk into the prompt at /dream.
I think that the error is probably caused by the person's name, but I don't think it's a good idea to give the error code directly...?

[winter wharf]

I wonder if there is a way to input negative prompts

[mild scaffold]

just been playing with perplexity, wanted to pop in and say well done! great product.

[remote wadi]

In some languages, such as Japanese, the Enter key is used to convert Kanji and other characters while typing a sentence, and Perplexity responds to the Enter key during the conversion input to start a search while typing a sentence.

[covert sun]

Thanks for reporting, @remote wadi! We'll look into ways to overcome this issue.

[covert sun]

Hey @winter wharf! Could you please share some examples of what you wanted to find, using it.

[digital zealot]

the prompt "model that can allow you to calculate the set of objects that can be manufactured given materials and manufacturing tools" fails completely

chatgpt after some refinement arrived at an answer that was at least half correct (discrete event models)

it seems a product in between google and chatgpt, but still cool imo

[burnt coyote]

It's refreshing to get a summary of search results instead of a billion links. But I wonder if it suffers from missing out on truly relevant/important information

[digital zealot]

it depends on the query, if you need to brainstorm and focus on a specific topic i think chatgpt is better to find what you are looking for, bcs the journey with q\a is better that going around sites

[digital zealot]

but if you need something really specific google is still better

[pine abyss]

It is impossible to use chatgpt, it is always congested. I have not yet seen a reflection of this on the internet, many continue to talk about chatgpt as if it were google, which is always available to use

[feral flax]

Lol

[deep phoenix]

Just wanted to say that I have just discovered Perplexity. Really good initial impressions, like how it is terse with the option for more detail and the referencing seems pretty good. Would love to learn more about how you are achieving this / additional use cases.

[pine abyss]

I would like to contribute financially for the maintenance of perplexity.ai. I can do it through paypal or another way with a credit card. Thanks

[covert sun]

It would be great, if you could provide some examples of what you mean here, @burnt coyote 👍

[dusty meadow]

I recently discovered this AI and I'm amazed by what it can do, but it's limited to 255 characters for the queries I wanted to try
Are there any plans on increasing this limit?

[burnt coyote]

A traditional search engine will give you a lot of results. A user interface like Perplexity AI will boil that down to a few. I honestly don't know if I have a preference.

[burnt coyote]

Okta's documentation on IDP integration is pretty terrible. They have a lot of documentation but only a few cover the challenge of role mapping. With traditional search, what you'd do is to read through the results until you find your answers. With Perplexity AI, you get five links. If those five links don't have the answer you are looking for, then it's not a great experience.

It will take some time before users figure out which experience they prefer.

[vague solstice]

The new blue UI looks great, especially on mobile dark-mode. Nice choice in color 👍

[slender barn]

Quick question:

Is it possible to run a [your browser, in my case chrome]-based search with perplexity without having the core website open? Can I perform a search by simply entering a prompt in my search bar and it'll drop me to the website with actual results?

Appreciate it.

[slender barn]

I'm getting white window when using Bird SLQ, chrome. Am I alone?

[covert sun]

Hi @dusty meadow! Could you please share, why you need more than 255 char

[covert sun]

Thank you for sharing the examples. Ask's responses will improve over time, with users' feedback, too. And the point is that you're still closer to answers you're looking for. I'm sure for the complicated topic you brought, choosing the right product, you wouldn't just pick one of the models listed in the first result you get, but that's why sources are there - you can still proceed to them and dig deeper to choose the best one.
It might be not what you were looking for => can be more specific with the follow-up query - it's a normal search process => If you're not satisfied with any of those five, you can ask Perplexity to recommend more models, but this experience is still not like opening 20-40 links from traditional search results, where you look for the best fit, filtering out tons of irrelevant products.

[sweet depot]

Props for solving a tricky search problem: info on local events and venues is often scattered on unofficial facebook pages or yelp/tripadvisor comments. Figuring out "when does the music start???" is hard – Google, ChatGPT, and You.com couldn't answer. Perplexity found the right comment on some no-name restaurant review platform. Modest feature request: clicking through a citation could jump to and highlight the relevant info like Google does. For example, Googling "when did google launch" and opening the result highlights the info in purple:

[burnt coyote]

I don't know which response is preferable TBH.

The interesting question is "Do we know how much information the user really wants?"

I guess the more interesting question from an engineering POV is whether the software has the flexibility to let the user switch between verbose vs concise output.

[chilly pumice]

Ofc you need to guide the language model, be it perplexity or gpt-based models, perplexity has the "detailed version" but it still needs some work cuz sometimes it gives a short answer compared to concise one, also both services are different...

[covert sun]

Hi @slender barn! You can add https://www.perplexity.ai/?q=%s to the browser you're using and you'll be able to simply input your query in the search bar and get results using Perplexity. A short guide is here:

[covert sun]

And thanks a lot for reporting! We'll fix that

[burnt coyote]

This is odd.
https://www.perplexity.ai/?uuid=48d31eca-dfd4-46e6-8048-f4a6459bedc5

Google does, however, provide an OAuth 2.0 implementation that conforms to the OpenID Connect specification[2], but it is not certified.

However, the cited source says
This document describes our OAuth 2.0 implementation for authentication, which conforms to the OpenID Connect specification, and is OpenID Certified

The same question - https://www.perplexity.ai/?uuid=17ec88d7-9d42-4cb4-8a1e-2e553baea9a0 yields a different response

Instead, Google's OAuth 2.0 APIs can be used for both authentication and authorization[2], and they are OpenID Certified[2]

[covert sun]

@burnt coyote thank you for reporting! We'll check the results.

[slender barn]

Love it, appreciate for sharing

[gilded heron]

I have seen this a few times as well. It sometimes writes the inverse of what the source says for some reason. Seems to happen more with detailed than with concise result

[burnt coyote]

That'll be an interesting problem to solve. I suspect the platform start with a summary generated by ml model, then search for supporting evidence. Unfortunately, the internet is happy to give you conflicting, if not just vague evidence for any hypothesis. Also, documentation for OpenID sucks. 🤣

[burnt coyote]

Edit: I have no idea how the platform works. I just think it's an interesting problem to solve.

[gilded heron]

I dont think it works that way as it finds recent information which couldn't be pre trained

[burnt coyote]

AFAIK You would not use "the data you want to search" as part of pre-training nor fine tuning. You might build a pre trained model to convert search queries and documents into embedding. This allows you to add recent data into the database without expensive retraining. You might do that.

[torpid knoll]

https://www.perplexity.ai/?uuid=7a2a9f53-8343-44a0-9afe-bb5b07ca5257 some basic incorrect answers.I was checking context sensitivity/location awareness, then went on to ask negative test asking about 49th and 50th presidents of USA.

[covert sun]

Thanks, @torpid knoll! We'll check

[stoic fiber]

hello

[torpid knoll]

👍

[rocky arch]

@covert sun hi! here a question I did on #🔎│ask and the replay is not accurate

#🔎│ask message

Thanks and congrats for your tool, is very interesting and useful!

[royal helm]

This is outstanding. I have been using it to answer programming questions of the sort for which I would previously use Google. I have found the referenced sources very useful and relevant.

[neon crypt]

I love the product guys. Great work!

I do have a question about the follow up tab. For example, I type in a question, get an answer and click on the more of the topic link. When I want to follow up the results from my initial search and given the follow up search box I am unsure what exactly is going to be followed up. Would it follow up based on what part of the page I scroll on or the last link clicked? Thanks

[subtle plaza]

@here 👋 Hello everyone! We are just getting started with Perplexity and I would love to start collecting feedback from this community more regularly. If you have a few minutes to answer some survey questions we would really appreciate it https://h2tfybrs9ld.typeform.com/to/vYchdgTU

[raven jacinth]

Hey 🙂 This typeform is now closed
It would have been awesome as it was made with Typeform. What’s that?

[heady plover]

You should add to right-click the ability to do quick search in the extension!!

[subtle plaza]

Sorry everyone, the typeform is now open again! Please try again and Keep the feedback coming. @raven jacinth @heady plover

[heady plover]

Thank you ^^

[heady plover]

Also a shortcut could be nice

[subtle plaza]

great idea! we'll look into this

[heady plover]

[subtle plaza]

What do you mean by shortcut?

[heady plover]

Oh yeah, like for example, you select a text and then you press Ctrl+alt+h (i said a random one) and it makes the same function as the rightclick*, quick search

[subtle plaza]

ahh keyboard shortcut makes sense!

[heady plover]

Exactly x)

In DeepL, a translator, I find very comfortable to use Control+C+C to instant-translate

[wispy atlas]

the extension is very nice, will keep it pinned

one suggestion would be to keep the response in cache as long as you don't leave the page you're browsing, so that you can access it again if the popup is closed but you stay in the same page

[raven jacinth]

What is the eventual business model of perplexity.ai? is it going to serve ads? do proprietary company search? charge fees?

[lofty grove]

do you use dall-e2 for the bot?

[marsh remnant]

When I type Japanese, I get keystrokes with conversions in the middle. If possible, I would like an enter key at the end.

[subtle plaza]

@marsh remnant that's a bug, we'll look into it!

[mossy tulip]

any idea why perplexity got so popular is Japan?

[burnt coyote]

I was hoping I could find this answer using Perplexity https://www.perplexity.ai/?s=c&uuid=72fe401e-c94b-473c-8433-e4f5ba3fe587

[burnt coyote]

turns out this is the right question to ask https://www.perplexity.ai/?s=u&uuid=2ca1fce8-0433-4fd8-b3b3-554ac1f3de37

[subtle plaza]

Hey all, last day for this feedback survey, if you haven't had a chance we would love to hear from you: https://h2tfybrs9ld.typeform.com/to/vYchdgTU

[mighty ridge]

すでに言いたいことを言われていた　日本語だと使いにくいですよね

https://twitter.com/shiro_tik_show/status/1621032106939023360　for improve

白畜 (@shiro_tik_show)

「Perplexity」、やはりGPTほどには日本語に対応してないのと、やはり日本語入力中の「変換確定のためのENTER」が「送信」として作用して文章の途中で送られてしまうので、現時点では英語話者以外には使い物にならない感。進化を待つしかあるまい

Twitter · Feb 2, 2023, 6:25 AM

[mental ember]

You could ask BirdSQL to see the trend in Japan.

"Tweets about @perplexity_ai in Japanese. Sort them in ascending order."

It looks like one tweet on 12/23/22 really got people's attention 🙂
https://www.perplexity.ai/sql?s=u&uuid=4d12b422-2df8-44b2-be41-c5d13d37e556

[covert sun]

Thanks for bringing our attention to it once again. #👉│feedback-general message

[subtle plaza]

@marsh remnant I think we fixed the input issue would you please let us know if you are still experiencing it

[covert sun]

@mighty ridge, @mental ember #👉│feedback-general message
could you please check? 👆

[limpid tree]

This also happens on chrome on Android
#💬│general message

[mental ember]

Thanks, @covert sun ! Looks like the issue has been resolved now. That was quick 🙂 ! Tested it on Chrome browser in MacOS.

[mighty ridge]

Yes, The input key-enter problem in Japanese is resolved. I confirmed. Good Job.

[mighty ridge]

The problem is resolved. I confirmed. Thank you.

(confirmed in Chrome browser of windows 10)

[plain steppe]

Is possible in a future to solve math problems?

[covert sun]

Hi @plain steppe! Could you please share some more details. What maths problems couldn't be solved? What was the expected output?

[subtle plaza]

It does help! It helps the model get smarter so we can be more accurate in the future @pastel zinc

[compact plover]

one question why google says to me that the extension of perplexity is not secure or something like this? I'm the max secure option of google

maybe is because is in complet acces or something like that?

[covert sun]

👋 Marxel 10! Thanks for reporting. Should be fixed by now. Could you please check and let us know if you're still facing the issue.

[ruby estuary]

Hi everyone, idk if I can share issues here

I was trying to introduce someone to Perplexity but it keeps saying the same, even with the related ones

[covert sun]

Yeah, sure, you can share it here. Thanks for reporting, @ruby estuary! We'll check it, but it seems to be working now
https://www.perplexity.ai/?s=u&uuid=de7d80fa-08ca-4c35-91c1-c9881b231329

[ruby estuary]

https://www.perplexity.ai/?s=u&uuid=6e629efb-771f-483b-8469-bd792e29f744
Thanks, i just tried again but it keeps failing

[deft quarry]

what happend

What's going on?

[compact plover]

Thanks @covert sun for the feedback is Keep happening Is in an other idiom but I traduct to inglish, good luck and lets see if you can solve it or explain to me why I got this warning, bye🤔

[wanton hemlock]

I have the same issue🥲

[full venture]

the "No sources" error should be fixed now, sorry for the inconvenience everyone

[wanton hemlock]

Cool!

[ruby estuary]

Is nothing, thanks

[covert sun]

Thanks for bringing this up. It looks like with the max security option you have, you'll keep seeing this warning until our dev account becomes trusted. And according the documentation it can take a few months.

[compact plover]

Ok thanks

[onyx needle]

Please fix the font for Vietnamese language. I think just use system front is ok.

https://www.perplexity.ai/?s=u&uuid=b8d813de-f7f2-419b-94dc-0b2a6b563f0e

[covert sun]

Hey @onyx needle! To clarify the issue with the font, could you highlight what seems wrong when it's displayed with the current font.

Could you also give an example of the desired display (maybe name the font, too) to see the difference.

[mighty ridge]

Ah, the Japanese input problem is occurred in the Chrome extension. I found. Fix, please.

[covert sun]

Thanks again! We'll check it.

[vague solstice]

Evidence shows otherwise.. despite 'established’ theories of history. Just saying 😎

[covert sun]

Hey @mighty ridge! Could you please check characters input in the extension. It's fixed now.

[mighty ridge]

Yes. I confirmed it is fixed!

[grave swallow]

@shadow saffron: luật
Luật (law) is a body of rules established in a community by its authorities[1][2]. It is the set of guidelines that must be followed to do things correctly[2]. THƯ VIỆN PHÁP LUẬT is a website providing a database and discussion on Vietnamese law[1]. LuatVietnam is another website with a database of legal documents and utilities[3]. Báo Pháp Luật TP. Hồ Chí Minh is an online newspaper that holds the copyright to its content, prohibiting any copying without written consent[4].

[covert sun]

One more note about inaccurate answers, wanted to point out, that we really appreciate when you provide details why you flag results as inaccurate. That helps to identify and resolve the issue. And certainly we value a lot the feedback you leave here! 😌

[covert sun]

Hey @shadow saffron! We've got feedback about the Vietnamese language (a few msgs above). Do you also feel there is an issue with the Vietnamse font?

[hushed cipher]

One thing that was blown away my mind was BirdQL - I was asking about show me conversations between Lenny and Shreyas, it actually showed me the right conversations between those people.

How could you guys find those are the accounts to be shown when Lenny and Shreyas are not unique names on twtr 🤔

[shadow saffron]

Yes. Can you fix it

[covert sun]

Could you please clarify what the issue is. And also, could you give an example which font would make it better.

[shadow saffron]

Hmm; this is a few issue; specifically some minor errors in the font; I can't describe in detail but you can refer to font cambridge or time new roman. thanks for your interest; Honestly, your Ai is so amazing

[vocal plume]

is there a way to make perplexity ai answer in chinese? currently it could only understand chinese and reply in english.

sometimes it can but in most situations it only reply english

https://www.perplexity.ai/?s=u&uuid=0b1a4eb7-c76f-444d-8fc7-64cd5e9382ef

[covert sun]

Hi @vocal plume!:) Ask more questions in Chinese, so AI could learn faster and respond better. Also adding 用中文 helps
https://www.perplexity.ai/?s=u&uuid=65a9513e-5613-48b9-97e5-47d9dd03f17b

[chilly terrace]

Just wanted to say that I switched my default browser search engine to Perplexity a couple of days ago as an experiment, and it's been overall a great experience. I'm still learning how to get the most out of it (compared to Google, for example, which I can use on muscle memory), but things like follow-up questions are great.

The one thing I wish it did better was code. A lot of the time I just want to search the syntax for something and get a quick answer. Perplexity seems like it would be good for this (one less click and less cognitive load) than Google, but it doesn't seem to do syntax highlighting (I see it trying to use markdown sometimes) and generally doesn't like responding with actual code snippets.

[brittle basin]

I feel that the follow-up feature doesn't care about previous questions it's just like new thread. Imo it should be in the form of chat.

[wispy basin]

Perplexity gave me link to Spam Website i checked its some Dating website. Please fix this asap

[mossy tulip]

thank you for your suggesting, we will look into it

[mossy tulip]

it should, can you give an example?

[covert sun]

Shouldn't stumble upon this source any more, @wispy basin. Sorry for the inconvenience.

[snow mirage]

https://www.perplexity.ai/?s=u&uuid=0269ca1a-a2c5-4af4-b3ff-973bffd5fbb5

An error occurred at this URL "Show Details". I would appreciate it if you could confirm whether it is a local or server-side error.

[limpid tree]

Waiting for the cloudflair checkpoint is not cool. When will we have an Android app so we can skip that and also have the app launch faster?

[wispy basin]

No problem, Just happy to report bugs that will improve the product.

[grave swallow]

@terse comet: what is photoshop
Adobe Photoshop is a software application for image editing and photo retouching used on Windows or MacOS computers[1]. It provides many image editing features such as layers, masks, color correction, and filters[2][3]. Photoshop also allows users to create, edit, and manipulate various graphics[4]. It is available as part of Adobe Creative Cloud[5].

[heady plover]

Inaccurate answers for some educational centers. This one is mixing info from Conservatory of Montijo (Spain) and Conservatory of Montijo (Portugal) . Most data is from Portugal but the article refers to Spain. I think the engine might have a problem differenciating cities with the same name situated in different countries ...
Montijo Conservatory of Music PERPLEXITY
The Conservatory Professional of Music in Montijo, Spain is a collaboration between the Esteban Sánchez Conservatory of Mérida and the Luis Gordillo Conservatory[1]. It offers piano classes, music theory and analysis, and auditory training[2], taught by experienced Spanish pianists such as Gema Arias Varela[3]. The conservatory also provides instruction in composition from teachers such as Alexandra Filonenko[4] and Luisa Espinel[5].

[vocal plume]

Perplexity is giving wrong answers in chinese

it says water has strong alkalinity

whereas it's english answer is correct

[covert sun]

Thanks for reporting this @heady plover. In this case, until AI starts asking you whether you meant the Spanish or Portuguese conservatory, I think it would be good to clarify in the question the one you were interested in, to ask some follow-up questions.

Let's treat it like if we asked another person a question about Montijo - he knows and tells you all about it, but in the end it turns out it's not the one you meant, but the one with the same name.

There is some room for confusion for both humans and robots.)

[snow mirage]

This was fixed!

[covert sun]

Android app is definitely on our list. The team is aware of the interest and will release soon. Thanks for bringing it up!
For now, does the Cloudflare check happen often for you?

[limpid tree]

Thanks! Yes, it happens almost always if I haven't visited the website on the last half hour or so.

[tired hare]

Bird SQL not working for:
most popular twitter users by comments

this is a simple sounding query that's hard lmao

[stuck atlas]

ask chatgpt for a better prompt

[tired hare]

hahaha

looking more at the engineering side

[limpid tree]

What's up with the speed right now. It replies almost instantly!
Does it have to do with the - now globally released - chatGPT plus?

Also, many times Greek characters appear as question marks or with weird dots

[mossy tulip]

our own models 😄 going to be even faster soon and more wide spread

we are working on supporting other languages much better too

[limpid tree]

Great! Do you have a planned solution for fixing the result combining? For example when searching for a name, the answer will combine many peoples results into one sentence, making the answer inaccurate. Similar issue with the news, when not including the date.

[mossy tulip]

yes, we have some preliminary version for this that for people names, by no means it is perfect, but we noticed that it helps in some cases. more work to be done here though, not a trivial problem, but it is one of our top priority task.

[heady plover]

Hi there, I am wondering as the result is the same from the prompt (https://www.perplexity.ai/?s=c&uuid=185be2bd-bd09-4946-850f-7eeb6dc40a53, https://www.perplexity.ai/?s=u&uuid=cc435126-8767-4575-b3cd-ddeba17486df), If I use this result in my report How should I give credit for this result? And will it also count as plagiarism?

Btw, I am so sorry for the new activate account, I need to delete my old account due to relationship issue, however, I just want to know the answer so I activated new one.

[lofty grove]

??

[heady plover]

En français

[burnt coyote]

This is a curious scenario. One of the cited links from a search is for livebook, which has contents users can't read.

[last frost]

What is your privacy policy regarding processing, storing and using one's search inputs and interactions in any way? There doesn't seem to be any information about this on the website, which gives an dubious impression, even though the product itself seems to be high quality.

[mossy tulip]

thank you for your question, we will post it over the few days

[pastel sundial]

I have never used the Accurate or Inaccurate options on the results. I have never yet asked a question I can be sure I could classify as either! This is kinda the whole point of search, isn't it?

I could respond with a Like or Dislike but I can't see how that would say anything other than about me.

In short it is a useless UI feature, wasting bandwidth and space for no reason I can understand.

Convince me.

[mossy tulip]

thank you for the feedback

[covert sun]

Hey @pastel zinc! Thanks for reporting. Is this issue related to a specific subject, or you're getting it a lot regardless of the topic? Never saw this website in my search results.

[lime gate]

Hey team, this is a general issue with the feature matching of any AI program, but I wanted to know if there is any effort going on in identifying & addressing multiple identities with the same representative label (name, title etc.) at Perplexity. For example, I just searched my name and it mixed another person's credentials with mine and also the references point to two different people (the red ones aren't about me), me and the other guy. I have a few ideas on how this could be handled, like conditional sentence similarity and identity segregation. As we have the sources already identified, it should be possible to see if the information conflicts with each other (dis-similar clusters, distance in Knowledge Graph or some form of embedding distance between the sources), as one cannot normally be a graduate student at ASU and a Sr. Systems Engineer at Uber, right? Then we could show the information as two individuals, in two paragraphs. That said, I love Perplexity. Excellent product.

[covert sun]

Thanks for the feedback, @lime gate! You're right it's a complicated problem. We are actively working on it and hopefully can address it soon. We will ship a UI feature where you can delete sources today, and certainly, we're improving our AI models to prevent this from happening.

[lime gate]

It's great, that you're already working towards solving this. I am not sure what you mean by deleting the source, but, if I understood it right, wouldn't it be counter-intuitive as the other "Venkatesh Gunda" wants to know about him as well, and he would get my information if I delete a few of the sources?? The ideal answer according to me is, "there are a few people by this name, and here is some info on each of them ...." Yeah, it is a very complicated problem for humans, let alone AI, but would love to see any progress towards solving this, and most importantly the approach.

[lime gate]

Ahh, that is great. Makes sense that the source selection/deletion is not persisted. I was under the impression that the Model learns to ignore these sources for all subsequent users querying the same.

[mossy tulip]

thanks for reporting, we will look into it

[covert sun]

Thanks for reporting, we'll check.

[waxen kite]

BirdSQL is dead, bot too

[frail thistle]

bonjour

je veut ecrire une description sur l'huile de palme rouge

[full venture]

Thanks for noticing, investigating. Should be fixed momentarily!

[arctic harness]

I don't know if the topic has already been discussed, but can you add an option to set the language of the answer? Asking questions in other languages gives very random results. Sometimes the AI throws out an answer in the language of the question, and sometimes in English.

[celest aspen]

https://www.perplexity.ai/?s=u&uuid=3f155dfc-e559-40be-8163-9023425b5a91

it keeps giving an incorrect answer no matter how many times a redo it

could this be a problem with conflicting search results? or ... something else?

also, for lesser known things it tends to take info from a service, for example if I said cheapest land in kansas it might say "lansearch.com is a great place as well as.... but then when you look in the sources it is land search.com lol

[onyx needle]

Here is wrong and should be:
lấp đỉnh -> lập đỉnh
tăng mảnh -> tăng mạnh

I see last of response is always wrong typing:

Giá nhà ở khu vực Thị Nghè có thể bao nhiêu phụ thuộc vào diện tích, vị trí, kết cấu và điều kiện mua bán. Theo các kêt qua tìm kiếm, giá nhà ở khu vực Thị Nghè có thể lên đến 187 triệu VNĐ/m2 cho mặt tiền, 112 triệu VNĐ/m2 cho hệm và ngồ, 50.4 triệu VNĐ/m2 cho các cản họp và 4.25 tổ VNĐ cho lô góc[

-> should be:

Giá nhà ở khu vực Thị Nghè có thể bao nhiêu phụ thuộc vào diện tích, vị trí, kết cấu và điều kiện mua bán. Theo các kết quả tìm kiếm, giá nhà ở khu vực Thị Nghè có thể lên đến 187 triệu VNĐ/m2 cho mặt tiền, 112 triệu VNĐ/m2 cho hẻm và ngõ, 50.4 triệu VNĐ/m2 cho các căn hộ và 🧐 “4.25 tổ VNĐ cho lô góc[ “ -> I don’t understand

[covert sun]

#💭│ideas message

[covert sun]

Thanks for reporting! We'll check

[limpid tree]

The "popular now" searches are not relevant at all. Most of them are the same for a week+. I would either remove them or make them actually useful.

Also, any information about a smart auto complete function?

[vocal plume]

Why is this happening

[heady plover]

Are you using a VPN by any chance ? That may be why it is showing this error

[grizzled coral]

There is some way I can reach perplexity from code? like an API?

[celest aspen]

Please tell me your using a vpn, if not I would highly recommend not doxing yourself

[covert sun]

Hey @celest aspen. Thanks for reporting the issue. Correct me if I got it wrong, but you want to ask if you can use a SIM card from one carrier to use the services of another carrier. And the system gets confused that you're asking if you can use another carrier's SIM in your phone, right?
Technically, it is correct when traveling abroad, you don't switch the card but register in another carrier's network. And we get to the point with more follow-up questions https://www.perplexity.ai/?s=u&uuid=266ba0a4-3cfe-425b-a404-2bb1681d51ef, when we clarify the question and the situation to make the answer accurate. Also, if you edit sources now to save the accurate answer it won't have an immediate effect, but will help the models for sure. Please, let me know whether the thread I attached contains the answer you were looking for.

[covert sun]

Thanks for the remark about 'Popular now'! We are working on updating it more frequently and showing relevant results.

[celest aspen]

Oh I forgot you can edit sources, and yeah I did know that about sim cards, I was just trying to use it to send to a friend

because it seems to usually be much more concise and to the point than I ever can

[unreal kite]

feedback: No one likes Everyone tags.....

[heady plover]

💀

[heady plover]

💀

[grave swallow]

@tidal summit: english is not my first language, what does "fix" a water bill means?
"Fix" a water bill typically means to identify and repair the cause of an unexpectedly high water bill[1][2]. Common causes of high water bills include leaks, which can be fixed by finding and repairing the source of the leak[2][3]. The phrase "fix a water bill" does not have any relation to the idiom "don't fix what ain't broke"[4].

[grave swallow]

@grim robin: Price vs value
Price is the total consideration paid by the buyer to the seller, while value is a more theoretical, mathematical exercise[1][2]. Value is based on how useful something is to the customer[3], and Warren Buffet famously said "Price is what you pay; value is what you get"[4][2]. The difference between price, cost and value is that price is what the company charges from customers for its product or services[5].

[covert sun]

Hi @grim robin! You posted this query in the #👉│feedback-general channel. Is there smth wrong with the result?🤔 Could you please, clarify.

[limpid tree]

I've found this issue. Share button appears further right than the visible viewpoint and causes some widths to increase. This in turn makes the page scrollable horizontally.

[covert sun]

Thank you! We'll fix this.

[brisk cedar]

Love this, made it my default search engine!

[limpid tree]

In my opinion, the language model should get a little more conversational in order to help itself. For example last night I gained access to the new bing, and when it stumbled on results for different things using the same name, it asked me what specifically I need to know. I'm noticing the same on another platform right now. It explains how one result says this and the other says that. It doesn't just combine them and show them to me.

Just a thought.

[stray talon]

I suggest:

• Time out error (lots of the times questions just loads forever)
• Reload button (in case of errors? just in case)
• Cancel button (instead of waiting 5 minutes for the answer to load before asking another question)
• Retain questions/answers on the browser extension

[mossy tulip]

we are working on it, thanks for the feedback

[mossy tulip]

thank you for the feedback, we will look into it

[winged iron]

It's a bit all over the place in this thread. It's giving inconsistent dates, switching subjects in the middle of the paragraph without clarifying, and ending with an irrelevant fact
https://www.perplexity.ai/?s=c&uuid=b68fbdb6-1c53-4ead-ab7c-d4698302548c

[flint stratus]

Unable to answer this

[covert sun]

Thanks for reporting, @flint stratus! It was down at the time. Should be working now.

[dark flint]

New here, Can we see the history or recent queries in the future updates

[covert sun]

Hi @dark flint, welcome!)
#💬│general message

[winter flare]

Hi guys, i'm experiencing some problems with Perplexity when using my ProtonVPN. Unfortunately, the search engine is completely unavailable if the VPN is turned on. Is it intentional, or it is a problem with my particular VPN provider/configuration? Thanks

[mossy tulip]

some of the attackers are using VPNs to attack perplexity, so we had to block some of those IPs. unintentionally it might block some normal users. we are looking into ways to eliminate this problem. please give us some time.

[lucid rock]

Bird sql cannot search tweets after 2.12, does anyone know?

[heady plover]

In this answer (https://www.perplexity.ai/?s=c&uuid=2ad1b5f1-ddce-4669-b680-daa17c322ec4) it give different result.

[mossy tulip]

thanks for the report, I will look into it

[still pelican]

Love the new update. Is there a prompt I can give the extension to summarise the content on the page?

[mossy tulip]

you should be able to just ask "summarize"

[still pelican]

Thanks. Works okay but draws from other external sources too rather than just that page it seems.

[unreal kite]

Please use @ here not @ everyone.... minor feedback.

[mossy tulip]

it sometimes supplements the results with external sources when it is hard to parse a web page

[stray talon]

Awesome. The new update is also good for explaining code in github gists

I'd like to recommend taking inspiration from merlin (https://merlin.foyer.work). I've been using it before it went paid.

Compared to Perplexity, Merlin has several features that I also find really useful:

1. Shortcut to open it as a popup on a site
2. Preserve progress/history/questions when the popup is closed (but not the site).
3. Ask about a certain context from the website, not the whole, kinda.

[copper gyro]

this broken it workd time ago help

[heady sedge]

I've been trying to use the domain-referenced research function that newly provided by the Chrome extension on a Japanese website, but it keeps saying "No results found" and the Sources are also shown as 0. Can anyone help me with this issue?

[mossy tulip]

could you please provide more context?

[heady sedge]

Sure why not, I just visited to the Japanese Lawyer's website and tried to ask perplexity like "この事務所の債務整理における特色をまとめて教えて！日本語で"（Eng: What is the strength of this law office especially in debt consolidation service）. Then it returns that I wrote in ↑. But now at some of Japanese site works well but some of them doesn't. From what I've seen, this may be a temporary issue because during the 1 hour that I've been trying, there have been times when it's working properly and times when it's not.

[mossy tulip]

thanks will look into it, can you provide the website

[heady sedge]

https://www.fukuda-yamikin-soudan.jp/
here. Please note that this is all written in Japanese..

[burnt coyote]

I have been using Perplexity to do research on a new tech initiative at work and I've been very happy with its results. For the most part, Perplexity has been able to summarize an overwhelming amount of results into a few relevant ones. It saves me a lot of time. It also makes it easier for me to refine my queries by asking follow up questions. Great product yall!

[covert sun]

Thank you for sharing your experience and your kind words!🙌

[mossy tulip]

thank you for kind words, over next week or so it will be much better! stay tuned!

[primal wind]

Would you be able to make it so you can open the response sources into a new window without first expanding using 'View List'?

[keen fiber]

it will be perfect if we can set by default is view detailed instead view concise

[mossy tulip]

there are some latency concern for doing this, but we are working on improving concise answers, please expect some great updates soon.

[keen fiber]

Thank you so much, but we find that information from the detailed view will be more accurate than the concise view.

[stiff hollow]

Any developer api in the near future?

[stray talon]

Please don't make the extension close when clicking any link from there...
I need to retype the question to click the all the sources
Use <a>, not <button>...

[stray talon]

#💬│general message
#💬│general message

[mossy tulip]

thanks for the feedback we will take a look

[teal narwhal]

Hi guys, I've just discovered perplexity and its a cool extension. It works in my Brave browser. However I'm recently using new Orion browser (Safari on steroids) which allows using any extensions, and I have to say that perplexity does not work... it just get stuck in any prompt..

[wooden narwhal]

The extension is not yet supported on Firefox and Safari. We'll be releasing for those browsers next week!

[teal narwhal]

I agree, however Orion allows to use any extension, either from Chrome or Firefox store, so it runs as a extension, but does not give answer to any prompt

I'm just letting you know guys. I know Orion is not popular and still in beta 🙂

[wooden narwhal]

Our API currently blocks any traffic that's not from a Chrome extension. That could be the reason queries currently hang even though you're able to install it

The traffic should be coming from a Chrome browser just to be more clear

[teal narwhal]

but it is installed from Chrome store directly!

[wooden narwhal]

Let me DM you!

[teal narwhal]

no it's okay 🙂 I'm just letting you know 🙂

extension works on Brave btw

[wooden narwhal]

Ah yest Brave is also Chromium based and that's why it works

Orion is not Chromium based and hence the problem: https://www.perplexity.ai/?s=c&uuid=e72bb168-020f-45bb-8970-23123002aea6

Hopefully that should resolve once we extend the support to Safari

[teal narwhal]

oh! I understand now!

thank you for explaining it to me

[wooden narwhal]

Np!

[teal narwhal]

yeah, you can extend the support, Safari is also quite a large market share for browsers, and lacks any ai prompting extensions

[wooden narwhal]

Absolutely, thanks for sharing the feedback!

[teal narwhal]

have a nice day guys!

[outer heron]

the first thing i wanted to try was to make it my in browser search tool of a page of text, i loaded a page (https://gist.github.com/h3xx/1976236 warning: 100000 words long, 100000+ lines, takes long to load, its the top 100000 most popular words in english), then clicked the chrome extension, then navigated away because it was taking time to load, on coming back to the tab, its gone, i have to start over... do you have to stay on the page...?

Next i want to know, is the thing just using the neural network to do everything it does ever for everything, or does it see opportunities to save processing power, and do some programming solution which does not involve ML?

the request i gave it, was list me all the words which contain more than 3 letter "e"

its a programming type question, but one which an AI should be able to handle, i'm trying to test it out

#1 do you have to stay in the tab for the parplexity chrome extension to do its job
#2 is it exclusively using machine learning, or can it save itself processing power by using javascript for programming type problems it detects

[stray talon]

Try using https://you.com/chat

It's incorrect tho

[outer heron]

Sorry but i think that doesnt answer the question 1 or 2

im interested in perplexity

[stray talon]

Should've mentioned "on a related note:"

nah ignore everything I said

[outer heron]

try another app/extension?

ok i can do that

perplexity doesnt know that there are letters on the page:

i guess its meant to be doing what bing does, and not be an advanced search tool for page contents

[stray talon]

I have different result:
"The words in the Wictionary top 100,000 most frequently-used English words that contain more than 3 letter "e" are addressing, neat, sphere, assisted, avez, smallest, bend, telephone, extending, memories, sublime, cabinet, danced, inclination, conclude, sadly, King, sickness, ancestors, Produced, pine, shouting peasant copper companies trees could erect painter sinking belle Wie Service disclaimer seal accepted board serious spread ways del marked face spot Colonel twelve wonderful questions prevent expect camp move winter excellent additional America Michael tax consider fait car faire distribution example silver glass share century engaged Every forms class start ease lake signs teach count worship comment: 2501 - 2600 stock Ireland White quantity kinds Holy earlier good interrupted apart voices confess bed convinced valuable sell roof growth commercial aan articles pardon record replacement development stepped base treatment hotel spend fled attend witness remark mysterious European woman's audience safety eternal skill guns located Love bow readable fierce earnest kings shalt destroyed sail bill suis extended cattle pushed sounds King's plants temple admiration alle band links med dance derive throat Royal hit desperate Next crowded meal impulse available resumed writers notre voor conviction declare flame Russia Helen wird port ideal Jim Princess silence Constitution vie union possible avoir haste route dread requires stayed Alexander Bishop
"

completely incoherent tho

[stray talon]

Wait you're opening it here, not on the gists?

[outer heron]

youre mixing up two different comments

#1 glad your attemtp worked i would like to try again with your prompt

#2 the letter E thing on the discrod page was meant to be a simpler test

but first i tried it on the gist

i tried perplexity on the website, its smarter

i asked it how many legs has ariel got, and it responded with a reference to the meme, although it did say she had 2 legs

so... 50/50

it seems the default concise mode is dumb, and the detailed mode is accurate

[full venture]

Language models aren't very good at separating words into letters because of the way the text is tokenized, so they actually can't tell which words start with e very well.

[outer heron]

so it never uses programming tools to save CPU/GPU time whatever, it always uses ML/AI

ouch

[potent grove]

I've been noticing this as well. Once I got it saying it couldn't answer my question, but when I went to detailed it answered it anyway. Sadly I lost the link because I did it in the browser extension.

[outer heron]

https://www.perplexity.ai/?s=u&uuid=02d810ed-5bf0-4a1a-9506-eaebda187c36

click detailed on the challenged points and it suddenly gets them right

it hallucinates to be concise

you have to scroll up on the link

[mossy tulip]

thanks for the feedback, we will look into it

[potent grove]

Not sure entirely what logs you keep or if this is useful, but I was telling it to search from https://pomsky-lang.org and was asking to to tell me where the docs were.

I can't seem to replicate it anymore, but then again I don't remember the exact prompt.

[potent grove]

Here's a case of the concise answer being misleading because of missing context. Even the reference preview takes the article out of context in a bad way.
https://www.perplexity.ai/?s=u&uuid=949c1f88-eaf8-4ba0-b02a-dcc68db1080c

Concise:

Electron apps are slow because they are essentially a browser without the user interface[1].
```Detailed:

Electron apps are not inherently slow, but they can be slower than a finely tuned native app due to additional abstraction[1].

It is slow ... At its core, an Electron app is a browser but without the user interface of a browser. It is just the renderer, and just like a web ...

Let’s take a look at some of these arguments:

<h2>It is slow</h2>

At its core, an Electron app is a browser but without the user interface of a browser. It is just the renderer, and just like a web app in a regular browser, it uses HTML, CSS and JavaScript to build an interface and provide functionality. Those are a lot of extra layers. And because of this additional abstraction, it can be slower than a finely tuned native app. Sure. But slower is not slow.

It looks like it's doing cherry picking of words that affirm what the question implies is true, even if it isn't. While the detailed answer gives what the article _really_ says.

In my experience so far, the detailed answer is always very high quality, but the concise one isn't always.

[full venture]

Thanks for the feedback, I agree, we have improvements in mind.

[celest aspen]

Could the ask follow up question be slightly more similar to a chatbot? Not dissimilar to you.com

[mossy tulip]

yes we are working on it, hopefully will be out very soon

[celest aspen]

Awesome!!

[brave bough]

Can perplexity read local pdf files like bing? I get this error using the chrome extension Failed to fetch the content of the webpage. Please make sure to use in-page search only on public webpages.

[pliant osprey]

Hi, just wondering how did you create the role selection screen?
Is it something that discord did? or it's unlocked by bosting the server?

[covert sun]

Hi @pliant osprey. It's one of Discord's features.

[full venture]

Hi jjboi8708, we do not read local pdf files: only publicly available webpages.

[burnt coyote]

This is an interesting challenge. Someone on SO gave a bad/incorrect answer to a question. Perplexity will report on the incorrect answer. What is the interesting challenge here is that if you read SO carefully, you'll note that the source admits that the answer was wrong. More in thread...

Actually, I don't know how to do threads in Discord...

https://www.perplexity.ai/?s=u&uuid=22cca50c-7ffa-42e8-af12-2eea48d79b38 Original question - "In rabbitmq is it possible to for a consumer to use a single queue, but selectively receive messages from that single queue?"

Answer is "Yes, it is possible with a citation"

https://stackoverflow.com/questions/11142071/rabbitmq-selectively-retrieving-messages-from-a-queue

The source admits that the given answer was actually not correct:

[brave bough]

Will you guys add the ability too?

[full venture]

We'll look into it, we are currently only indexing the open web because we want to respect privacy and permissions.

[brave bough]

Ohh okay. It would be awesome if you guys gave it that ability because bing can do it but I always need to restart the topic because of the limits 😭

[potent grove]

Oh, you're indexing yourself? That's pretty cool. Will there be a way to request indexing of a domain in the future like on Google Search Console? I tried to test Perplexity on some of my own websites, but it only had one indexed.

[brisk cedar]

@full venture https://discordapp.com/channels/1047197230748151888/1047649527299055688/1079148306875301898
Regarding the codeblocks in perplexity:
Do you guys work on syntax highlighting and differentiating them with prefixes like py, js, bash, etc. as you can see them in markdown for example?

[covert sun]

We'll consider having such a service in the future, for sure, thanks for bringing that up!

[full venture]

currently the syntax highlighting is basic (default python) but will work on differentiating it by language, thanks

[quasi bay]

Does anybody get the infinite cloud flare verification loop? I get it on pretty much every single page reload but here is a catch. It only happens on my android tablet (using Edge) and only when desktop pages are requested automatically. It doesn't happen on my android phone (also using edge) or on my windows PC. I tried this with both VPN on and off and VPN is certainly not the issue here.

In fact, after verifying myself once and getting back to the verification screen, all it takes is clicking on settings and loading the mobile version of the site to get to the homepage.

Unfortunately, as soon as I leave the browser and the page needs to be reloaded, edge reloads the desktop version thus forcing me back into the verification loop

Just some info that's hopefully useful

[brazen nova]

Love the app. I tested a few common medical queries and I definitely see the potential to create an AI medical assistant

[heady plover]

it does not answer long contextual question.

rest all good.

[brazen nova]

I would love the ability to select sources prior to searching and have a default collection of sources to search from.

[fringe pawn]

I have been trying you.com and perplexity for few weeks now replacing google search. I also tried few other ai search engines like neeva ..etc. Without doubt the one i keep using the most now is perplexity. All the others have hallucination problem where sometimes they will provide sources but that source wont have what the ai said. Perplexity so far has been the most accurate and useful for me. So kudos for that. 2 things i would like to see:

• Live info like sports score ..etc perplexity has some issues.
• perplexity is my home page now. But there is reload happening often if i take a break for 30 mins or if i want to start a new search for eg. Some way to avoid it like google does where new search can be done without reload would be great.

[brisk cedar]

you could try to set your default search engine to perplexity with the get parameters for the search, instead of setting it as homepage

to fix the timeout issue

[fringe pawn]

Ok, that works 👍

[fair locust]

I've been loving the search on this page/domaine feature, thanks for implementing it! It would be neat to have 'search the highlighted' option for long articles/papers. e.g. sometimes I found myself wanting to know more detailed summary of a selected paragraphs of content instead of a short summary of the whole article/paper. (I've also tested view detailed vs concise, most of time the level of details presented are similar to each other so far)

[covert sun]

Thanks! The last sentence is the result of calculation of the numbers found in the sources above it.

[mossy tulip]

great idea, we will look into it

[copper panther]

Hello and thank you for letting me join this group. I'm new to discord. I'm not exactly sure how it's supposed to work. It works with chat GPT. Check GPT is a new game changer. It revises my words that I want to use. Then again I have to make sure it is accurate. Chat GPT can have some inaccuracies. It's always good to check after it revises a paragraph or a description.

[tight nymph]

im using perplexity as DE FACTO fact checker

[river junco]

1. Définition de la veille technologique

[slender barn]

When executing a search, it encounters some sort of malfunction. - I've been having some issues with the Chrome addon recently.

It's like going in a loop trying to parcel out the info 5-7 times with no success. - Website version works smooth.

[covert sun]

Thanks for reporting, @slender barn! Could you share the page that generated this loop and gave the error? Did it happen with other pages? Is the extension working fine now?

[wooden narwhal]

Hey @slender barn, this should be fixed soon! Sorry for the inconvenience

[slender barn]

Hey @covert sun , I'm not sure I can be a help here - because it didn't generate the link that could be shared. No I believe it happened only on one page (with different prompts). Just executed a couple of searches now with the same prompt and it works perfectly for now. Kindly let me know if I can contribute with other info.

Hey @wooden narwhal , works smoothly for now, thank you guys for your efforts, I'll share updates in case of troubles. Appreciate.

[runic turret]

How to config language of Perplexity?

[carmine sage]

This is the Chrome extension. What is this error?

[covert sun]

Hi @carmine sage! Thanks for reporting! Our team is aware of the problem, it should be fixed soon.
#👉│feedback-general message

[covert sun]

Hey @runic turret! You can ask Perplexity questions in your language. No additional settings needed. Sometimes replies can switch to English, we know about this and are working on improving models answers, too.

[gaunt flame]

Hello! Is possible to add the bot to one of mine server? How can i do? Thankkss and good job with Perplexity! 👍

[mossy tulip]

not yet, should will be soon, please stay tuned

[lucid current]

Error occurred when i tried to see detailed of the result.

[celest reef]

Hi

[covert sun]

Thanks a lot for reporting this @lucid current! Could you share the permalink with the error. Does it still show this error when you try the query again?

[lucid current]

https://www.perplexity.ai/?uuid=935529d7-8d6e-498b-9975-60739c349161, It still shows the error.

[covert sun]

I checked that detailed version works fine on the site and in extension. Happens to the /ask bot question when you proceed to the website via permalink and detailed version is not available. We'll fix the issue soon, thanks again for reporting.

[round spindle]

does the chrome extension work on arxiv? I keep getting

[mossy tulip]

it does work but arxiv has strict rate limits, so if many users use it it starts rate limiting our IPs, we are looking into it

[tawny coral]

This is way better and faster than bing ai

[true shell]

summarize

[tawny coral]

Hmm... I don't care for trump or biden, but I care that a search engine won't show me what I'm looking for (for whatever reason)

Maybe I'm writing a book. Maybe I'm studying dictators. Maybe I'm studying history or public opinion. Maybe i'm the other 50% of the country that voted for the 'wrong guy'. Maybe I'm the 50% of the country that voted for the 'right guy' but want to see how the other 50% think?

But when an AI starts to act like a purple haired librarian... 🤦‍♂️ hiding books from me that are clearly there...

But otherwise, Perp is amazing

I think it makes sense to hide information that could cause physical harm, such as recipies for explosives and so on, or poisons

But an ideal search engine otherwise shouldn't have an opinion one way or another

Because otherwise, you'll end up with a council as arbiters of truth/ correct opinion - who are human and like it or not, either biased, bribed or compromised in some way

[mossy tulip]

thanks for reporting this, this is a very interesting example, obviously not our intention at all. we will investigate

in general though we are not a chat bot and not optimized for queries unrelated to search, for example "write a poem"

[tawny coral]

That's very great to hear - I'm really loving it and recommending it to my friends

Best search engine - I've stopped using google/ bing, tbh

Also, this linkable to the search result is amazing. This way, you can share with people or in projects you are doing: https://www.perplexity.ai/?s=u&uuid=7172ec7b-86ce-410a-9dbc-cb9e2b4bc2f0

[mossy tulip]

thanks for the kind words, it will only get better, we have some exciting things in the pipeline

[hushed cipher]

I'd love to join the beta group if and when you @cunning goblet team opens up the API 👀

[grave swallow]

Please try slash commands!

[rotund summit]

i love this AI

[burnt coyote]

https://www.perplexity.ai/?s=c&uuid=46bbcf44-2496-4907-89f0-05192169fdef This is close to working out. Note that it includes markdown in the output, which is not user friendly.

Note that the code in the #🔎│ask channel handles the output gracefully 😉

The images lead to broken links. But I suppose it's the fault of the underlying search data, or poor documentation from MS 🤷‍♂️

The first cited link (azuredevopslabs) has relevant screenshots. So it's interesting to see if the system can pull those screenshots...

[mossy tulip]

thanks for reporting, we are not rendering images yet, but we are looking into it

[burnt coyote]

It's almost there though. It does kind of include the desired outcome in the results.

Magic of LLMs

[dusky lark]

Is it possible to get the previous message "related" prompts back? sometimes I'd like to check more, but after choosing one the others disappear

or does it make things complex if they wouldn't disappear?

they are so good that choosing one gives me FOMO 😊

[mossy tulip]

the reason we did this is that otherwise it makes it confusing in term of UX, but I see your point

[dusky lark]

yeah, I thought there would be something like that - currently I try to delete the last answer as a workaround

also is it possible to save threads? Like doing research on different things?

[tawny coral]

One suggestion

When you share links, it scrolls to the bottom

https://www.perplexity.ai/?s=mn&uuid=1e324c0a-0ce5-4a1f-851d-dcc02d8a7253

I think it's better to scroll to the top of the page?

But I love how easy it is to research with this

Would love an Android app.

Hmm can't find video content

Perplexity AI: Show me some video tutorials on DaVinci resolve https://www.perplexity.ai/?s=mn&uuid=91481e32-837f-448b-9a1f-f498d5cf6fa8

[full venture]

Should be fixed now, thanks!

[tawny coral]

That was fast. I just tested but it seems to still be scrolling to the bottom instead of top

https://www.perplexity.ai/?s=u&uuid=f40cfe30-4b44-4961-8586-8ee2c5b20840

But maybe it hasn't propagated to me

[full venture]

Sorry the deploy is rolling out, one moment

[tawny coral]

Amazing

[full venture]

Defintely also good ideas re: android and better video support, we'll look into those next

[tawny coral]

I was thinking that the web interface is still really nice even on mobile. Maybe an Android 'app' can just be a simple browser that loads perplexity. (So no need to have something complex)

Another suggestion

I hear there is a chrome extension but I haven't seen the link to it on the site

More people would probably install if they saw the link

[full venture]

oh it's in the top right corner on desktop but it's hidden among the other links

[potent grove]

Sometimes when I try to open the extension popup it's completely blank. I don't know why I haven't thought of this until just now but next time it happens I'll inspect the popup page and grab the error if there is one.

[tawny coral]

Confirmed scrolling to top!

[dusky lark]

Can somebody explain this to me? I'd like to chat, but this was unexpected 🙂

[acoustic star]

I wish we had a greater query character limit than 255. Sometimes I just need to search about some longer pieces of texts to try and find their sources, but the character limit makes it impossible. Does anyone have any idea why is it exactly 255?

[full venture]

Thanks for finding this, should be fixed now!

[dusky lark]

Thank you. So... it can chat? Easter egg? 😉

[acoustic star]

I am curious if it is really not using search results

[acoustic star]

I guess 😃

[limpid tree]

Greek answers cut off abruptly, mid paragraph.

[tawny coral]

Another suggestion, the text returned by Perp could be formatted

vs bing

[tawny coral]

The bold and bullets make it much easier to read

[mossy tulip]

thanks, we will improve this

[tawny coral]

Here is another example of Perp 'hiding' information

It seems unable to play devil's advocate

Compare with Bing

Bing is giving information (whether 'true' or 'false') and supplementing it with an opinion or fact check

Perp seems more leaning towards what authorities consider 'correct opinion' - I'm not a particular believer in objective truth since everything is hazy and in flux

Bing gives the 'correct opinon' but also goes - "on the other hand, here is the other side of the argument" - which is more helpful, I think

"Why is Russia on the right side of history"

Bing is more balanced giving both sides of the issue

In truth, right side of history depends on the particular flavor of propaganda we all choose to follow. But at least Bing is serving both sides seemingly equally

[mossy tulip]

thanks for sharing these examples, we will look into it

[limpid tree]

Is might be because of the new GPT-3 turbo, might also be because of prompt tweaking, but I have to say that this is now resolved! No more combining and false information from different search results. Kudos 🎉

[acoustic star]

Omg I just can't get enough of how fast it responds. Everyday you guys are making it faster. What a great experience. I love it!

[mossy tulip]

hopefully it will be even faster in a couple of weeks, we treat latency very seriously

[acoustic star]

Very excited! I have never thought that such a small team can deliver such a product and other major updates this fast. As a software engineer, it makes me appreciate every bit of it!

[dusky lark]

Ok, I know there are many great and awesome features, but I'd like to especially thank you for the follow up questions - while Bing limits the chance to go deeper into a topic, here you can wildly explore related, interesting and useful information 🫶

[mossy tulip]

thanks for kind words, we recently improved related questions so they are more relevant now

[dusky lark]

awesome, I really love it - and would it be possible to reroll related questions exactly like how you can reroll "popular now" questions?

[mossy tulip]

ah interesting idea, yeah it should be possible will add it to our future features list

[dusky lark]

thanks!

[limpid tree]

Is there a solution for the weird characters appearing instead of words for the Greek alphabet, or it's a core issue with the model used?

Also, many times paragraphs end abruptly (token problems?)

[limpid tree]

Just noticed that it says 2021. The answer is correct, just the year is wrong!

https://www.perplexity.ai/?s=u&uuid=58479d1d-86ab-4cd4-8861-90add54adb77

[potent grove]

It would be great if the links for the sources were real anchor elements in the DOM. Right now as divs you can't right click to copy the link. I also like to right click and open in a new tab to ensure links don't override my current tab (even when I know Perplexity always opens in a new tab) when I might be clicking multiple from one page, and this hampers that muscle memory.
​https://gazou.kyza.net/317bffa6d08c4fe2.png

[covert sun]

Hi @potent grove, thanks for sharing! At the moment right-click menu, where you can copy links and open them in a new tab, works when the sources section is expanded.

[potent grove]

So it does. Thanks for telling me.

[modest kestrel]

Can you create a strategy on pine script where the main indicator is harmonic pattern with smart money concept and block order & all price action like Bearish and bullish Chart Patterns & Bullish Flag Chart Pattern & Double top and Bottom Chart Pattern & Elliott Wave Chart Pattern &Falling Wedge Chart Pattern & Head And Shoulders and inverse Chart Patterns & Inverse Head And Shoulders Rectangle and flag Chart Pattern & Rising Wedge and Triangle Chart Pattern & Triple Bottom and top Chart Pattern & Volume profile Auto Candlestick patterns with Fibonacci Extension / Retracement / Pivot Points

[covert sun]

Browser-extension-feedback

[stray talon]

Or maybe make a forum?

Also that thread is locked

I don't like that the "summarize" button loses progress when the extension window is closed.

Example of good extensions that doesn't do that:

• Merlin
• Mendeley

[mossy tulip]

thanks we will address this

[haughty geode]

when you ask for an inappropriate/illegal answer, the sources for that inappropriate response are still shown (only for the first question in a thread)

e.g.

[cyan musk]

Regarding the new feature summarise, the plugin runs perfectly but when I want to see the summary in a new tab instead of the Chrome extension, it loses the summary created for the webpage where you created the the summary with the Chrome extension. I sound repetitive I know but it's to make the point clear. Cheers.

[mossy tulip]

thanks for the feedback, we will track this issue and address it

[hidden adder]

I have a problem: even if I enter a query in a language other than English, the answer still comes in English. Only after that, if I send the next message asking for an answer in the language I want, it will reply in that language.

[limpid tree]

While speed great, I find that sometimes an answer takes even 5-10 seconds. Is this something GPT related or back end related? Could it also be the server you use(EU S)?

[autumn current]

Hey @mossy tulip I recently used your Crome extension and noticed that the UI/UX can be improved. I am pleased to inform you that I made some changes help navigate through the extension more easily and improve their overall experience. please let me know if you are interested in seeing my changes. I will be happy to send them over 😁🙂

[mossy tulip]

thanks, I replied to your email

[toxic current]

Oui

[autumn current]

@mossy tulip thanks for your precious time 🙂😁

[autumn current]

Hey @mossy tulip I have fixed some changes on my previous mock-up of your Crome extension. I look forward to hearing back from you soon.

[mossy tulip]

thanks for kind words!

[autumn current]

Hey @mossy tulip check it out Perplexity Apk concept 🙂

I hope you like it !!

[gilded idol]

Hi

[dusky horizon]

next to summarize there is an icon with a chain of 2 elements. What is its function?

[signal owl]

i keep asking questions, but no respond. is this a server issue?

[covert sun]

Hi @dusky horizon! You can copy and share permalinks with your search results. When you press the button the link is copied and then you can paste it in a post, a tweet or message.

[mossy tulip]

copying a permalink so you can share the answer

[stray talon]

Retain suggestions/related from previous prompt results instead of just showing the latest one. Sometimes there are multiple good suggestions that I'd also like to send.

[covert sun]

The sharing button with a chain of 2 elements - when you get the answer.

[somber aurora]

Hello Perplexity team! I'm a member of the media and have a request re: an upcoming event I'm planning and would love to have someone from Perplexity be part of a panel and demo. Who can I reach out to?

[covert sun]

Are you referring to the web or the extension version? There is a "Summarize" button in the browser extension interface.

[heady plover]

Hello, when is an Android app coming?

Also, can you add chat history like ChatGPT?

[clear imp]

hi everyone 😀 really nice tools 👍 good job

[mossy tulip]

yes stay tuned

[scenic cedar]

Question: Is there a way to select languages other than English with the summary button of the Chrome extension? As a Japanese speaker, even if it's a summary, it takes me time to read in English.

[mossy tulip]

not currently, but we will add it soon

for now you can try to ask it to summarize in Japanese

[heady plover]

Hi

[heady plover]

when i followed this tutorial:https://www.youtube.com/watch?v=usgqmQ0Mq7g&ab_channel=Aitrepreneur , i got this error ,can anyone help me?😢

After i clicked this ,it didn't run the SD,just show this error..

[scenic cedar]

thanks!

[dusky lark]

how can I make it more strict on dates: some of these are from February or January https://www.perplexity.ai/?s=c&uuid=0bbbb059-8146-4596-9652-d748bca25d6b

[old pagoda]

A glitch found. In this Detailed thread evoked through PerplexityAI bot, ref 5 failed to be embodied while can be opened as a reasonable source. https://www.perplexity.ai/?uuid=0f7c1e0c-c07b-4946-b5a5-df51f8fe1d70

[floral nova]

Want to highlight for one-click summarize that everytime I click view detailed it turn out to be shorter

[heady plover]

Is there a plan for an Android app though?

Would be great 👍

[unkempt atlas]

It'd be great if the bot could link sources in some different format like this one. Currently the links can be very close to each other making it hard to click one, plus it's always good to know what site you're going 🙂

[glass dawn]

Is there any group in Spanish?

[slim yacht]

First -- thank you for an awesome service. 🙌

I noticed two issues (My Query:https://www.perplexity.ai/?s=u&uuid=978e8084-7465-4c07-875c-63b0286d43e7)

1. some of my answers are summarily cut off mid-stream
2. View Detailed often ends up being significantly shorter than the original answer.

Another issue:
When I use the extension to contextualize my query to the current domain it works as expected. However, when I click on any of the related queries at the bottom, it loses the domain context, and answers in general. This may be useful in some cases, but if you are trying to dig down to a useful answer, it is frustrating to see the scope widen unexpectedly. Those related query links also exit the extension and go to the site instead -- where there is no UX to contextualize answers to a domain or page. To achieve my search goal, I have to click the related question, copy the question text, come back to the site I queried on and do the contextual query again from there.

[covert sun]

Thanks for reporting, we'll check this issue!

[covert sun]

Thanks for spotting this, @old pagoda! We'll investigate.

[covert sun]

@floral nova, ありがとう for bringing this up! Our team is aware of the issue and is working on it.

[covert sun]

Hey @glass dawn! There's no separate Spanish group on the server, but there are lots of Spanish speakers, who you can talk to or ask for help here. You can also write your feedback in Spanish we'll translate it anyway.

[heavy forge]

@elder harness: polska po polsku
"Polska po polsku" can be translated to "Poland in Polish". It can refer to various things, such as a language learning project called "Po polsku po Polsce"[1][2], a Polish-language dictionary[3], or simply the use of the Polish language in Poland[4][5].

[south ether]

Hello everyone,

Thank you for your interest in providing feedback to Perplexity. We are always trying to improve the product experience for all of our users. As such, we would like to invite you to participate in a 15-minute chat with us to provide your feedback directly.

To participate in this interview, please fill out the form below. I will contact you directly to arrange an interview at a date and time that suits you.

Your help would be greatly appreciated, and I look forward to speaking to you soon.

Daniela
User Research at Perplexity

https://perplexity.typeform.com/interviewform

[unkempt atlas]

Link doesn't work because of the brackets

You don't need that formatting on discord

[vapid mason]

Your discord COMMANDS no longer work - /dream is no longer asking "about', and /ask is giving a red error " The application did not respond "

[primal wind]

See "anouncements" section, they deprecated /ask

[radiant kindle]

Hi. Discord formatting is not the same as standard Markdown, and links don't work like this. The URL in your message got messed up, and clicking on the link leads to an invalid URL (and then redirects to Typeform home page). Please fix.

(For those who need it, here's a link you can click: <redacted> Go above for the updated link! )

[dusky lark]

fair idea, but I'd still like to see the results even if there is one - maybe some note or a warning?
"Please, try to rephrase you query, as there are too small amount of results, so the answer may not be accurate" 🤔

[radiant kindle]

I can't reproduce your situation (it gave me a slightly more detailed answer with 3 sources), but asking it for more sources in a follow up question did make it give a few more sources, and it explained the sources in the output.

(for reference, my follow up prompt was Can you give me more sources to support your claims?)

[rugged whale]

I keep recieving Error in processing query

[old mountain]

me too

[versed musk]

Hello there! I have the same issue: "Error in processing query"

[radiant kindle]

back end likely died or something

[hearty frigate]

I'm receiving "error in processing query" too

[heady plover]

It's back up now

[radiant kindle]

Wonder if we can benefit from some sort of way to have people to talk in other languages, like an "other languages" channel, or language-specific threads/forum channel posts

[radiant kindle]

... Perplexity seems to cache responses and return the cached response whenever a similar question is asked... despite I made a meaningful change to my question. Is there a way to bypass said cache, other than rewriting my question to be completely different-looking? Alternatively, may I know how long it holds onto said cache?

[halcyon nimbus]

I can't add the discord bot to my server, It gives me this error

[glossy zealot]

Same

[mossy tulip]

sorry about that, discord imposes a restriction that a new bot cannot be added to more than 250 servers in the first few weeks, and we reached this number already. we are trying to work with discord to get this resolved.

[steep tree]

why only in chrome, what about firefox

[dusky lark]

not sure if other reported similar but the response get cut off https://www.perplexity.ai/?s=c&uuid=8d0bed28-530e-4a71-a009-6f95c5d41f8d

[safe wing]

If a response cuts off, say "continue with the text".

This respones look normal, you only need to click on "view detailed"

[dusky lark]

when I make a follow up I can't delete it - so I can't go back to use recommended questions
I still need the choice to keep the recommended questions kepping visible so I can continue from anywhere anytime

[radiant kindle]

It is exactly the DETAILED that was cut off

[radiant kindle]

^

[elder barn]

can it not make a proper table right now?

also can we not edit out the sources anymore?

[elder barn]

also it seems like perplexity does make few errors when it comes to numbers. For example Google Pixel 6a 5g has 6.1-inch display not 6.2. I know it's not that big of a deal but still thought of sharing it here as it might help the developers to improve the AI. It has also mentioned that google pixel 6a 5g has dual camera (50MP + 12MP) but that is not factually correct. It has 2 12 MP cameras

[radiant kindle]

It seems to output tables in a format known as "Markdown". However, markdown tables is supported by neither Discord nor Perplexity Web :-/

[safe wing]

it is a table in markdown. But discord dosen't support markdowm and the web of perplexity also dosen't support it

[safe wing]

It uses the internet sources, look at the sources

thats why there are here

[safe wing]

you can, https://www.perplexity.ai/?s=u&uuid=4ed20a0e-0194-4211-9ce2-40945e4fefd5 , ask him for it

to not include

here without saying it you see the sources next to the text: https://www.perplexity.ai/?s=u&uuid=830d9ea5-e4cb-4b68-bc3f-729dccb00ad5 (click on link lol)

sometimes buggy but

[elder barn]

It still doesn't make proper tables, even without asking it not to mention sources link

[elder barn]

Then where am I supposed to see the table? I usually just copy paste the response from perplexity into Bing ai and ask it to make the table

[radiant kindle]

Sources is not the thing that breaks the table.

I'm gonna ask for Markdown table support for you in #🌐│feedback-web

[fringe pawn]

If u paste it in any markdown viewer you will see table output. Search for online mardown viewers and paste there

[elder barn]

thanks

[radiant kindle]

I really, really wonder if the bots should be disabled for channels outside of #🔎│ask and #🎨│dream .

[limpid tree]

Greek answers cut off. This is probably a token issue, since not all characters are the same size. Can you look into it? It's a shame having to click detailed answer just to get an "almost" complete answer (because at one point it always starts typing greek words with english characters!)

Example (view detailed): https://www.perplexity.ai/?s=u&uuid=114dd1c2-d9dd-4280-b0e5-816b9662c16a

[safe wing]

Thats also not the anwser i provided you, you see that i just wrote don't include sources 🤦‍♂️

It was based for your anwser "#👉│feedback-general message" can't we edit out sources?

Here he is just not including sources, as i mentioned, he can't make Markdown as i also said

[elder barn]

ooh 😅 Sorry for the confusion

[safe wing]

Np :D

Didn't mean to offend or smth :)

---

Hello, today I would like to give my feedback on perplexity. Personally, perplexity has helped me a lot as a student, because it not only does an output (like chatgpt) but also adds sources. I can check these sources manually and verify their authenticity, which is very important as a student for research or otherwise. The user interface is very easy to use and I love that I don't always have to log in - as you don't have to register. Perplexity helps me with a lot of things and it's a great tool, I hope the development continues!

📚 The only thing I would improve about Perplexity is the Markdown problem: when Perplexity inplememts a table or Markdown type text it will NOT render in Markdown, which is a shame since the text becomes buggy! I would like that its going to be fixed - but its not a fundamental thing.

However, all in all, Perplexity is a very helpful tool for me, thank you for the development 🤖

[mossy tulip]

thanks, yes markdown is our top priority, we will address it very soon

[radiant kindle]

• A #1083863273566371961-discord channel for feedback for this server rather than the AI
• some limitations on the usage of the bots outside of #🔎│ask #🎨│dream , also some rules/guidelines to discourage using the bot outside of these channels
• some help info on how to use the bots, either in a channel description or an info/announcement channel. More specifically, mention these because people often get them wrong: ping bot to get text response, /ask doesn't work, /dream is for getting images not text.

[dusky lark]

can the site do a summarize? Especially in other language?

[covert sun]

Hey @dusky lark you can ask questions in your language on the web version, too, and if sometimes it responds in English instead of your language, you can specify it like: en Espanol, 用中文 or some other language you need.

[dusky lark]

ok, I'll try - and can it summarize? so putting in a link of an article? I can't install the extension everywhere

[radiant kindle]

... You can try, but this one test result wasn't satisfying.

But at least it could potentially do something when you give it a URL, which is already half not bad...

[covert sun]

yeah, you can try to just input links after "summarize".
got a better result than in @radiant kindle's message, you can experiment with it.
here are my examples, in English:
https://www.perplexity.ai/?s=u&uuid=a57dc04b-2935-4577-9baa-fa84796fe63
and in Spanish
https://www.perplexity.ai/?s=u&uuid=4ef1f732-7820-4102-b8b4-b063a43f395a

[dusky lark]

hm, not bad, thanks, I'll try it

would it be possible to ask in language X the search would happen in English then the answer would be in language X?

[radiant kindle]

Ha

[elfin gyro]

You all aren't verified? How are you in 250 servers then?

[radiant kindle]

The bot isn't verified, and the bot is in 250 servers and can't join more.

[mossy tulip]

thanks, we are looking into this

[mossy tulip]

we sent a request for verification to Discord, hopefully it will be approved soon

[snow cedar]

Hi, can we have a conversation history? Great idea overall

[radiant kindle]

For now, looks like you'll have to keep a share link when you end a conversation.

[limpid tree]

The viewpoint scrolls past the starting point of the answer (when tapping enter)on edge mobile. On chrome it scrolls less so the whole answer is visible.

Also noticed that you show text all at once today, like bard. I'm not sure if it's better imo.

My last point may be irrelevant since now it loads text word by word as it always had 🤔

[edgy umbra]

I’ve seen multiple messages on this but no answer yet— is there a plan to offer API access?

[lucid current]

When i see detailed results in Korean, they show unfinished sentences.

[heady plover]

I use perplexity more than Google now. No complaints so far. I really like that it can give me answers from the deepest corners of the internet. The old archives which are still written in HTML and no CSS lol. I was looking for past test papers and was not able to find them on Google but was able to find them using perplexity, I am presuming because of the same reason. The archives were like ancient websites so Google did not prioritize them. So, yeahhhhh I'm really glad perplexity does not use that weird system and gives me the accurate answers that I need. Okay, sometimes it may confuse itself in thinking that a certain website is related to something even though it is not. For example, I searched for "algebraic geometry" and it recommended me ACT's website which had keywords like algebra and geometry but that is not similar to algebraic geometry which I was looking for. But I am pretty sure that can be fixed! Maybe adding a shorthand key to toggle between images would be good too but I can see the pros and cons of that. Not even complaining about it. PERPLEXITY IS SO GREAT I WISH THAT I WAS THE PART OF ITS DEVELOPMENT TEAM 🥺

Can I really get an internship lol

[vapid mason]

It cannot do math reliably:

2164663517184000 byte to TB
PERPLEXITY
2164663517184000 bytes is equal to 1.964 TB (terabytes)[1]. This can be calculated by dividing the number of bytes by 1,099,511,627,776, which is the number of bytes in one terabyte[1][2].

(Correct answer is 1968.75TB)

You should train it to recognize math-related questions, and tell it how to use tools like these: worlds best online math tool free
PERPLEXITY
There are several free online math tools available, but some of the best ones include Desmos, GeoGebra, and Microsoft Math Solver[1][2][3]. Desmos is a website with interactive math activities and a graphing calculator, while GeoGebra is a suite of free math tools used by over 100 million students and teachers[1][2]. Microsoft Math Solver is an online math solver that provides free step-by-step solutions to algebra, calculus, and other math problems[3]. These tools are recommended by teachers and are accessible to anyone with an internet connection.

[radiant kindle]

What's underneath Perplexity is a language model, which wasn't exactly designed to do math.

[vapid mason]

@radiant kindle I know (I'm an expert), which is why I suggest that it be coded to detect math-related things, and hand-off the work to those, rather than spew incorrect results. This is a MAJOR problem in all NLP's and also in Google search engine itself - so solving this here in Perplexity would give it an obvious level of superiority to those competitors.

Also, related - Perplexity staff: You're not making use of the brains of your users enough - you need to be USING us humans to help train and direct your AI without wasting your own time. For example, anything simple (or even complex) that can detect math-related questions should ideally be asking us if we just asked a math question, and if we click "yes", hand off the work to a math engine (and same for units conversions, weather/emergency/other situations that might yield wrong answers, and so on). Work smart, not hard!

Sometimes, experts need to ask questions before they give answer: Perplexity.AI is not doing that. e.g. legal questions probably require asking jurisdiction, and many other things probably require additional context to give accurate results. Asking us so you get it right first-time is much less dangerous to users than giving them possibly improper answers.

[lone gazelle]

"Dangerous," in my mind, would be relying on an AI to do all the thinking for you.

🤷‍♀️

[mossy tulip]

we do have a calculator plugged in, unfortunately it didn't trigger for your query but if you ask your question slightly differently it works:
https://www.perplexity.ai/search/ad9ea706-ba49-4473-b9c7-89545dfda461?s=u

[teal siren]

Is it possible to summarize emails? I tried a few things and could not get it to work. (The emails are too long to be copy/pasted as a query)

[covert sun]

hi @teal siren as for summarizing your emails you can try it on Discord sending pm to Perplexity bot. They will fit as messages on Discord, I assume.
#💗│sharing message

[vapid mason]

@mossy tulip - excellent example of why you should be making use of us users a lot more: I could have "trained" your calculator to understand when it should have been used - had you given me a way to do this. Same for millions of other professionals who get wrong answers. Imagine if your code only got it wrong some limited number of times (enough to draw consensus among trusted users) and then forevermore understood how never to get it wrong again? Us humans are the resource you need to exploit to make that happen... (remember the saying: if it's free you are the product). Also, while I have you ear, it's worth keeping in mind how PayPal became successful: by utterly fanatical concentration on market-share above everything else. Your team asked me about paying earlier - you probably need an all-company meeting to decide whether or not you want your company to succeed, because destroying your userbase with paywalls (and wasting staff time even thinking about revenue) is a pretty quick way to destroy market share. The value of your business is our eyeballs, not our wallets.

[mossy tulip]

great point, we are actually moving towards this direction where perplexity will be personalized and working together with users, please give us some time 😄

[radiant kindle]

https://www.perplexity.ai/search/4e7fc48a-afeb-4621-833c-660ef4c8ff88?s=c
Kinda wondering if this is a bad thing or not - for the last two questions of the thread, I asked about a particular limerick and its author, but for both responses, it also explained what a limerick is, which I feel kinda redundant for what I asked...?

[versed musk]

Do you plan to have broader access to OJS, EBSCO, JSTOR, DeGruyter, and other similar scientific services?

[celest ibex]

Perplexity has been immensely helpful to me, thank you all so much! Doubtless hours of manual internet scouring have been saved.

One small thing I'd like to see: Could we have the reference links below the answers be actual html links so the full url will show on mouse hover?

[oblique topaz]

could you let us paste multiple lines? when I paste something that has multiple spaces it only paste the first line

[versed musk]

If you click on view list, you can see the sources, and if you move your mouse over the titles, your browser should show the URL.

[rocky arch]

Just asked for an example of writing Swift code to make a basic app UI and it works great on the “Swift playground” app on my iPad ! 😳

https://www.perplexity.ai/search/3910f64b-0bc3-43fc-b665-031141893fad?s=u

I have zero knowledge of writing code 😆

[rocky arch]

Was trying to recover this “conversation” with perplexity but I have not option to follow it and ask it

I’ve bookmarked it yesterday thinking I could follow the conversation today and change some parameters to continue the writing of the code

[radiant kindle]

Perplexity, or any AI ever (including ChatGPT, GitHub Copilot and the like), are NOT perfect. I've seen Perplexity give me code that is blatantly wrong more than once. If you are completely new to programming, I'd recommend checking out some online resources to get you started with understanding the basics, so that you can potentially figure out whether Perplexity is talking nonsense with their code.

"Saving a conversation for later" has been suggested multiple times here. No idea if it's already on their backlog.

[rocky arch]

Thanks for the reply. I’m watching some tutorials to learn swift code, at least some basic info to understand it

I found this

https://useadrenaline.com/playground?ref=futuretools.io

[flint lodge]

@heavy forge list gpt's that can summarize large documents like .pdf and .doc files

[covert sun]

@flint lodge you can ask this and many other questions in the #🔎│ask channel.

[celest aspen]

I think so far the iOS app is a great start!! Just a few suggestions, maybe a widget for a random daily question, Oled/amoled support for deep blacks, and some sort of settings menu to toggle various settings

[shy oriole]

good

[vapid mason]

You're not fixing bugs we report. This shows that you do not have a working procedure to record those bugs in anyplace that gets attention from your programmers. This is not a rude or snarky comment, so do not "shoot the messenger". Let us know when you've set something up to properly record bugs we report, and some method to take note when we record them - and I will re-report the same bug I already reported that's still not fixed (simple 10-min-to-fix UX CSS mistake causing serious problems)

[covert sun]

Hey @vapid mason! Thank you for sharing your ideas and helping us to improve, much appreciated! We work on the fixes for the bugs that users report and we do use a ticketing system. Would be great if you could report what you mentioned one more time, and sorry if we've missed something.

[radiant kindle]

Just saying

• some bugs aren't simple.
• they can be busy and/or low on manpower and/or having a task with a higher priority.
• them not being transparent with stuff doesn't necessarily mean they aren't listening, taking notes and taking actions.
• they have fixed a bug I found, at the very least.
  I do see that sometimes no change is seen after reporting an issue for a long time, but I'll most likely assume they are just being slow/busy/stuck, not they didn't care.

Looked through their past messages. I don't see a message from them about a "simple 10-minute-to-fix UX CSS mistake". I found a message about bot commands not working, and a chain of messages, starting here <#👉│feedback-general message> about an improvement idea to the AI.

[unreal kite]

Good to see that issue is mow resolved #🔎│ask message

Will likely use the app a lot once the android version releases.

[rocky arch]

Congratulations for the iPhone app! It works perfectly 👏👏

Can it be universal? I’d like to use it properly in my iPad too

And it would be just perfect if it could understand Spanish speaking

By voice

[storm hamlet]

Perplexity can't make tables, although it uses the same method that Bing AI does. Here is an example:

[mossy tulip]

working on it, will be fixed soon

[storm hamlet]

Thank you!

[light abyss]

Error in processing query

¿Perplexity is working?

[latent urchin]

will perplexity have api access any time soon?(maybe if integrated like in the extension it would be easier)

[silk aspen]

There was a bug on iOS app on first launch after giving microphone access and voice recognition. Simply did not pick up any audio or words. Button stayed grey. I had to restart the app to fix it. I’m running iOS 16.4

[analog mulch]

Hey team. Love the iOS app so far. The only thing that needs to be addressed ASAP in my opinion is that code examples are not shown inside a code-block, and rather as a normal text

[manic roost]

While the bold text looks very good with small prompts, when using larger prompts with for example some code, it isn't very readable. And the input field doesn't scale with text like it does with Chatgpt. This makes large prompts harder to read. It would be nice if the input field was larger and that the indentation of text stays the same.

[radiant kindle]

Try the discord bot until they improve this part...?

[manic roost]

Yeah, maybe that's better right now.

[mossy tulip]

this will be fixed soon

[manic roost]

Great!

[willow moth]

Please perplexity team can you work on perplexity to be installed on chrome / I don't know much about the technicals but like any other PWA which has a chrome based app

[willow moth]

@covert sun I think this is the most demanded feature before moving on to making the android version for perplexity

[mossy tulip]

could you elaborate? we do have a chrome extension already https://chrome.google.com/webstore/detail/perplexity-ask-ai/hlgbcneanomplepojfcnclggenpcoldo

[willow moth]

Yes you are right there is a chrome extension only for desktop, I meant if we can have a chrome installable feature to access it like a native app but on the web (basically a PWA)

As shown on the screenshot below for example on the mobile version of Twitter , there is an option to install it on chrome as an app /webapp

[radiant kindle]

Is that even a thing on PC?

[heady plover]

api access would be really nice to have

[spice monolith]

@mossy tulip any update on API, a month or so back on LinkedIn you said should be soonish. Thank you!

[celest aspen]

I second this, that would be very helpful

[proven seal]

Hello, Perplexity Team! is there any api for perplexity so i can send query from terminal or via python? I want to simplify life for me and use perplexity to generate figma files.
I want to do something like this:

1. pass figma file to perplexity (it is json) and query to adjust it.
2. get user recommendations/code to adjust figma file(or just it's part) via figmascript(figma plugin api)
3. enjoy experimentations with generating more complex and complex designs

[heady plover]

thw api is not public, you can only try to hack around, thats what ill do today probably

[proven seal]

there are api.perplexity.ai but nothing interesting in it

[heady plover]

ik its beinf worked on or smt
but i mean, i found somd stuff put yesterday by seeing what it does and hacking around and making your own api doesnt seem too hard but idk, ill see

[proven seal]

keep me update please i'm interested on your work

[heady plover]

yeah, ill only be able to release it or share it if its complient with tos though

[latent urchin]

i also made the way around but i dont know how to release(perplexity "chrome extension") because its original extension works by contacting and sending messages to the extension by locating its key on your chrome but if it is not that same key for you manifest json file on your extensions it will now respond because of the wrong key and you will be stuck at the message animation when you ask something. this means that because when you release a new separate extensions you need to have a original key that no extensions ever had before so basically you have two options: 1. make the extensions but with another key so that it can be your extensions but it will not send info and the user will be stuck when asking something, or 2. you can add the same key and it would work but once the user download your extension the original extensions will override it so none of your design or functionality will be gone and it will load the original.

[icy goblet]

In the website, I think you need to make starting a new search just one click away. Now it requires pressing +New Thread first. This is an extra step that should be eliminated.

[radiant kindle]

I wonder how this can/should be done, because there has to be something that separates "I'd like to start a new thread" from "I'd like to continue this thread"

[covert sun]

We have already applied for our bot to be verified by Discord, as it reached the number of servers (250) where it can be invited without verification.
You'll be able to invite the bot to your server as soon as Discord verifies it. We're waiting for it, too. Stay tuned for updates and thanks for your support! 📌
#💬│general message

[vast jetty]

im sure this has been requested before but I would love a copy button for both the web app and the iOS app!

[willow moth]

Try to make it chrome installable @covert sun @mossy tulip

[latent urchin]

recently stard+ted pubishing my extension so does anyone has any ideas on how to make the extension popup have rounded corners

[unkempt atlas]

I think some spacing between the user's question and the answers would be nice