💬│general

[tame current]

8k chars aren't nothing

[arctic spade]

pay for playground or poe then

for my use case the pplx 3.5 ft model is perfect

[tame current]

Do you know anything about the finetuning data of perplex?

[arctic spade]

here's a tweet from pplx with some data: https://twitter.com/perplexity_ai/status/1695102995325710484

Perplexity (@perplexity_ai)

Exciting update to Copilot on Perplexity with @OpenAI's GPT-3.5 fine-tuning API! We've improved speed, reduced cost, and matched GPT-4 performance. Copilot now engages faster without losing quality. Here's what you need to know: 🧵

Likes

767

▶ Play video

Twitter · Aug 25, 2023, 3:57 PM

[lilac nacelle]

@latent wharf is there a way to see previous threads on android that were done via web on my laptop?

[carmine linden]

Hey @lilac nacelle! You can use the bot in the #🔎│ask channel. Tag @ PerplexityAI and ask your questions.

[grizzled canyon]

all information up front, first up, taking it all in like a firehose

[balmy dew]

what is the best model offered by perplexity to write long-form text / ebooks?

my plan is to do market research, come out with an outline, then, ask AI to develop an in depth outline based on the research, then, ask to break down and using prompts for each subchapters and paragraphs by asking exactly the tone and how the chapter should develop

so i don't expect to have an ebook with one click

but which model do this best? i was thinking chatgpt4 but the total context of 32k means that it can remember around 15k words, i assume half of it would be inputs and so after 7-9k words of ebook, it will start to forget what was previously written and this can create duplicate content and redundancy, no?

what's the best way to do this

[velvet dirge]

good question

ping me when u get the answer please

[halcyon coral]

You could use Claude 2 and/or start a new thread after a few messages and give your previous text (as a file), this way you can avoid hallucinations/duplications for the most part.

Of course, this requires a Pro subscription.

[restive olive]

What are the main differences between paid ai chatbot services like ChatGPT, Claude 2, Poe, Perplexity.ai, etc? Also looking for pros/cons, tokens & character limits… I current pay for ChatGPT but wondering if there’s something better bang for my buck wise. I’ve used all the services mentioned above but I’m sure there’s things I have missed with how fast things are developing

[worn monolith]

I recently switched my $20 subscription from ChatGPT to Claude 2. I find the latter is significantly better at summaries and outlines the way that I do them. It can also write engaging essays. I didn't have that much luck with ChatGPT. Perplexity is incredibly useful for researching the internet. So I can use perplexity to research a topic, provide a good outline to Claude, and get a pretty good essay in response.

[undone olive]

So chatgpt vs Claude isn't a big difference since they both semi operates on the same metrics meanwhile Bing, Bard and Perplexity is on a whole another level.

[karmic river]

Excuse my ignorance but actually what is the difference from using chatgpt or claude to using these two services through perplexity?

[orchid magnet]

i thought if i paid for perplexity it was using gpt4?

[wraith crow]

You can set the model it will use in the user settings as a pro user

[orchid magnet]

I did, it says thst it was set to gpt 4. But when I talk to the ai it says it is gpt 3.

[wraith crow]

do you see gpt 4 here when it answers?

[orchid magnet]

Or it says it’s not gpt 4

Yes

I do and it’s selected

[wraith crow]

Then its gpt 4. It says 3.5 since it trained on data till sept 2021 where gpt4 wasnt a thing

[orchid magnet]

ohhh

I got it

Sorry for the confusion!

What is Claude 2?

[wraith crow]

Claude 2 is another model from another company. Its good at writing text. It also has a higher context limit of 100k tokens instead of 32k with gpt4

[orchid magnet]

I wanna get a prompt like this “Generate slightly longer two-part motivational quotes for TikTok videos targeting teens, young adults, and people seeking inspiration in the areas of self-confidence, self-improvement, and relatability. The quotes should be inspiring, relatable, and visually appealing when displayed as text on the screen. Divide the quotes into two parts, with each part being a separate line of text. Ensure that the first part serves as a captivating hook, while the second part evokes a sense of relatability and connection for the viewer. The quotes should be slightly longer than usual to increase the time viewers spend reading them, but not too long to lose their attention.” That’ll spew out quotes for me that I can use for TikTok content. Is gpt 4 or Claude 2 better for this?

[wraith crow]

The only way to find out is to try them both! And see which answer you like more.

[orchid magnet]

Or wait, what if i asked Claude 2 for a gpt 4 prompt?

Or vice versa

[thorn moss]

You need to experiment and see for yourself

[orchid magnet]

true

But I get access to both with this? Thst is awesome

Canceled my gpt + subscription for it

[thorn moss]

We cant answer everything cause AI has lot of potentials

i cancelled mine last week

i had enough. 😂😂😂

[orchid magnet]

😂

I love pwrplexity so much

[thorn moss]

Everyday i am reading complaints. 😂😂😂

[orchid magnet]

U canceled gpt+ for perplexity?

[thorn moss]

I have both befofe

befor

for me perplex is so useful

[orchid magnet]

Same

[thorn moss]

somehow chatgpt was so goos before

[orchid magnet]

The sources makes it so much better

[orchid magnet]

Government regulation

[thorn moss]

but unfortunately open ai has a big problem with money

nah

Money issue

they cannot maintain their platform

if that was true but i think its legit

so many subscribers unsubscribe

Thats why i learned to accept that openai needs time for now. Embrace perplexity.ai

[orchid magnet]

Oh

Yeah

I am

For the same price I get Claude and got 4

[thorn moss]

the only problem with perplexity is input limit and output

how i wish someday input limit gonna expand

and output limit like 50 percent of claude 2. lol dream 😂😂😂

with that perplexity is really monster

but i hope openai will have a comeback

[orchid magnet]

We shal see

[balmy dew]

input is not a problem with perplexity if you upload a txt file, it is a little bit unconvenient, but it does the job.

the output i didn't find any restriction on the output? i hardly get longer text with gpt4 ? i have both

deleted my subscription for perplexity

that's the thing i don't like about these chatbots, especially Claude, i don't think it's that good, i tried it thru poe.com and here and i always get bad results

i am using it to write product descriptions, doing market research, write some long form text and it always gave me very short output, maybe 500 words, and never elaboraated or in depth as GPT4

Perplexity i am still testing, but i think it is running on GPT 3.5, no?

[crude crystal]

Yes I think it's 3.5

[crude crystal]

I'm also using Claude for content writing but it doesn't seem as good as GPT 4 is

[stark lichen]

i like the internet research it's quite formidable

[chilly forge]

FineTuned 3.5

[orchid magnet]

They advertise thst when u pay u get gpt4 tho

[chilly forge]

There are separate models for each one of 'em

[orchid magnet]

So it is gpt 4

W

[grizzled canyon]

you have to change the model yes

[orchid magnet]

Trying to have Claude work

[grizzled canyon]

just select it in the settings

and it's on

[eternal kestrel]

yo

[calm parrot]

Would be nice to have the interpreter features in the perplexity app 👀🚀

[stark lichen]

is there a way for perp to read the current page i am on?

[halcyon coral]

Yes, you can use the chrome extension:
https://chrome.google.com/webstore/detail/perplexity-ai-companion/hlgbcneanomplepojfcnclggenpcoldo

[stark lichen]

Hey @halcyon coral problem is it tries to acces the page from preplexity server and it cant if the page is private

[halcyon coral]

Perplexity only has access to public data.
The most you could do is download the page and use the file upload.

[balmy dew]

How can I leverage perplexity to increase the sales of my books? Or to help with the creation of my books / market research / user persona research / marketing etc.

Looking for ideas I might not have thought about

[flat agate]

You could ask perplexity about a rundown of the book, and later a more of less detailed explanation of every subject or chapter, if they are literary works you can ask for research help on every point or more creative prompts to explore ideas, in more academic books it can help you to create the skeleton of a chapter or order your ideas... 🤔

[fluid mauve]

I just used perplexity to work on some bullet points for a website. I gave it my initial suggestions. It came back with its 10 suggestions based on my originals. I then got it to spit out 4 variations on each of its previous results. I now have 50 examples of what I was thinking about. all delivered in about 5 minutes from ideation to creation.

Maybe feed perplexity your ideas and see what it comes back with. I was using the Claude 2 model and focus was set to all.

[flat agate]

I have the intuition that Claude2 approach is better for this kind of creative process, yes 😊

[grizzled canyon]

in my experience it's much better at intuiting and providing more original output

[flat agate]

Yes, it is more human like on that and gpt4 more skilled on less creative stuff, it is great to have access to both, the only thing I use in open ai directly is the code assistant that is still better than perplexity at that work, but but perplexity will go evolving

[lethal bramble]

Hey guys! I created a discord bot with gpt, anyone wants to try to hack the system prompts? if yes dm me I'll be glad ! 😊

[white laurel]

#💭│ideas message williamx did small website just for this: https://perplexity-clean-mergd.vercel.app/. A bit late, but you still might be interesting

[mystic adder]

Hi, I have a question regarding perplexity pro

Does asking questions get the same answer as I would get from using chatgpt4?

[chilly forge]

Mostly Yes but really it depends on the question since both have different system prompts applied

[arctic shuttle]

Are there known issues with mixed up conversations when using the perplexity deployed code-llama? Because I just encountered a quite wired one..

I am talking about when you get a response that was ment to someone elses conversation

[orchid magnet]

Any content creators here

[chilly forge]

#🦙│feedback-playground

[arctic shuttle]

great thanks, I don´t see that channel I don´t see that channel in my discord though!? How do I get to that setting do you know that? I remember entering sometime a server where I was prompted which channels I want to see (like after they did some changes IDK). Apparently this is the same issue here, bc obiously that would be quite a silly question if I already see feedback-llama 😂

[arctic shuttle]

#2 nvmd, found it. Literally just 2 klicks away there is the setting..🙈

[chilly forge]

@frank rivet

[halcyon coral]

Thank you, he has already left the server

[rare fiber]

9/11. We remember.

[regal frigate]

I have subscribed to the pro version, and I would like to ask where to use gpt4-32k. I chose the focus on writing and uploaded an article with more than 5,000 Chinese characters, but it seems to recognize only a small part of the content. When I ask about the content at the end of the file, it knows nothing. How can I solve this problem?

@halcyon coral

[halcyon coral]

Hey @regal frigate!
It will automatically switch to GPT-4 32k if the context becomes too large.
Please try again to ask for the exact text elements that are at the bottom of the file.
You can also switch to Claude 2 and test if that works.
Can you please send me the file if that doesn’t work? (DM)

[prisma parcel]

@halcyon coral And @regal frigate, but is it possible to execute a proofreading process on medium sized documents like blog posts (a few thousands words), using Perplexity? Would be a great use case for the writing focus ....

[regal frigate]

@halcyon coralI can use the Claude2 official website, and it works well for the same question. However, it doesn't work with Perplexity.

[halcyon coral]

Thanks for the reply, then the problem seems to be with Perplexity.
This is the fourth time such problems have been reported to me.
I am sorry for the inconvenience.
Hopefully, the problem will be solved in the near future.

[halcyon coral]

From the context size it would work, but only with the Claude model, for the other models it is probably too big.

[prisma parcel]

Yes, uploading the file as a markdown or text file does manage it as context. But it doesn't seem capable of returning the result: it breaks the work in chunks and then jams, also using Claude...

[regal frigate]

However, the document only has a little over 5,000 Chinese characters, which does not exceed the GPT-32k length limit. When I asked it about the content of the last paragraph of the document, it only answered with content around 1,000 Chinese characters in.

[halcyon coral]

Hmmm…. Could you please create a bug report for this? #1140622008086970420

[regal frigate]

I uploaded a PDF file, and I just asked the same question to Poe's GPT-4 32k, and it worked.

[prisma parcel]

done

[halcyon coral]

Thank you 🙏

[gaunt stratus]

hello

can change language in web? im spanish and my english is poor

[flat agate]

El interfaz está solo en inglés pero puedes preguntar en español y pedir que te conteste en español

[fluid mauve]

I dont care what mt wide says a Crisp sandwich is such and under-rated culinary delight

[tame current]

#💭│ideas message @halcyon coral tokens or chars? how much is it now?

[halcyon coral]

Tokens, 1.5k

[jagged geyser]

Best bot ever fr

[balmy dew]

Am I the only one who thinks CLAUDE2 is terrible?

the output is always subpar, it doesn't understand well the prompts

the content is meh.... the output is always shorter than GPT and always more generic, no matter the use case

[muted linden]

I fully agree. For me who mostly deals with technical / software topics, Claude 2 has continuously produced misleading or wrong output. 🤷‍♂️
To be fair, I didn't use it for coding yet, more for software architecture, APIs, concepts etc.

[jagged ivy]

I like claude 2, but don't use it because Claude 2 often doesn't make citations in the text

[tame current]

L

[wraith crow]

do you still see undefined after reloading the web page?

[stark lichen]

There is like 5 tools i am testing right now for ai assisted browsing. Preplexity is hands down the best at just raw research. Will we have ability to have it crawl even further? Instead of 1st page of google maybe first 10 pages?

[gilded breach]

Hello all! is there any word on Perplexity offering an API?

[halcyon coral]

No plans about that

[gilded breach]

ok thank you very much!

[tame current]

so many people ask for an API here, I'm curious what the use case would be

[noble quest]

Probably to integrate it into their own project

[chilly forge]

1. Pay as you Go prices
2. Automation for the workflows like lets say replying to an email

There are tons of usecases with API

[tame current]

that's what I'm wondering, what kind of projects

[tame current]

yeah that makes sense, I didn't think of automation

[tame current]

hello

where can i sign up for the android app waitlist?

[halcyon coral]

The android app is available in the Playstore.

[tame current]

lms

thank u, appreciate it

[edgy fiber]

To me the main differentiator for a Perplexity API, relative to, say, the APIs for ChatGPT, Claude, etc. is the Internet search component. It's basically a Google Search API but rather than sifting through a JSON response of a bunch of links I can get an actual answer in whatever format I'd want. I'm surprised Perplexity hasn't offered it yet because as far as I can tell there's no competition offering a similar thing yet and the first-mover advantage should be huge.

[sick hill]

Could you name them?

[tame current]

name them 🙏

[tame current]

gotta find better alternatives

get phind

uses gpt4 by default +free

💪

[hard ermine]

YOu use plexity to debug your code as opposed to GPT code intrepreter or claud2? Or something like https://github.com/KillianLucas/open-interpreter

[tame current]

the previous guy got banned for sharing a website i dont know if theyre rlly open to suggestions

[grizzled canyon]

they're trolling

[tame current]

who?

[stark lichen]

chapagpt, merlin ai, harpa ai, best regards?
best regards, merlin & harpa allow me to use my own key
each one does something special. Perplexity so far has been an amazing research aid. Faster then doing manual research.

most useful so far is perplexity i just wish i could use my own api key and pay less / month

[carmine linden]

Hey @formal parcel! You can use the bot in the #🔎│ask channel. Tag @ PerplexityAI and ask your questions.

[dense kraken]

#📢│announcements but the channel link is "no access"

[signal hamlet]

Hi @dense kraken! I see that #🅰│web-alpha-feedback channel is open to everyone. When you proceed to the link, it doesn't open the channel, right?

[formal parcel]

Would Perplexity consider implanting open-interpreter?

[chilly forge]

free
Only for 10 uses / 24 hrs (GPT4)

[icy abyss]

Is it possible to "bookmark" or set aside a particular thread?

[dense kraken]

I can now see the channel link. (idk if it's a problem with discord and/or my internet; when I saw "no access" it was on my phone, and now on my PC I can see the channel link properly.)

[halcyon coral]

This was fixed 😉

[jagged ivy]

I don't have access too

[halcyon coral]

Should be fixed now

[halcyon coral]

Can you see the channel in Channels & Roles?

[wraith crow]

its private again

[halcyon coral]

Should be public checked it with another account

Maybe you have to refresh discord

[wraith crow]

I did..

I had access to it like 20 mins ago

[halcyon coral]

Probably a discord bug 🤔

[wraith crow]

It is a discord bug.. I have access on pc

[tame current]

I can access it

[jagged ivy]

Will check this when I come home

[muted linden]

Now the access is gone for me too. Ha access like 20 mins ago

[wraith crow]

are you also on your phone?

[muted linden]

Yes, Android app

[wraith crow]

Seems to be a bug with the discord mobile apps. It's working fine on web/ pc

[muted linden]

Can confirm I have access on the web

[fluid mauve]

@halcyon coral probably worth getting this android entry fixed as the app is available now.This was in Channels & Roles

[muted linden]

Whoever has problems accessing the web-alpha-feedback channel through the mobile apps: it helps to logout, delete all app data and login again.

[halcyon coral]

Another solution would be to switch "quickly" to the alpha 😅

[unkempt prawn]

Does alpha.perplexity.ai use the same LLM as www.perplexity.ai?

[halcyon coral]

Yes it's just for testing the new thread management update.

[unkempt prawn]

Does that mean the functionality will eventually be available at www.perplexity.ai when it's done with field testing?

[halcyon coral]

Yeah, sure!

[unkempt prawn]

Thx!

[wheat vine]

It says I’m using a vpn when I’m not

At least I don’t think I am

They’re all disabled so I’m confused

I cant Access perplexity

[halcyon coral]

Are you using any extensions? Did you verify yourself at labs.perplexity.ai?

[ebon zodiac]

I find a new bug that I can't right swipe my thread to delete it now.

in my iPad

[halcyon coral]

Hey, a message in #1140622008086970420 is enough. Thanks!

[ebon zodiac]

OK

[signal hamlet]

You are right, @fluid mauve, the App is out there already. I updated that question. Thanks for bringing that up!

[silver laurel]

If upload pdf with text in image will it still work?

[fluid mauve]

No worries. Happy to help.

[granite ocean]

hi

[ivory lava]

I’m sorry if you don’t understand what I’m trying to say, I’m using a translator (Deepl). Have a nice day, everyone! I just want to ask if Claude 2 is using the 100k? I asked him to give me the formulation of an argument in a section (6.3, for example) because he has not. I go to Claude.ai and Poe and it does it correctly. But I ask this from Claude 2 in Perplexity and it gives me a wrong answer. I thought it was my prompt: turns out it is not. I came here and uploaded the pdf file in 4 pages on Perplexity (out of 32) and asked him again to give me the formulated argument: he did it correctly. This makes me think that Perplexity does not use 100k.

[halcyon coral]

Hey, Perplexity's Claude 2 uses 100k context.
However, several file upload issues have been reported with Claude, but the devs are working on improvements!

[fading rose]

what's Perplexity's position on password sharing?

like, is there some netflix-like measures to block it? or an intention to do so?

[halcyon coral]

You mean share account?

[fading rose]

yes

[halcyon coral]

This is not allowed!

[fading rose]

I use my account on multiple devices (and multiple networks, as I study far from home)

will it get flagged or somethign?

I know netflix flags it

which is why I only use it like twice a month

[halcyon coral]

Not as long as only you use the account, sharing with other people is prohibited.

[fading rose]

so I'd assume similar positions would have similar policy? how does it work?

ok but is the check done by verifying if the IP changes too often, or what? because moving a lot across the country I'd not want my account to get suspended

like, do you know what behavious it "looks" for? because I'm pretty sure my actual identity doesn't get leaked to the browser

so there has to be somethign else

if that thing is position, it's quite possible that it'd think I'm multiple people, because I study around 270km from home

[halcyon coral]

I am not aware of any bans due to location changes.
But presumably Cloudflare will make additional checks.

[fading rose]

and my vpn sometimes makes me look like I'm quite far, too. Free tier Proton only has servers in nl, jpn and US

[fading rose]

cloudflare does check a lot when I use the vpn

[halcyon coral]

Yes that is normal

[fading rose]

I assumed as much

just wanted to make sure, before getting a premium, that I'd not have issues with switching locations often (which between travel and vpn happens a lot)

[halcyon coral]

I don't think you should have any problems.
And if there are any, you can always contact support 😉

[fading rose]

will do!

I've been around for long on the free tier, but claude is really pushing me to try the pro

claude itself is already really powerful, with internet results it could be insane

[balmy dew]

is it possible to change my password on perplexity if i signed up with google? (in order to have a different password for perplexity than Google)

[halcyon coral]

No, that is not possible.

[radiant tundra]

Is there an easy everyday way to switch between gpt-4 and claude 2 or is that only available in the settings?

[wraith crow]

only in the settings as of now, but it has been suggested by several people to make this easier. #1145894530239975474 message

[wraith crow]

Since you're asking this question I assume you're a pro user. You can get access to the pro channels by leaving this discord and joining again with the invite link in the pro section on the settings page.

[radiant tundra]

thx!

[radiant tundra]

an easy switch would be useful, given the models are good at different things

[tame current]

Or not

I definitely wouldn’t choose Perplexity for that

[radiant tundra]

perplexity is like chatgpt plus (literally the same llm in the background) but with transparency regarding the sources the llm draws from for its answers. would def prefer perplexity to chatgpt for this use case, because this way you can easily check wether the llm got anything wrong, why etc.

[silver laurel]

I just curious how it picks those few sources in the answer. I heard it based on google search to find the sources

[tidal stream]

what model does everybody use? perplexity, gpt4 or claude2?

[tame current]

I think it just reads all the sources and decides which ones to use on its own

[tame current]

I use perplexity and gpt4 personally

[tidal stream]

how do you use both? I think we can only chooes one

[tame current]

ah yeah, I meant to say "or"

[tidal stream]

fine. so in what reason you switch the model? when you find the answer is not good and then try another?

[silver laurel]

Claude as the answer is more human and creative centric

[tribal gale]

hello!

[tribal gale]

I'm using claude, but the output in not in Mark Down, it won't output in mark down, any way I can work around this without having to manually format the text?

[granite ocean]

hi i am a pro user and i want to ask is there a way to get perplexity gpt-4 responses without searching

like bing #nosearch

[halcyon coral]

Yes you can use the writing focus

[granite ocean]

oh thanks let me try

[granite ocean]

will i be able to code?

[halcyon coral]

Sure

[granite ocean]

thank you very much!

[tidal radish]

i have a pro subscription. I reloaded my perplexity tab this morning and it shows "0 left today" though I have not yet used it at all today. anyone know what's going on?

[halcyon coral]

Currently there are overloads, please do not open bug reports and please be patient.

Perplexity is stable(er) atm at https://alpha.perplexity.ai.
Thank you!
https://discord.com/channels/1047197230748151888/1151514486008717352

[tidal radish]

ah okay cool. thank you. will use the stabler version for now.

[fervent oracle]

Thanks just saw that

[candid falcon]

same here

[fiery dust]

Omg I LOVE this new feature! 😍

[stable burrow]

ah okay - time to break for coffee

[balmy dew]

what's the difference between the discord for the pro and the normal? I left the server, joined again but i only see the old channels

[foggy marlin]

hey all, i hope you are doing well. I wanted to check in again on some work i'm doing with the US Embassy in Moscow. I'm presenting a series of topics using AI, and usually demonstrate Perplexity (sometimes Pi) because they're available in Russia...and i think Perplexity is pretty awesome.

one of the assumptions i have to make about the audience...all students within Russia, who theoretically do not use VPN...is that they have the same functionality in Perplexity that i do. To that end, i had a couple of questions i was hoping a mod or someone might address:

1. is all current Perplexity functionality (including copilot and libraries) available in Russia? I'm assuming Russians can at least try Copilot, even if GPT and Claude are not directly available in Russia

2. can russians even pay for a paid subscription? I didn't know if payment was possible with some of the international restrictions now. I only present the "free" version, but would love to tell the audience they have access to a lot more if they do subscribe.

any thoughts are appreciated, and thanks in advance. Perplexity is a real gem in the AI world and i really like telling audiences about it.

[halcyon coral]

Hey @foggy marlin!
It should all work in Russia!

[halcyon coral]

Hey, please leave and join again with the link in your settings.
If nothing happens, please contact Alex Romanov via DM.

[pulsar sparrow]

Gn

[keen citrus]

Is the Android Beta still running?

[wraith crow]

Yes, you can sign up for the beta on this google form

[tame current]

great website!

gets progressively worse

[wraith crow]

you can report weird answers on the website by clicking on that flag at the bottom of the answer. it will get send directly to the ai team so they can improve the output.

[tame current]

i dont feel like it though

[foggy marlin]

Ice, ty so much!

[lofty dagger]

hiya

i need to vent

[trail herald]

I am interested in how perplexity conducts its searches based on the free text queries that we give it.

When Perplexity is in search mode the search text used can look simplistic compared to the question you have asked it.
Where are these searches being conducted? Google?

What determines the syntax used for conducting these searches? How can we be sure the search strings will produce optimal results?

Also, how is it possible that when Perplexity says it is “considering” 30 sources, that it takes only a few seconds to produce output from those sources. Simply loading 30 internet pages takes way longer due to latency and other factors, so it makes me wonder: how deeply does Perplexity actually “read” the pages?!

[tame current]

theyre using a serp api

using google search previews as sources

it does not extract text from webpages

otherwise the latency would be higher

the product itself is gimmicky in my opinion

an extension can easily replace it

[trail herald]

I’ve been using Bing chat lately and trying to compare the two. It’s all a bit smoke and mirrors for sure.
That said I do find perplexity useful!

[hallow linden]

I'd love to see an extension that could do so. I've found perplexity to be the most useful implementation of LLM I have come across

[silver laurel]

I think it all based on your prompts

[formal parcel]

tbh, I think the OG logo was wayy better

[twilit wigeon]

The alpha.perplexity is having a different backend, right

It is returning a wrong answer

While the original site works

[halcyon coral]

Can you please share an example in #🅰│web-alpha-feedback?

Alpha has now moved to https://perplexity.ai 🔥

[twilit wigeon]

Done

[twilit wigeon]

No wonder now in main site the answer is becoming wrong

What's is the token size limit of normal vs pro of pplx?

[halcyon coral]

Input limit or context limit?

[twilit wigeon]

input limit

Was referring to input limit, but also good to know context limit (which I have never reached before)

[halcyon coral]

It's 2k, context GPT-4 32k, Claude 2 100k.

If the input is too large (for example, if you copy a website), it is converted into a txt file.

[twilit wigeon]

Thanks, I will probably be trying pro because 32k is generally good enough for my use case.
But one more question before subscription, android app does not seems to support gpt-4 even if I am a pro, is it right?

[twilit wigeon]

I mostly use it for coding, it outperforms gpt-4 official and some other competitors

[halcyon coral]

Not atm, but you can change the model in the web app and then change back to the android app.

[twilit wigeon]

So android app will remember my setting, if I am correct

Thanks for the support, gonna try it after work.

It's working hour here.

[granite ocean]

how can i know if the model is using gpt-4

[wraith crow]

It will always say it’s gpt 3 no matter what since the cutoff date is sept 2021. You can check it it’s gpt4 by asking question that gpt3 can’t answer

[granite ocean]

please give me a question like that

i am trying to find one

[wraith crow]

make sure you’re in writing mode for this so the internet searched can’t change the answer in a weird way since it’s not a normal question

[wraith crow]

Solve the puzzle
1-d
2-i
3-t
4-c
5-a
6-s
7-e
8-n
1,5,3,5_,6,4,2,7,8,4,7 = ?

[granite ocean]

yup

ok tysm

[wraith crow]

The answer should be data science

[granite ocean]

hmm right

[wraith crow]

gpt3 will just make up a random word since it can’t solve it

[granite ocean]

btw chatgpt solved it too

cause its 3.5 ig

btw thanks for the thing i will save it for future use

[wraith crow]

for me it returns datascence on chatgpt

closer than before ig

sometimes I got the weirdest answers

[granite ocean]

oh

[wraith crow]

and you have selected 3.5?

[granite ocean]

yes

[wraith crow]

lemme check if I can find another one

[granite ocean]

oke

try to find a hard one

cause even i can solve it with my mini brain XD

[wraith crow]

I have 32 apples now. I ate 4 yesterday. How many apples do I have now?

answer should be 32

[granite ocean]

ok

perfect

chatgpt cant solve it

thanks

[devout sun]

perplexity.ai/pro?referral_code=6HW74JVG

[jagged ivy]

GPT 4 😭

[wraith crow]

that's weird.. I can't reproduce it no matter how many times I try could you report that answer so the ai team can improve the output?

[jagged ivy]

Perplexity AI: I have 32 apples now. I ate 4 yesterday. How many apples do I have now? https://www.perplexity.ai/search/I-have-32-AxqX0xx7RC6bGQ2p1_o3yA?s=mn

[wraith crow]

can't access it did you make it secret?

[jagged ivy]

Hm, it can be because I added thread in folder

[wraith crow]

could you try it without copilot

see if that might be the problem

[jagged ivy]

And all models make the same mistake

[wraith crow]

Okay one last try could you disable your ai profile. And add 1 ? to the end of the question

[jagged ivy]

Didn't understand about ?

[wraith crow]

So the question will be I have 32 apples now. I ate 4 yesterday. How many apples do I have now??

[jagged ivy]

Wow, GPT 4 without ai profile is so fast..

And yes, I have the correct answer now

[wraith crow]

I might be the ai profile..

Could you create a bug report and send both the answer in that thread. With ai profile enabled and disabled hopefully they can check what went wrong

[jagged ivy]

But using web search focus it suggests to subtract 32–4

[wraith crow]

I can confirm that yes. Seems like the internet sources can affect the answer when asking questions like that. Could you maybe add that thread as well to the bug-report. So the team can further improve the answers given.

[muted linden]

Seems to also depend on the AI profile. I get the correct answer even with GPT-3.5 FT and quick search. Perplexity AI: I have 32 apples now. I ate 4 yesterday. How many apples do I have now? https://www.perplexity.ai/search/I-have-32-G1pozePhQFKg8D4qHCwpiQ?s=mn

[jagged ivy]

I am making the folder with all bugs about it

And it's interesting that with ai profile search focus mode doesn't do this mistake

https://www.perplexity.ai/collections/Apples-0ChHe7DvShGPiHrPJT9sng?s=c

[wraith crow]

It still says to substract it and with the 2nd question it only returns what you had initially
nvm I see you haven't added the ai profile -search -gpt-4 to the collection

[jagged ivy]

Now it makes a mistake 💀

I have added thread with AI profile search mode

[steel escarp]

hi , i just jumped in here new , maybe the question been asked before but i will give it a shot : will Perplexity have a plan to add LAMA model soon to pro sub ?

[muted linden]

It's available in perplexity labs already, including code llama: https://labs.perplexity.ai/

[steel escarp]

aha ok, thanx . but i ment it will be in the app?

[proven rain]

Thanks for explaining this. I just came here with the same question. Will cancel ChatGPT Plus now since I prefer the Perplexity UI in addition to some other features and they both use the same LLM. What do you think?

[tame current]

perplexity doesn't have code interpreter

that's one downside

[icy raft]

I have a problem. Do I need to start a new thread and re-upload a file if I want to restart a chat with the file? Re-uploading seems like it could slow down the workflow.

[halcyon coral]

The devs are working on improvements!

[jagged ivy]

Claude 2 writes KaTeX not correctly while gpt 4 writes it correctly
https://www.perplexity.ai/search/MeL4fl6HRXuEE41cIs1RdA?s=mn

[wraith crow]

threads are now private by default ^ link is not working

[mental moth]

Is gpt 4 and Claude are pro version?

[dusty bobcat]

I am trying it for the month but not sure I can move away from code interpreter as I do alot of coding. I've tried the labs they offer but it's still not as good

[floral pine]

hey is the website down?

when i search something up it gets stuck at summarizing

[wraith crow]

Do you still have this issue after refreshing the page?

[floral pine]

let me try

nope it fixed, thanks!

ill stay in the server so if something happens i can get help

[zealous bough]

Is the new collections feature desktop only?

[flat agate]

I suppose they will implement it on mobile apps as soon as possible

[craggy pollen]

👋

[zealous bough]

Bruh

I've been using 1 thread for a general topic

Like all finance questions would be in 1 threaf

[tame current]

Wait what

Don’t the desktop and mobile websites function the same essentially?

[halcyon coral]

Yes, the downloadable app was meant.

[zealous bough]

I use the app

For mobile

[twilit wigeon]

seems the apple discussion suggest the profile makes the AI dumb.

i'd rather clear the profile

not seeing it helps actually.

any promotion code? 🤨

[tame current]

Ah. Seems like a fairly simple concept so I assume the app has or will eventually have that feature

I personally have the website as a bookmark since
[1] I’m assuming it’s being worked on more thoroughly
[2] I’d need internet in order to use it either way so why not

[halcyon coral]

Yes, at the moment the web version and the app version are on a different level in terms of functionality, but over time the functions should also be integrated into the app.

[tame current]

Will there ever be a desktop software for Perplexity or will it remain a website for desktop?

[halcyon coral]

There are currently no plans for this.

[tame current]

Fair enough

[dusty bobcat]

https://www.makeuseof.com/websites-desktop-apps-windows/#:~:text=Launch Microsoft Edge and open,a name for the app.

this is about as close as you'll get to it

[tame current]

The website is sufficient

[dusty bobcat]

I have clients who do this with webapps that I provide.

@tame current ☝️

[tame current]

Oh, I know how to do similar things. I was just wondering if they were planning on making an official app for Desktops

[dusty bobcat]

I wouldnt know what additional functionality you would get and justify with a desktop ap

app

[humble swan]

I hope discord bot gets some attention

[wicked blade]

Hey guys. Perplexity currently uses which search engine to search for information on the Internet?

[humble swan]

its own i assume

[wicked blade]

I do not think so.

[tame current]

Mainly just convenience. Having a software isn’t too expensive so you don’t need a ton of justification

Many AI prompt alternatives have done it so I was just curious

[muted linden]

People on this discord assume it's Google, for various reasons. The most official statement that I found is from this interview: https://youtu.be/jksGQhMtXjo?si=l0tr6kg828qYOaH3
Where Aravind says something like "we've started with Bing and now we're using multiple sources". Honestly: whatever it is, it is definitely better than Bing. I've been using Bing AI since it came out and I can tell the difference 😅

[dusty bobcat]

Software is more expensive than you would think, depending on the current tech stack they would need to potentially code a 4th tech stack which would be more time and money (ios, android, web). Using flutter or ionic could reduce this from the front end side of things but that's based on their requirements.

[tame current]

I mean, they made one for IOS mainly for that reason I assume

[tame current]

I’m not saying they need one

I was simply curious

[wicked blade]

Do you think Perplexity outperforms Bing Chat AI in aspects such as information accuracy and reliability?

[muted linden]

For me personally it definitely does. In Bing Chat I often saw citations which actually did not contain the claimed information which I personally didn't experience at all on perplexity. Aside from that, I love the feature set, the different LLMs, and the speed of perplexity which led me to cancelling my ChatGPT Plus subscription in favor of Perplexity Pro. No regrets so far, but again: only my personal opinion, your experience may vary depending on what you actually want to use it for.

[wicked blade]

And for programming? Do you also think Perplexity has an advantage? Or does Phind still sound like the better option?

[carmine elbow]

Is there any way to make the perplexity site specific pages where the information was found from the uploaded PDF file?

[muted linden]

I haven't done a lot of programming with perplexity in the last couple of weeks tbh. I'm more into software architecture and cyber security topics. But you can use the "writing" focus which is basically "pass-through" GPT-4, so it should be at least comparable to ChatGPT. Only thing to note is that perplexity has a response token limit of 1,5k, IIRC

[humble swan]

test it for yourself and your specific use cases

[proven rain]

"For instance, the website Retraction Watch has reported on a preprint about millipedes that was written using ChatGPT; it was spotted by a researcher cited by the work who noticed that its references were fake.

Rune Stensvold, a microbiologist at the State Serum Institute in Copenhagen, encountered the fake-references problem when a student asked him for a copy of a paper that Stensvold had apparently co-authored with one of his colleagues in 2006. The paper didn’t exist. The student had asked an AI chatbot to suggest papers on Blastocystis — a genus of intestinal parasite — and the chatbot had cobbled together a reference with Stensvold’s name on it. “It looked so real,” he says. “It taught me that when I get papers to review, I should probably start by looking at the references section.”"😂

[slow copper]

Can anyone help on how to get perplexity pro to help with my writing. I have 8000 words and even after my prompt stating I want similar word count after editing for flow and headings - it keeps summerising to around 500!

[nova shoal]

does the free perplexity copilot use gpt4?

[true lark]

Hi team-

Can someone help me please?

I've been seeing the following screen in my browser when trying to visit perplexity for a few days now. I see it on some other cloudflare-protected sites too. It's an endless loop and there's no way for me to bypass. I assume this is some kind of bot protection, but I'm just a regular human user and it's preventing me from accessing the site.

Things I've tried unsuccessfully:

• Cleaning browser/cookies/cache
• Renewing DHCP lease
• Changing DNS servers
• Alternative browsers (same result no matter the browser)
• Disabling ad blockers

Does anyone have any suggestions? 😦

[rocky sigil]

I think maybe I'll need to split it up? What model are you using? Would prompt are you using? I would go for GPT4 if you can but that would require you to split it up into chunks and feed it to the bot. But I'm no expert!

[icy raft]

I've experienced the same problem with Chrome, but interestingly, it works fine on Microsoft Edge. I've been switching between both browsers for my tasks, but it's a bit of a hassle and slows down my workflow. If anyone has a more efficient solution, I'd love to hear it.

[signal hamlet]

Hey, @true lark, @icy raft! Please try the following steps:

• go to https://labs.perplexity.ai/
• solve the captcha there
• navigate back to https://perplexity.ai/
  Let us know if it works fine for you then.

[icy raft]

Wow, it actually works! You've solved my problem. Thank you so much!

[true lark]

Yeah for me it's wacky on chrome and edge, but works fine in safari? Also annoying and def messes up my workflow. It's also affecting archive.today for me (unrelated, but they also use cloudflare protection I think).

Are you behind a pihole on your local network by chance? I am, and I'm wondering if that has something to do with it

[true lark]

Holy crap, that worked! Any idea why that is, or what's up with the main site?

Thank you so much!

[wraith crow]

it’s because of the perplexity chrome extension. That is causing this issue for some people.

[worldly perch]

I'll just leave my 2cents here regarding pro-pricing. I tried pro and I liked it, but 20$ is a bit too steep for someone who is not a "power user". I use it 1-3 times a day, usually as a full google replacement.

I would gladly sign up for ~$10 and stay signed up and never had to think about it again, but for $20 it kinda stings I have to admit.

Maybe a Pro/Lite Plan (with half or even just a quarter of copilot/gpt4 quota) could serve the "occasional pro feature user" better

just saying ❤️

[wicked blade]

The only problem is that, even if you pay for the Pro version, the use of GPT-4 is restricted to just 300 responses, correct?

While in Bing Chat IA usage is unlimited.

[wraith crow]

With perplexity Pro you have 600 copilot/gpt-4/claude uses a day which is almost the same as unlimited. Never heard anybody about reaching the limit of 600 even power users.

[wicked blade]

Oh, so there are 600 users at a time? I thought there were 600 responses, just like ChatSonic. 😅

[twilit wigeon]

you sure bing chat ai is now unlimited?

were having a limit which I hitted when coding for long

around 30 x 5 conversation, bing ai were complaining too much

[wicked blade]

Yes, I mean using GPT-4, not one-on-one conversation. The individual chat remains for 30 shifts. However, I'm always deleting and creating new shifts.

And always using GPT-4.

[wraith crow]

You have 600 total uses a day. Every time you send a message it will reduce the amount by 1 (if you're using copilot/gpt-4/claude) and after 24 hours you'll get that 1 use back.

[twilit wigeon]

Hi Ice LavaMan, my query contains only 15133 tokens. what's wrong?

start a new thread don't work either.
it forces my input to become a txt, which in the response it failed to parse.,

i got a brain fog, it is able to parse. Start a new thread can workaround, but still very inconvenient because it will assume it is a txt despite i tell him it is csv.
need to brainwash it a few time such that it believes it is csv.
ideally the follow-up should also support 32k ?

[twilit wigeon]

it affects how it generates code to read the source data.

[wraith crow]

It shouldn't affect the way it will read the data here is an example of a csv pasted as .txt https://www.perplexity.ai/search/Create-a-table-.JVTEeUuSRCUGInxoF1vJw?s=c
also the total context of gpt-4 is 32k. at the moment you can only upload a file at the start of a thread. After that you can ask questions up to 2k tokens (around 8k characters) for each question.

[twilit wigeon]

i was asking it to preprocess the data
it gave me

Read the file

with open("paste.txt", "r") as file:
data = file.read()

Split the data into lines

lines = data.split("\n")

Filter out irrelevant lines

relevant_lines = [line for line in lines if re.match(r'^[A-Z]{2,}', line)]

anyway, thanks for clarify the token limit.
may i seek for an option to disable auto-convert to txt?

i accept whole context 32k. follow-up 2k
but just give an option to do NOT convert to txt

otherwise i have to type 32k by hands.

[wraith crow]

more than 2k tokens isn't supported in the input field so it will automatically convert it to a txt file if it's above that limit

[wraith crow]

Could you share the Link to the thread or if it contains private data you could DM it or try to recreate a new thread without private data. That way I can help you better

[twilit wigeon]

if it helps the team to troubleshoot, no problem. i will do it probably tmr.

[calm idol]

Good morning ☀️ I’d love to share my interest in the open role as a Frontend Engineer! I’m amazed to see such a powerful tool and would love to help the team and company grow its already established success. I’m more than happy to share into details my background and portfolio. 🚀

[halcyon coral]

Hey @calm idol, you are welcome to do this at https://jobs.lever.co/PerplexityAI/, where you will also find a detailed job description.
Good luck!

[tame current]

when using perplexity with copilot, I see that it accesses live online sites and compiles them into an answer. but also when I dont use copilot, the answers have sources that also link to live sites. so what is the difference between copilot and no copilot?

[wraith crow]

Copilot is an interactive search assistant that personalizes results by asking clarifying questions. In the #📘│faq channel you can find some more information about it.

[light spear]

I’m not sure where to ask this. I was wondering since I saw the announcement about the new Collections feature for Perplexity, is that related to Google Collections? I got 3 emails saying random people wanted to share collections with me

[signal hamlet]

Hey, @light spear! It's not related. But it's cool to have lots of invitations from other users to explore different topics together!

[valid venture]

I am trying to get Perplexity to also add images when building an article. Does anyone have suggestions on how to instruct Perplexity to scrape an image, related to the title of the article, and embed it into the article.

[halcyon coral]

Was possible, but somehow doesn't work atm.
Maybe Alex Romanov can give you more information 👀

[signal hamlet]

Hey @valid venture, my example is not too detailed and articles would defs take more work, but just as an example for every query you can ask the model to add images and get some related ones at the end of the answer for now: https://www.perplexity.ai/search/c46b7aa7-e44e-4e7d-a29d-8c199c9f9492?s=u
You could also get more pics in the follow-up questions.

[lofty dagger]

guys

i know who the new mario voice actor is

wanna hear. a sample in the call

hint hes not a real human

[frosty vortex]

Question about the prompt in the starred channel. How do you atually put it into action? Do you enter this every time you ask a question?#⭐│starred message

[halcyon coral]

You can put the prompt in the Intro section in your AI Profile https://www.perplexity.ai/profile

[steel escarp]

Isnt perplexity connected to the internet 🤔?

[wraith crow]

yes but not if you're choosing the writing mode

[steel escarp]

What u mean ?

[wraith crow]

writing mode doesn't search the internet

[eternal saddle]

Hey, who knows of a great chrome extension that summarizes YouTube vids? One I was using broke. Actually, any site works as long as the summary is HQ

[steel escarp]

I'm sure i donr use writing 😔

[wraith crow]

And what if you put it another way like What do you think about www.cncgoldmine.com

[tame current]

lol

[tame current]

Can writing more be used offline then?

Meaning used on a phone that doesn’t have internet access

[zenith atlas]

No, as model itself is on the servers and not on the device

[halcyon coral]

No, an API from another provider is used and this can only be done with internet access.

[tame current]

Ah, alright

Wait, wouldn’t writing mode be way less accurate then?

[wraith crow]

Not necessarily, the offline model has information up to september 2021. So if you ask for information after that, yes it will probably be hallucinating.

But writing mode can give you a more chat like experience. And can be good for writing text and coding questions.

[tame current]

Wouldn’t turning it off be better for coding related stuff?

Well, depending on what exactly you ask

[tame current]

Chat like experience?

[wraith crow]

Yes like you’re talking to someone. The online modes tend to search everything you say on the internet.

here is an example

[tame current]

Oh so the ability to imitate human conversation

[wraith crow]

Yes that is better with writing mode

[tame current]

That’s pretty neat

[charred dagger]

its cant gather urls that well

[silk flower]

Why doesn't it support the Arabic language?

Why don't we use /search ???

[halcyon coral]

Arabic is supported!

[halcyon coral]

What do you mean by that exactly?

[silk flower]

like this

[silk flower]

I mean that on Discord

[silk flower]

nothing

[halcyon coral]

You can just add "Answer in ..."

[rain silo]

Did search for myself using perplexity and my picture showed up 💀

[paper viper]

NEW LOGO

WTHAT

HOTTTT

the blue is sick

http://zip.yi5.dev/u/faP7b3.png

https://cdn.discordapp.com/icons/1047197230748151888/a_9d48d7c3f30d0df7b8a61d6b0ab44d02.webp

[signal hamlet]

#📝│server-ideas message

[paper viper]

love it

great idea

[paper viper]

The end of the typeform link which says "discord" links to the root site: https://www.typeform.com/explore/

This questionare: https://perplexity.typeform.com/to/pv9jyhte

[vapid onyx]

Hey everyone

New to Perplexity. Any suggestions / use cases y’all currently use it for?

[proud robin]

Hi

[heady jay]

Amazing research tool. I’ve used it to replace just regular website google/bing searches. The results are phenomenal in my opinion 🙂

[vapid onyx]

I like the fact that it has sources associated. Is there any other differences compared to other AI tools such as ChatGPT?

[heady jay]

Funnily, I hit the Ai track backwards. Perplexity I discovered first from some source; and honestly haven’t needed or used ChatGPT

Perplexity has given and hits the use case I need. Don’t want to bounce around to “all” the Ai options; seems like there’s new products or services every day that are just wrappers.

I’m looking and hoping to see Perplexity grow and evolve, I like it a lot. Now if the Gods of Perplexity would just give me an opportunity for a onetime lifetime license price… I’d be willing to pay right now 🙂

[vapid onyx]

Is there a major difference between iOS and Web for perplexity? Haven’t used it yet on web

[heady jay]

Had used web briefly in the beginning, now solely iOS app.

Not sure if there’s any substantial differences

Just as a testament of how much I appreciate, respect, and utilize Perplexity… it’s located on my bottom app bar for both my iPhone and iPad 🫡

[turbid iris]

Hi, I have a question... When I activate Copilot, I would be using GPT-4, right?

[fervent halo]

Only for summarizing the information gathered from sources. Search, source selection and information collation are performed by fine-tuned GPT 3.5. This collated information is then summarized by GPT 4.

[turbid iris]

So it would be a mix between GPT-3.5 and GPT-4? 🤔

[fervent halo]

Yes. In the past it was all GPT4, but the devs figured that it would be more cost-effective and faster to use fine-tuned GPT3.5 for search and source collation. Although there were some accuracy and performance issues when initially introduced, Copilot's now roughly(with some caveats) on par with how it performed when it utilized pure GPT 4.

[turbid iris]

Great then, thank you very much for this information. This puts Perplexity on par with Bing chat, basically.

[fervent halo]

I would argue that Perplexity is on the whole much less prone to hallucinations than Bing chat, the citations it provides are much more relevant and accurate than Bing Chat, and it is much much faster.

[turbid iris]

Yes, that is true

[turbid iris]

I have a question about this: would it apply to the use of Perplexity with the Copilot function, or to the use of Perplexity in general?

[fervent halo]

Perplexity in general(the default and only model available to free users when not using Copilot) uses fine-tuned GPT 3.5 for all its actions. That is, search, source selection and collation, as well as the final summarization of information.

[turbid iris]

Ok, thank you

[tame current]

Have you ever thought about ordering openai’s enterprise version package? That package includes Unlimited Advanced Data Analysis! !

[flat agate]

I can see the share button has improved the UX, loving it! ❤️

[silver laurel]

Anyone know the prompt to call out previous related topics from the same thread

[tribal gale]

gm

Oh yes, the new logo is sweet, LED, LAZERish pew pew'in goodness

[lavish flint]

Hello

[silver laurel]

Hello

[tame current]

the response cuts off when i ask about stop sequences in openai api...

not sure if you could do anything about it since its just a wrapper but still wanted to share

[wicked blade]

Good morning guys. Greetings from Brazil. 🇧🇷 Could you tell me if Perplexity is currently the best performing option for programming assistance? Or does Phind still have a slight advantage?

[fervent halo]

If you're looking for just programming and dev work nothing beats phind, if you also do research and work with quite large files, then pplx beats phind in functionality.

[wicked blade]

So basically Perplexity would be a kind of GENERAL personal assistant, just like Bing Chat IA, correct? For any kind of queries whereas Phind would be purpose specific. Am I right?

[fervent halo]

Yeah, pplx is essentially a much more accurate and faster version of Bing chat, with multi-text and pdf file upload support.

[wicked blade]

And what is the character limit in the Perplexity input field? On Bing there are 4,000 characters.

[fervent halo]

the character limit is ~16k tokens. Above 2k tokens, the input field behaves like Anthropic's official Claude 2 web UI, that is, it gets pasted in as a text file upload.

[wicked blade]

Hmmm, interesting. This is only in the Pro version, right? I'm asking this because I'm evaluating the possibility of paying for the subscription, but still wondering if it's worth it.

As I'm Brazilian, converting dollars to reais is a little expensive here. 😅

[fervent halo]

Yeah, I'm quite sure that large file uploads are only available in Pro, also GPT4 and Claude 2 access is only available via Pro.

[wicked blade]

Considering how long you've been using Perplexity in the Pro version, could you tell me if it's worth the subscription compared to using Bing Chat IA for free?

[fervent halo]

For me, as someone working in the Bioinformatics and Data Science fields, its definitely worth it for its ability to read larger files and its much higher accuracy. There are some limitations you should however be aware of. For agent-type work like booking flights and accomodation, Bing chat is better than Perp. However, for research and file-analysis pplx is definitely miles better.. So, my verdict is that it depends on who you are. If you need to do lots of large file analysis and web-research then Perplexity Pro is for you, if you're purely into software and dev work, then go with Phind, otherwise, if you're a more casual user, just needing to use AI for the occasional cheap flight and such, use Bing chat.

[wicked blade]

Well, I'm a software developer. To help with programming, I already know that I will adopt Phind. For general use, such as answering my personal doubts, studying, analyzing what the law says on a certain subject (as I am also a military police officer) and asking for advice, I was in doubt. Until then I only used Bing Chat IA, but it sometimes has some very annoying bugs and inaccuracies.

Thank you for the feedback. It helped me have a better understanding when deciding whether to subscribe to Perplexity's Pro plan. I will sign and test.

[fervent halo]

You're welcome 😃

[wicked blade]

I've noticed and find it annoying that there are still certain things we ask of Perplexity that it doesn't deliver like Bing Chat AI does. For example: I asked him to bring me the names of the drone models used in the war between Ukraine and Russia. Perplexity said there were no model names in the search results. Bing clearly showed the drone models used, with a brief description of each.

[wraith crow]

Perplexity AI: Give me the names of the drone models used in the war against Russia https://www.perplexity.ai/search/Give-me-the-4WRIzs3XT1yikYYUS_KnNg?s=mn

do you mean something like this?

It may be the way you phrased the question

[wicked blade]

Exactly. Strange that my question was exactly like this: "What FPV drones were used in the war between Ukraine and Russia?"

[halcyon coral]

Since you originally wanted the names of the drone models, your prompt should also look something like this: "List the names of all drone models used in the Ukraine-Russia war."

[wicked blade]

I understand.

Does anyone here also invest in shares and funds on the Stock Exchange? If so, what did you think of Perplexity's performance for financial market analysis?

[wicked blade]

?

[halcyon coral]

Perplexity can, if you ask, analyse the latest news on the economy and especially on the stock market.
However, you must be aware that Perplexity may also use outdated data and Perplexity is in any case not liable for incorrect predictions.

[wicked blade]

I understand.

Can't Perplexity also estimate the arrival time from one destination to another by analyzing metrics from Google Maps or any other map?

[halcyon coral]

Google Maps is not used directly, but the information is usually very accurate.

[wicked blade]

I did more than one test, to estimate the travel time from my home to my work, but I only received the same answer.

When I do the same test through Bing Chat IA, although it doesn't directly use Bing Maps, it brings route information from that map.

[halcyon coral]

I think it doesn't work so well on short routes.
https://www.perplexity.ai/search/How-long-will-vs2VH2X0TfmP.9Verli7tQ?s=c

[wicked blade]

Was this test you did using Copilot?

@halcyon coral

[halcyon coral]

No, normal GPT-4!

[wicked blade]

I thought the normal GPT-4 was just the Copilot mode. 🤔

Could you explain the purpose of Copilot and its difference in relation to using only GPT-4?

[halcyon coral]

Copilot uses GPT-3.5 FT (fine-tuned) + GPT-4 or Claude 2. You can use more imprecise prompts with Copilot, as Copilot then asks queries back to you so that the answer contains exactly the information you want.

[wicked blade]

So the Copilot chooses which model to use, between the GPT-3.5 FT, GPT-4 and Claude 2, according to the context of the prompts?

[halcyon coral]

GPT-3.5 FT is always used for the internal routing process, for example to make the queries. You can select the main model, GPT-4 or Claude, in the settings: https://www.perplexity.ai/settings

[wicked blade]

Perfect. Thanks.

[hard cedar]

hello

[lament imp]

hello

[bronze wigeon]

Hello

what is your name

[wicked geyser]

Has anyone figured out a good way to use perplexity with speech-to-text?

[bronze wigeon]

ay

frio

[true shadow]

Heelo

[waxen imp]

I'm quite curious if perplexity has been achieved with the fine-tuned GPT-3.5 Turbo? The product seems to have struck a good balance between speed and accuracy.

[lofty dagger]

hey

my mom thinks theres a company called ai inc

she says it. over and over

to piss me off

shes taunting me

[undone olive]

Idk, it sure does work a lot better then before. Everyone still says that perplexity has the latest knowledge but gpt 3 not 3,5

[balmy dew]

I have a quite long document, it is a test prep, I initially wrote the draft by using CHATGPT4, before discovering perplexity, now, i have 300 questions with 300 answers, in the meantime there's been an updated in the codes refering to the questions and answers, I copy and paster my Q&A and i asked Claude2 with online research if the q&a were up to date, correct and relevant, it did say that everythins is correct, even tho it is not, chatgpt4 hasn't enough memory context, can you please advice me how to tackle this issue?

What is the best way? to break down the document and use GPT4 with browsing support and ask to correct both the qustions and the answers referring with the new, 2023 updated codes, or is there a fastest, more efficient way?
how can i efficiently correct my codes with the updated codes? how can i double check the accuracy and fact check?

[lament imp]

Personally, I feel like Perplexity does well as long as there's at most 3 questions being asked in a single input. I believe it to be due to the fact that it search specific things that it displays to you when compiling information. So that's why it may say a blanket statement of True, due to the fact that it searches out specific questions from your Q&A.

[tame current]

couldnt care less 🥱

[foggy gorge]

I don't know when we will be able to use plugin functionality like an interpreter

[halcyon coral]

This is not planned until now, Perplexity is an AI search engine and not a chatbot like ChatGPT.

[foggy gorge]

Ok

[halcyon coral]

Perplexity can only perform a certain number of searches per message, so for best results you should only have about 3-5 questions checked at a time.

[tight gull]

Is there any article or post somewhere that explains how perplexity is able to give you references? I know approximately how ChatGPT works (it generates tokens autoregressively and uses beam search or similar to find the most appropriate answer), but that doesn't search the web for answers to your questions, and it can't give you references. So how does perplexity do it?

[tame current]

there isn't afaik

I think they just feed search results into chatgpt and tell it to answer the question with citations but we don't know how they actually do it as they never revealed that

also, chatgpt did have a search plugin

[stray gorge]

Yeah, that's the thing I've always wondered. Like, how is it capable of reading 30+ separate pages all at once and in just seconds too?

[muted linden]

Here's a good video on this: #👉│feedback-general message

Seems perplexity uses its' own index and other sources + they optimize for speed as a USP.

[rain silo]

Cache

[sweet locust]

yow guys im new here i have paid version of perplexity what do you think best model (gpt 4 ,perplexity or claude ..) ?

[granite minnow]

Ask perplexity.... 😉

[sweet locust]

so perplexity better than gpt 4 hmmm

[flat agate]

It depends on the context, for creative writing better Claude2 normally is better and it's context is bigger, but for other usages gpt4 could be better, I suggest you to make the same prompt on both and see if one is more appropriate for you

[halcyon coral]

GPT-4 is probably the best, Claude 2 for file upload and writing.

[flat agate]

Yes 😁

[sweet locust]

im talking about integration models in perplexity plateform i knoow chat gpt open ai plateform is the best of the best xd and nothing worth it more than chat gpt 4

[flat agate]

I think you should try booth and compare but as @halcyon coral says gpt 4 is better for the most part of tasks

[sweet locust]

thanks body 😄 i just want know whats the difference between free version and paid version because the free version doing work well

[sly sigil]

Hi, I read an article that Perplexity incorporated LLama and had a chat option called Llama Chat, but I can't seem to find it. My goal is to address coding issues based on Code Llama. How or can I do that within Perplexity?

[muted linden]

Llama and Code llama are implemented at https://labs.perplexity.ai/
Llama is not part of the main product.

[wicked blade]

Hey guys. Greetings from Brazil! 🇧🇷 One question: can Perplexity read images like Bing Chat AI?

[sly sigil]

Thank you

[grand mauve]

Hi! Not yet! 😄

[tame current]

New announcement is pretty nice

[tame current]

Damn

Wait I pinged you, sorry!

[sterile wedge]

Hi

[tame current]

I won’t delete that since that would simply ghost ping

Hello

[proud robin]

When will the Perplexity page in Spanish be released?

[tame current]

Doesn’t it already support Spanish?

[proud robin]

Yes, but when you write later it puts you in English or sometimes in Spanish

[tame current]

Oh so it forgets

[modern crown]

Is perplexity looking into the memory issues around co-pilot? It seems that it doesn't remember conversations that well.

And also are they looking into adding where you can switch focus modes in on thread without having to change it?

[modest python]

Agree, in some conversations seams Copilots goes wrost than a simple prompt. But, Copilot is good enought if you ask for step-by-step planning 🙂

[river zephyr]

With GPT4 allowing Zapier integration to create Zaps, is there any work underway to allow Zapier integration in a similar way with Perplexity?

[autumn socket]

Hi! Is there a school or education subscription for perplexity?

[halcyon coral]

Hey, @autumn socket!
Not yet, there is only the referral link.

[halcyon coral]

Do you have a Pro subscription? With this you have access to GPT-4 32k and Claude 2. With the free version it can be a bit tight with the context size.

[halcyon coral]

Yes, interest was definitely shown on the part of the team!

[autumn socket]

Thanks!

[grizzled canyon]

So which features of perplexity use the 32K model?

[fervent halo]

I believe GPT4 32k is only available to Pro users due to cost.

[grizzled canyon]

I am asking which features provide access to 32k, because as far as i know the inference cost is different for 32k compared to 8k and there is no explanation where it is utilised over 8k

[fervent halo]

32k is utilized dynamically as a conversation thread gets long enough to warrant its use, as determined via some threshold that hasn't been explicitly clarified by the devs. Copilot use whilst on a Pro subscription likely uses 32k as well.

[wraith crow]

Correct it will automatically switch to 32k model for answers that need more than 8k tokens of context. This is only available for pro members.

[grizzled canyon]

cool cool that's delicious, thank you good people

i had "pro" before but only saw benefit for conversations with longer context and wasn't sure if that context was available

this clears it up

so if i search, lets just say across all the web generically, that will switch to 32k in theory if it goes beyond 8k tokens in aggregate?

[fervent halo]

Yes, it should

[grizzled canyon]

❤️

[wicked blade]

😢😢😢

[formal parcel]

Does Perplexity use 32k?

[fervent halo]

No, 8k I believe, for regular Copilot, 32k is only available for pro users.

[formal parcel]

I am a pro user, so I have it?

[fervent halo]

Yeah, you do then. Assuming you have GPT4 set as your default model. It switches dynamically from 8k to 32k if your conversation gets long enough. Also Pro Copilot summarization is via 32k.

[scarlet zephyr]

Oh I didn’t know that

[grand parcel]

Hi

[modern crown]

@halcyon coral I do have the pro versin

Ill do more testing

[storm cobalt]

Hey all. Quick q:

When I upload a PDF and ask perplexity to analyze it, in the past it would analyze it. But now when I upload a PDF, perplexity says my PDF link is not publicly accessible. Did this change recently?

[halcyon coral]

Hey, can you please share the thread?

[storm cobalt]

is the PDF accessible if I share the thread?

[halcyon coral]

No, the PDF is not visible, but you could ask questions about the PDF. You can also send it by DM.

[scarlet zephyr]

I just tested it and I don’t believe it to be true

[scarlet zephyr]

Oh wait. Let me test it some more

[tame current]

not another rebrand

[scarlet zephyr]

I don’t think it’s GPT-4 32k unfortunately

[wraith crow]

Are you a pro user?

It has been confirmed by Denis it will use gpt4 32k when the context size exceeds 8k tokens (for pro users) #👑│pro-lounge message

[scarlet zephyr]

Yes I am a pro user

I put in a text of 25k words, and asked it to point to the last sentence of the article. It wasn’t able to give me the right answer.

It’s possible that the paste txt function is not working properly

Let me try a bit more

It’s not even working for Claude 2 on Perplexity. I tried the same thing on Claude.ai, and it worked perfectly there.

[rough wren]

🥴

[scarlet zephyr]

It just doesn’t work on Perplexity. This is very disappointing

[outer valley]

my body is ready

[wraith crow]

If you leave this discord and rejoin with the link in the settings page you get access to the pro channels. Maybe you can share your threads there where you have this issue.

[lofty dagger]

hewow

hewwo

uwu

[rough wren]

😨

[steep shuttle]

👍

[twin prawn]

Anybody know how we can use perplexity to build an mobile app?

[halcyon coral]

A great question for: #💻│coding

[lofty dagger]

hey

i have chatgpt enterprise if anyone is intrested

and google

free google bison bot

[woven plinth]

How? I'm interested.

[meager charm]

does perpexity actually surf the internet or is that a gimmick? The reason why i'm asking is beacuse the results are usually very wrong and then I click the links and it says something completely different

[abstract kite]

yes i have the same question,

[signal hamlet]

Hey @meager charm! Could you please share some examples with permalinks in the feedback channels or DM those in case you don't want to share publicly, so we could also check and investigate.

[thorn moss]

Thats what i love in perplexity moderators, they are fast as lightning in addressing concerns. 🙂

[pulsar patrol]

hi all

quick question, whats the advantage of using GPT4 over the perplexity engine?

[halcyon coral]

The perplexity model is FT (fine-tuned) GPT-3.5, GPT-4 will usually give better results.

[nocturne wyvern]

did you get a 10 billion dollar investment?

[vernal cedar]

Is the source code of the Perplexity discord bot publicly available?

[halcyon coral]

No, it's private.

[vernal cedar]

Pity, but thanks

[stray gorge]

what about cache?

i'm paying for the premium of perplexity - month 2 - how do I get access to the private server

[halcyon coral]

Please leave the server and join with the pro invite link in your settings.

[balmy slate]

hey is Perplexity just a fine-tuned version of GPT-3.5 or 4? I wonder if the question is sent to the openAI directly withouth going through a vector DB. I ask this because for some use cases it seems pretty fast as if I was using GPT-3.5 in openAI's website. On some other cases, like if I upload a document and ask it to summarize it, it takes some more time and streams the response a little slower, like GPT-4.
So what makes someone choose Perplexity over GPT?

[stray gorge]

ok thanks

[fervent halo]

Web search and response citations/accuracy mostly

[wicked blade]

I think he meant: what makes someone choose Perplexity (with the GPT-4 model selected) over GPT-4 itself directly from the OpenAI website?

[fervent halo]

Again, its accuracy, because the perplexity team pre-prompts GPT4, and modulates its responses using something akin to CoT I think to ensure greater accuracy than the vanilla GPT 4 available on OpenAI

[balmy slate]

@wicked blade thank you both

[wicked blade]

😉👍

[wicked blade]

Does Perplexity maintain creative abilities such as writing poems, songs, etc.?

This is at the level of GPT-4 and Claude 2.

@fervent halo @halcyon coral

@neat elk

@signal hamlet

[halcyon coral]

@wicked blade You don't have to ping half the server....

[wicked blade]

Sorry. Sometimes I have doubts about who to ask for help.

[halcyon coral]

Want to know if PerplexityAI can be as creative as ChatGPT?

[halcyon coral]

In Writing Focus, Perplexity can be very creative, I personally think the Claude 2 model is best suited for this.

[wicked blade]

Yes

[wicked blade]

Thanks.

[jaunty breach]

My perplexity can't answer any questions!

only today

[halcyon coral]

Hey @jaunty breach,
please share some examples in the feedback channels!

[jaunty breach]

[OK]

[magic cargo]

question: I saw someone list custom instructions for the responses that they get from Perplexity in the prompt-lib channel. How does one create the custom instructions? Is there a field in the AI Profile?

[halcyon coral]

You can put the instructions in the "Intro" field.

[mystic kelp]

Question: Perplexity focusing on Writing still browse on the internet or no?

For plagiarism purposes only. TYIA!

[wraith crow]

Writing mode doesn’t have internet access. You’re talking straight to the model. And it doesn’t have knowledge of things that happened after september 2021

[mystic kelp]

But I created an article using Perplexity focus on writing, and I pasted it on Duplichecker for checking plagiarism. It has plagiarism, 38% plagiarism, and 62% unique. So it means the AI can collect some article section on SERP.

Any thoughts on this guys? TYIA.

[orchid magnet]

New perplexity update!

V

[thorn moss]

@orchid magnet what update?

[orchid magnet]

1.20.1

[thorn moss]

Android or ios?

you mean?

what's with the update btw?

[orchid magnet]

Wait I tho this was the Minecraft server my bad

[thorn moss]

yeah you said perplexity update then you thought mine craft server. kinda weird for you. 😂

[lofty dagger]

looking for a job

im a large language stupidity model

i will demonstratemy java knowlegd

[opaque cobalt]

Hey I wanna ask. How does Perplexity Pro compare to ChatGPT plus? Is it worth it to pay for Perplexity Pro or GPT Plus is already good enough

[halcyon coral]

The big difference is essentially that PerplexityAI is an AI search engine, while ChatGPT is a chatbot.
As long as you don't need to rely on the code interpreter, a Perplexity Pro subscription is the better choice.

[undone olive]

Perplexity Is a search engine so it will mainly focus on providing you with information but Chatgpt is a chat bot so it will mainly focus on writing text and code.

[granite ocean]

Why when I click purchase the stripe website just keeps does this: purchasing ... And never stops

[fervent halo]

I'm unsure, probably an issue with stripe itself, and nothing to do with Perplexity tbh, maybe try and email support at [e-mail]. Stripe also has support options located here:
https://support.stripe.com/contact/login

[delicate rune]

Hi! Out of curiosity (and ignorance), what will Perplexity's distinguishing benefits be when OpenAI allows browsing again?

[halcyon coral]

PerplexityAI has focused purely on browsing.
To be honest, OpenAI's browsing function has never impressed me.
Perplexity already has a lot of experience when it comes to searching and evaluating sources, so I doubt that OpenAI is a serious competitor.
ChatGPT does not have a collections function either.

[lunar wharf]

Perplexity Pro (https://www.perplexity.ai/pro) states that there are 300+ uses of Copilot per day. When I use it, I see 600+ uses per day. Is this change official? Perhaps update the official site? My question: if I am not using Copilot, how many questions can I ask to Claude+GPT? 300+ or 600+ daily? I would assume 600+. Thank you.

[wraith crow]

The 600 uses a day is for copilot/gpt4/claude together. If you use 1 of those it will reduce the amount of available uses by 1 for every question you ask.

Also if you’re a pro user you can leave the server and join again with the invite link on the settings page in the pro section. It will give you access to the pro channels when you join with that link.

[glass solstice]

FYI, posted an article today that leaned on Perlexity https://www.etftrends.com/disruptive-technology-channel/ai-the-post-hype-golden-age-is-here/

[lunar wharf]

Thanks. So if I use Copilot+Claude, my uses will go down by 1. If I use Claude without copilot, my use goes down by 1. Basically, use goes down by one for every question I ask, regardless of what options are turned on and off?

[wraith crow]

correct. the only thing that is unlimited is the Perplexity model without copilot

Well basically the other options are also unlimited because you will never go through the 600 uses each day. But perplexity model without copilot wont reduce your uses

[lunar wharf]

Perfect! Exactly as I thought. Perplexity is a great tool. I have been on ChatGPT+ since last March, but I think it is itme to switch. Note though the introduction of Dalle-3. Are there plans at Perplexity to do anything with imaging?

[delicate rune]

Thanks.

[tame current]

Got a question

[halcyon coral]

You are welcome to ask this question here!

[sacred verge]

My perplexity can't answer any questions! I just ask 5 question, but I am pro subscriber, why?
only today

[tame current]

share the thread so devs can look at it

[sacred verge]

how to share the thread?

@tame current where to get the thread ?

[tame current]

there should be a share button at the top right

[sacred verge]

https://www.perplexity.ai/search/perplexityai2-gPfu6gKATWSeID7dhy_8dQ?s=c

I really like perplexity.ai, but this issue is affecting my usage. Could you please help me solve this bug? Thank you very much. I am in a hurry to write my paper.

[halcyon coral]

Perplexity uses a dynamic return of usage, which means that if you use Copilot/Claude 2 or GPT-4 you get exactly that usage back after 24h.

I just saw that you posted a screenshot in another channel.
The problem should be fixed if you log in again or reload the page.
Sometimes waiting also helps.
@sacred verge

[sacred verge]

I just logout and re login, but seems the same result, while I can use chatgpt4 as usual...., so I think this might be a statistic bug??

[halcyon coral]

Normally this is because of high demand of Perplexity. It should work again after some time.

[sacred verge]

Ok, I can use coipot as usual, Thank you！I hope this small bug might fix later

[fervent halo]

Faster responses and much greater accuracy, plus web image retrieval.

[gilded breach]

I've really come to enjoy the accuracy of Perplexity and for my use case I'm really really hoping that an API becomes available. But in the mean time, is it possible for Perplexity to access a webhook I have set up?

[carmine linden]

Hey @gilded breach! Perplexity AI does not currently provide an API.

[vestal wing]

Does perplexity pro support images? Like PNG?

[halcyon coral]

Hey, @vestal wing!
Image analysis is not (currently) supported.

[carmine linden]

Hey @gilded breach! Perplexity AI does not currently provide an API.

Hey @gilded breach! Perplexity AI does not currently provide an API.

[delicate rune]

Thanks.

[lofty dagger]

guys

i got the raw gpt4 model

its gpt54

gpt5*

i cant talk about it

its twice as fast as gpt4

talso they are the raw ones

instruct

also i need a coding budd y badly

im sick of coding alone

[tame current]

how lol

[tame current]

Use gpt4 for your professors

Professors won’t know

[dusty bobcat]

Without proof this is a horrible statment

[tame current]

Is using ChatGPT to help you understand and structure an essay allowed or is that a form of cheating?

Like I read the book chapters that I wanted to use and asked GPT to help summarize it and help provide a reference to answer the questions

[steady mason]

Thanks for posting! And I'm with you--I haven't opened Google since opening Perplexity...

[heady birch]

Yo where did the left hand nav go

Perplexity is useless to me a one-off search

[tame current]

?

[undone olive]

?

They are still working on bugs and fixes on gpt4 but gpt5 already?

[halcyon coral]

No, he simply told untrue things. GPT-5 has not yet been published.

[warm lichen]

Is there any macbook app that can convert dictation to text so that I can write long documents without typing all the way and the application should be open source

[chilly grove]

I just read they haven't even started training 5 yet. But perhhaps he meant 4.5. That is in the works

[zenith atlas]

You should check it with your school\university as their policies about AI are not universal and can differ from teacher to teacher.

[fervent halo]

The usual advice for assignments applies here, use the AI to enhance your understanding, but don't outright lift entire paragraphs or even sentences the AI provides. You can use the sources Perplexity provides within your research, as part of your references, and you will be fine as long as you've looked through them thoroughly and understood how they integrates with the assignment you're doing. I would recommend writing the entire assignment yourself, and after having done so, you can ask Perplexity to suggest improvements in sentence structure and whether there are any grammatical errors that need fixing. And choose for yourself, which improvements you wish to incorporate, and wish to forgo.
Finally, and perhaps most importantly, don't leave things to the last minute, since doing so is what usually tempts people to lift an AI's response wholesale, when they're tied for time.

[stoic wave]

Its allowed for learning purpose and nothing wrong or ilegal with that.

[flat agate]

I believe it is the best advice I have read about generative IA usage

[rustic gate]

Check to see if you have any dark mode installed in your browser, or turn off extensions and try again. That is what happened to me with the dark mode extension. Switched to another extension called dark reader and I could then see the whole perplexity ai page, collections etc. good luck

[undone olive]

No, It's not. Using ChatGPT or any other AI for references or learning is totally acceptable but directly copying from chatgpt without learning either taking reference is cheating and not acceptable

[tidal radish]

for some reason .. i just reloaded "https://perplexity.io" on my browser and I get a blank page

[wraith crow]

the official website is https://www.perplexity.ai

[solemn trench]

i want to export all of my data then delete my threads

is that possible?

[fervent halo]

I don't think it is atm

[solemn trench]

I think it should be it is an obvious step to be able to export the data the platform has about you, very common across mature platforms

[fervent halo]

Actually, its a bit more difficult than you might imagine, if its just thread queries and responses, then maybe. But, if you include all the file uploads as part of that data, then it could easily run into the hundreds of gigs/terabytes. Plus, I believe file uploads are only retained on-server for 30 days, so if they enable such a feature, it likely won't include the respective files you uploaded as part of the data retrieved.

[solemn trench]

i think i would be interested in just my own threads

but i recognize one can opt out of using the data for training, which i guess is my main concern

[muted linden]

This is actually mandated by EU law (at least that is must be possible to export and delete all data from a digital platform). While this might be not the sexiest feature to implement, I think perplexity should consider it 😉

[solemn trench]

i think california has a variant now

[halcyon coral]

Yes, there is a separate regulation from the EU "General Data Protection Regulation" (GDPR), but there is no obligation to add export or delete functions.
If you or @solemn trench would like to view or delete your data, you can always contact Perplexity Support at [e-mail] 🙂

[muted linden]

I'm good, thanks 😉 Just wanted to explain why "the big players" have it - so they don't have to fulfill every request manually. In the end, as long as it is possible in any way (manually or automated), it should be in line with the regulations 👍

[solemn trench]

well also another thing is usability, i have too many threads lol

i want to delete them from the platform to make it faster, but dont want to lose the info in case i ever see a gem

[halcyon coral]

All right, but I also think that automation would make sense, especially because of the increase in the number of users.

[real peak]

Hello team Perplexity! You've got a great app. Love it! But one suggestion. Can you make the "return" key when pressed down, go for search instead of adding a new line break? This is in the ios APP's search bar by the way with palceholder "Ask anything..."

[narrow silo]

can anyone with paid perplexity get me ansswer to one question?

[wraith crow]

what is the question

[narrow silo]

ok, I will try to be as precise as i can

near the enf of the movie Tar Lydia Tar is watching video on tv of conductor giving a speach about music. who was that conductor?

current answer I get is

Therefore, Lydia Tár is not watching any conductor on TV tapes, as she is a fictional character in a movie.

which is not true

[tame current]

just normal copilot doesn't seem to be able to find the answer

[wraith crow]

“Near the end of the movie "Tár," someone is watching a video on TV of a conductor giving a speech about music. The conductor in question is the fictional character Lydia Tár, played by Cate Blanchett[2]. The film, directed by Todd Field, follows the story of Lydia Tár, a renowned conductor whose life unravels after a series of scandals come to light[1].“

[narrow silo]

so this is what you got on paid?

that someone is Lydia Tar , of course

[wraith crow]

so this is the right answer?

[narrow silo]

but name of conductor is still????

[tame current]

lol

[narrow silo]

coz she'd watching male conductor who inspired her

talking about music

i liked that speach and was wondering who that guy is

[tame current]

The conductor was Leonard Bernstein

dunno if right, found it from a comment on a random article

[narrow silo]

yes, that's right

[tame current]

fastgpt couldn't answer but it found the page

https://kagi.com/fastgpt?query=near+the+end+of+the+movie+Tar++Lydia+Tar+is+watching+video+on+tv+of+conductor+giving+a+speach+about+music.+who+was+that+conductor%3F

[narrow silo]

hmm... so no good help from AI in this case

[tame current]

it did find the article

just couldn't answer it

this is another reason it's important to check the sources

[narrow silo]

should i report this to perplexity team?

[tame current]

sure

[wraith crow]

It is able to find it if you’re a bit more specific https://www.perplexity.ai/search/Who-is-the-mn5m_.oXTHq.B2LrjTpFHQ?s=mn

[narrow silo]

thank you

[meager plaza]

I'm not quite sure where to ask this, but when it runs out of tokens in its context window, does it remove working out (such as the results of previous websearches) or does the window just crop out the top tokens first?

Because it starts doing irrelevant searches about things that are on my profile after a certain amount of messages have been exchanged

I'm going to assume this is the wrong place to ask. I admit I don't understand discord that well.

[tame current]

https://www.youtube.com/watch?v=5rEZGSFgZVY

[tame current]

no, you're in the right place

[smoky parcel]

I want to love Perplexity so much, but I can't pay $20 a month for an AI that can't understand the actual question and therefore does a worse job selecting search queries than I would if I tried to search myself, doesn't understand basic context often (like what day it is, or simply common sense understanding of context around a question), and most frustratingly, that like 40% of the time replies asking me to answer the exact same question that I asked.

Sucks, because it's not the actual AI that's the problem. It's just dealing with every question it's asked in a rushed manner with seemingly very little attempt to actually understand the question beyond identifying keywords in it.

I really hope that these pieces get addressed. I don't need more custom instruction profile fields or anything else. I need it to slow down, understand my question, devise a search strategy and maintain context throughout a thread. If those things happen, you'll get my $20/mo again (that price is wayyyyyy too high anyways fwiw)

Oh another one, it should recognize when it's about to answer a frustrated follow-up question with the exact same response it gave before the frustrated follow-up question and very much never ever ever do that. I can't afford to buy a new computer when I smash my computer screen lol

🙃 Oh ok then.

[outer glen]

uh copying large text into the ask anything box just converts it into a "file.txt" and then wastes a upload even if you dont proceed witht he question thing is this a bug

[halcyon coral]

This should definitely not happen 😅...
Please report the answer with the flag icon below the answer.

[lament imp]

This also happened to be at one point.

I thought it was a feature

[dusty bobcat]

Does anyone use perplexity and phind pro?

[tame current]

You can share things while not signed in?

[knotty marsh]

#📱︱feedback-mobile Threads aren't syncing on the Android App. I created a thread in a web browser. The thread syncs with a different web browser and the iPhone App, but the thread does not sync with the Android App. I signed out of the Android App and signed back in, but still no joy.

[dark mortar]

Is there a great github repo search, or can prplx help me with it? I'm often looking for sample github repositories to build upon, but websearch engines suck at finding good github repos.
Example:

• I want a sample nodejs prorject using twilio apis with more than 100 stars and updated in the last one year

^ This is reponse of perplexity, the linked repo is archived and only have 79 stars, the prpx only gave one result 😦

(Sorry if I came across a harsh and demanding user. I really really want Perplexity to succeed and think you folks are doing a great job)

[night lance]

Does anyone know an ai that can humanize a text?

for free ofc

[narrow silo]

What do u mean by that

[night lance]

I need the text that is given by perplexity ai to not be detected that it is an ai written content

[narrow silo]

Did u check grammarly

[night lance]

grammarly isnt good enough and someone said we needed to use the academic mode but u gotta pay for it

[narrow silo]

Unsure what would be good enough then

Anyway that is your starting point

Like, ask which pages are similar to grammarly and go from there

[proven rain]

i get these sometimes the blank answers not sure why

[zenith atlas]

You can do it with perplexity itself. Ask it to rewrite the text so it will look like non ai written. You can also add different stuff as like “use formal but simple language” and other stuff. I used to do it this way in chatgpt and worked good

[grizzled canyon]

yes just prompt it to speak plainly, give it some pieces of text and do so then use that as a template/format to emulate

[grizzled canyon]

you can put links into the search for it to cite

[honest yarrow]

I’ve been getting “something went wrong” nonstop where the query will sometimes resubmit multiple times automatically and continuously fail. iPhone beta app via test pilot v2.3.0

[honest yarrow]

Using copilot

[wraith crow]

please post the feedback for the beta versions in the beta channels #🍏│ios-beta-feedback

[molten wren]

Hi New here

[lament imp]

Yeah, what's up?

I used Phind Pro but currently am using Perplexity Pro primarily.

[storm cobalt]

hey all. quick q. what is the daily limitation for focus: writing with claude 2 selected?

[timber garden]

You have 600 copilot/gpt4/Claude2 request every 24 hours.

[storm cobalt]

Kk ty

[storm cobalt]

does anyone know if 'copy to clipboard' has a keybind?

[thorn moss]

loving the android update. perplexity working hard. love it!

[scarlet zephyr]

Good app update

[jaunty breach]

can not use anymore，iOS，super bad update

[halcyon coral]

Feel free to share your concerns in #🍏│feedback-ios 🙂

[jaunty breach]

same

[cunning sky]

Hey everyone.... anyone wanna be frens?

uwu

[noble quest]

Were there any updates on copilot mode? Did it become better?

[thorn moss]

Oh my i have noticed late there is an image and video now in android app. woooo. very cool. Thanks!

btw was the copilot search buffed? seems provide a better results.

[honest yarrow]

Post in feedback channel there is thread

[zinc folio]

bro

[zinc folio]

ok

[vapid onyx]

Test

How do I get premium discord access if I’m a subscriber?

I clicked the button in my account and it brought me here

[modern crown]

Which model do most people use here? Claude 2 or GPT-4?

[undone olive]

Perplexity

[modern crown]

@undone olive is perplexity GPT-3.5?

[undone olive]

No it's gpt 3 with latest information

3,5 is better

But 4 with latest info is op

[modern crown]

I was talking about the models in Perplexity that is offered @undone olive

[fresh grail]

Leave the server, then click the link to come back and you should have access to the pro channels

[proven rain]

https://www.axios.com/2023/09/25/testing-chatgpt-support-images-voice-search

[grizzled canyon]

no it's not gpt3 it is gpt3.5 fine tuned

[vernal moon]

iOS bug

[vapid onyx]

Thank you! that appears to work

[orchid magnet]

Hey guys

[orchid magnet]

Bruh what

How

[orchid magnet]

Is anyone here

[storm cobalt]

hey all quick q - when does copilot reset? is it a specific time?

[flat agate]

Every 24h you recover the uses you spent 24h before, so if you spend 1 use now it will reset tomorrow at this hour

[devout sun]

The speed of answering questions is getting slower and slower

[dim igloo]

Is Perplexity currently down ?

keeps saying "Checking if the site connection is secure"

[wraith crow]

Do you have the perplexity chrome extension installed?

[dim igloo]

not really.

Do premium users benefit by using the extension?

[wraith crow]

You can use gpt-4 and claude with the extension as a pro member.

[wraith crow]

If you're still facing this issue try to follow these steps hope it helps:#🌐│feedback-web message

[dim igloo]

just not the same 😦

any ETA's for the website to come up online ?

[dim igloo]

Unusable to say the least :/

[halcyon coral]

You tried #🌐│feedback-web message ?

[dim igloo]

logs me on my non-Pro account and I'm unable to log out off ..

[halcyon coral]

So you visited labs, then went back to perplexity.ai and logged in with another account?
And "sign out" doesn't work?

[dim igloo]

correct

[halcyon coral]

Please try to delete all browser data, cookies,... for Perplexity

[dim igloo]

worked 🙂 but no answers so far 😮

been 15minutes waiting

[tall horizon]

there used to be an universal Perplexity mac app. is it still available?

I found it useful since the cloudflare security check on the website became more and more annoying... 🙈

[halcyon coral]

Is it still not working for you?

[dim igloo]

logs me in ... but no replies are generated

[icy raft]

I have the same issue.

[tall horizon]

If you are on a PC, you can install an Android emulator and use the android app as an alternative way... It sounds tricky but actually its quite simple and fast.

but really wish the team can fix the Cloudflare issue though

[modern crown]

I really want Perplexity to succeed, but it seems that it gives a lot of inaccurate stuff

[undone olive]

No it doesn't you just have to adjust your prompt

But I wonder when perplexity will have a continue function