💬│general

[rigid gull]

Edit: fixed the problem. For some reason the app is not working when I have the cloudflare warp vpn app running. Old post; Has anybody had problems with the iOS app of late? Starting around yesterday the app would not ask questions. I deleted the app and reinstalled, when I tired to log in via google it just kept spinning trying to log in but would not get any farther. Web works fine

[tame current]

i kinda love perplexity

the UI is like ctos

[tame current]

yeah

perplexity kinda look like it dont it

[tawny egret]

So silent

[thorn moss]

We go loud when perplexity release incorporation of GPT 4 vision when Openai release API. 😂😂😂

[sweet ore]

same thing, warp is on, perplexity web works well, app not

[gusty blaze]

Thanks

[patent nexus]

how would i bypass ai content detection? are there any new methods?

[digital siren]

Are there any plans to introduce advanced data analysis for gpt-4?

[thorn moss]

I cant wait anymore. hypeeed. 🙏🙏🙏

[sleek vortex]

Tbh it can’t be that hard

Really all it is like some backend that can run docker containers or an isolated environment

And then a fine tuned model that runs code as part of its CoT

Maybe through function calls or something

[digital siren]

It’s a wonderful feature for anyone in an analyst role. Visualizing data in graphs or just getting another take on it is amazing.

[storm cobalt]

im assuming since perplexity launched API for LLMs, eventually API for account will be released? would be so much easier if we could query via an API each day.

[glacial forge]

How do I feed Perplexity tables, diagrams etc

[sleek vortex]

You can give it tables as text

Though it can’t take images (yet?)

[glacial forge]

Whatchu mean

[tame current]

Damn perplexity just recommended a piracy site to me

[digital siren]

I used GPT-4V to have it code the tables I had into python

[tame current]

W

[hoary holly]

Just saw the ad on Reddit. Well done it looks sick

[ruby pulsar]

what does the term "autonomous agents" mean? Is it mean that we can finally get rid of current pplx's copilot?

[halcyon coral]

Hey @glacial forge,
while Perplexity cannot (yet) evaluate images with graphics, you can, for example, upload a .csv file with data.

[halcyon coral]

Please file a report in #1140622008086970420 with a permalink to the thread containing the site in question.

[tame current]

Ah, I already closed the app

[halcyon coral]

Were you not logged in?

[tame current]

I don’t even have an account to be honest

[halcyon coral]

All right, if you see a website like this again please report it, so something like this can be avoided 🙂

[tame current]

Will do!

[tame current]

Well, I just asked it a related question and it sent another piracy site (unless I’m wrong)

I’ll report it now

[rigid gull]

How do we report this? To Feedback-general or something? Seems like the perplexity support staff need to contact cloudflare.

[halcyon coral]

You can report bugs in #1140622008086970420.
You can add your information to: https://discord.com/channels/1047197230748151888/1162633810769362974

[muted linden]

Random thought of the moment: "Googling" feels so archaic 😆

[steel wagon]

Perplexity-ing doesn't quite have a ring to it though 😂

[muted linden]

Agreed 😅

[patent nexus]

plexing it

idk

[jagged ivy]

Does gpt 3.5 with copilot work better with ai profile than without copilot?
1 pic it's without copilot
2 pic it's with copilot
3 pic it's ai profile setting

[south kindle]

This is unrelated to Perplexity but didn't feel like going to Reddit to share. Was having a chat with Bing about it's image recognition abilities which lead to this interaction. Something about it just made me laugh at the exchange

Follow up image showing it searching for how to refuse a request poliety. Again, i don't know why it needed to search the web for things in it's training data. regardless, it's attempts at politeness are interesting.

[signal hamlet]

Plz, plex what you found today! 😎

[sweet ore]

report to
https://discord.com/channels/1047197230748151888/1162633810769362974
I think maybe it’s perplexity server bug, not cloudflare

[tame current]

sup folks

[rose sparrow]

Can someone explain to me exactly what is the difference between copilot and regular search. Is chatgpt only available if copilot is on? Also, is there a way to have perplexity remember context longer. Sometimes two questions in it’ll start acting as if it had no idea what the conversation was talking about.

[halcyon coral]

Copilot is an interactive assistant, it reads your input and asks you questions so that your result is as accurate as possible.
In addition, more sources are used.

[muted kelp]

I found that asking questions with copilot yielded much more accurate results.

[random cloak]

at the begining i think copilot is used for coding 😂

[leaden solstice]

If I give input more than 4k tokens will convert to text file. like it shows paste.txt

[halcyon coral]

That's correct, it's a feature.

[worthy nest]

Hey, im curious, one of the benefits of perplexity pro is access to GPT4. Is there any modifications or alterations that perplexity makes that would make it give noticeably different answers from OpenAI's GPT4? Does anyone here who cannot answer the previous question have any personal evaluation as to whether Perplexity GPT4 or ChatGPT plus gives better answers?

[halcyon coral]

Hey @worthy nest!
The big difference is that Perplexity gets sources from the internet. This means that the answers are more up-to-date.

[worthy nest]

Ah okay

[worthy nest]

Out of curiosity, does Perplexity have any statements on what they are doing to further the cause of AI alignment? I cannot find anything in the FAQ or on the website unless im searching the wrong terms

[halcyon coral]

No, no specific statements were made on this 🙂

[random cloak]

does perplexity support multi-turn dialogue? i feel it cannot answer multi-turn request as well as gpt 3.5

[sleek vortex]

the word copilot is so overused

perplexity copilot
windows copilot
microsoft 365 copilot
github copilot
microsoft sales copilot

[short gale]

Microsoft has a fetish for the word

bottom four all them

[carmine linden]

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

[wraith crow]

Hi @leaden solstice, please don't share links to projects like these. These are against the TOS.

[leaden solstice]

Sorry, won't happen again

[tame current]

what was it?

[eager prairie]

where can i access the Pro discord?

[halcyon coral]

Now available for you 😉

[full island]

/info

[digital siren]

I’m also curious how I can access to pro discord? I just presumed this was it?

[zenith atlas]

There is a special link to this discord in your perplexity settings, you can leave and join again with this link to get a pro role and access to pro chats

[sweet jasper]

Just a qq, with pro what is the max output response size? I understand claude has 100k context length for both request and response.

[halcyon coral]

Hey, it's 3k with Claude.

[sweet jasper]

Thanks

[tame current]

grrrr

[cunning leaf]

Is perplexity down?

[tame current]

I can't use it either.

[halcyon coral]

Perplexity is currently unavailable, sorry for the inconvenience.
Please do not create bug reports and be patient 🙏!
<@&1151914708107014226>
Make sure to get the role to receive status notifications.

**Resolved! **

[tame current]

Bro how to get perplexity api key without pro

[split comet]

thanks for prompt notification! We'll patiently wait 🙂

[halcyon coral]

Not possible atm. The API is in closed Pro Beta.

[twin oak]

How long ?

[halcyon coral]

Normally it only takes a very short time 🙂

[strong tiger]

I have the same question, I left and re-joined but I don't think I have any Pro Discord access. (Unless I'm misunderstanding.)

[halcyon coral]

You should have access now

[strong tiger]

Thank you very much!

[frosty berry]

For whom Perplexity doesn't work either? Loads for a very long time

[halcyon coral]

#💬│general message

[tranquil monolith]

Does not work for me

says try again later

[frosty berry]

Okey. Thank you

[halcyon coral]

#💬│general message

[tranquil monolith]

thank you

[split comet]

GPT 4 is not yet working for me, anyone experiencing the same?

[rose hemlock]

Hello @halcyon coral, is there like a pro discord channel or something where we can share tips about perplexity and more deeply about how to do seo with perplexity ? 😁

[cunning leaf]

How do we get the API key with pro?

[halcyon coral]

Perplexity is still down, sorry.

[halcyon coral]

In your settings

[rose hemlock]

@halcyon coral Thanks 😁

[cunning leaf]

Thanks

[ancient owl]

Is it possible to use Copilot's functionality through the API? If so, could you please explain how?

[maiden elm]

What are you doing if it is not an indiscretion?

[halcyon coral]

Not possible, sorry.

[ancient owl]

Thank you! From Japan.
ありがとうございます。

[halcyon coral]

Can you please explain what you mean by that?

[maiden elm]

I'm wondering why do you close the site temporarily.

[halcyon coral]

It's not planned 😉

[scenic skiff]

Is Perplexity working for you all? It's not working for me at the moment. I tried it both on the web and in the app. My friend also tried it and had no output. The website is loading, but the model is not outputting

I am i Europe, but I tried it with VPN to the USA and had the same result.

[halcyon coral]

#💬│general message

[cobalt belfry]

where do i join the pro discord?

[scenic skiff]

Well, I got an output now. So it's fixed? But It wasn't working like 2 minutes ago

[halcyon coral]

At the moment it is still unstable

Perplexity should work again.
Thank you for your understanding and patience!
<@&1151914708107014226>

[wraith crow]

You should have access to #👑│pro-lounge now. @inner zinc

[maiden elm]

Anytime 😊

Thank you for perplexify us

[junior lance]

Scared me when all my Threads dissapeared

[inner zinc]

Yes got it thanks! Perplexity is incredible 🚀

[halcyon coral]

@tame current It's against the TOS.
Please do not point this out to any users!

[kind vault]

hi everyone, sorry about the downtime, small technical issues may have resulted in some bugs over the last couple hours, should be fixed now!

[wraith relic]

Are the data send to the API encrypted?

[kind vault]

What level of encryption are you looking for? In transit it's encrypted with https/TLS as most modern services are

[wide stratus]

Wow, Perplexity (free version) uses a less sophisticated GPT model than Bing AI 😳

And since when is Copilot free to use for everyone? 😱

Why is GPT-4 a financial expense for Perplexity while GPT 3.5 doesn't seem to be?

[sleek vortex]

Gpt4 api is 32x more expensive

[wide stratus]

How so?? Doesn't GPT 3.5 also have to run on Nvidia H100 (or A100, I forgot which one is older)

[sleek vortex]

This is just my thoughts/guesses but

Gpt3.5 is like 350b whereas gpt4(v) is like 1T+ parameters

so id assume more ram is required first if all

and then it would also require more compute since gpt4 is multi modal, whether or not that’s on the API yet

[sleek vortex]

If the gpus are older then it’s cheaper, isn’t it

Pretty sure gpt4 was trained on 25k GPUs

Gpt3 on 1024

[halcyon coral]

The price list explains it quite well https://openai.com/pricing

[sleek vortex]

Yeah more or less

Behind it is just the fact that GPT4 cost a ton more

1024 GPUs vs 25K in training

Which is already huge compute costs

Then inference is slower and larger context which means more time spent computing so more cost there

Plus the fact it’s multimodal

All adds up, I guess

[wide stratus]

And why is Copilot free and uses GPT-4 while Perplexity uses GPT-3.5 by default and doesn't use Copilot by default? Now I'm fully baffled...

https://tenor.com/view/excuse-me-wtf-baffled-gif-14056947

[wide stratus]

"25K" what?

wait... 25.000 GPUs? 😆 😳

[wide stratus]

"1T" = 1 trillion?

[sleek vortex]

Mhm

[sleek vortex]

Yeah, we don’t know exactly but rumoured around 1Trillion parameters?

[sleek vortex]

GPT-3.5 (fine tuned by perplexity) is cheap, and is what is used on normal searches

Copilot (free) is GPT-4, which they can provide a limited amount for free due to money made from perplexity pro

Pro subscribers can choose the model they want between GPT4, Fine tuned GPT-3.5 or Claude 2, obviously since they paid for it.

[wide stratus]

Sorry it's hard to understand the price differences if the units are not even the same... 😖 (8K context and 32K context vs 4K and 16K in GPT 3.5)

[sleek vortex]

So since perplexity is going to feed as much context as possible from the web to support answers, let’s assume perplexity uses the 32k model and 16k models for most queries

Price per 1000tokens for output on GPT4 is 0.12

[halcyon coral]

The important values are the price per 1k tokens, the values 8k/16k and so on only give the context size.

[sleek vortex]

0.004 on GPT-3.5

Which is literally 30x more expensive on GPT4

0.12/0.004

Even if perplexity has a large discount from openai or something, it would still cost much more than gpt3.5

Hence why gpt4 (copilot) is limited uses per day and more uses for paid users

[wide stratus]

ok. Have you compared Copilot replies to Prplxty GPT-3.5 replies yet? Do they differ much regarding reply-length and difference in quoted sources? 🤔

[sleek vortex]

Copilot usually asks me more questions and the output is usually less of a summarised paragraph and more closer to what i want

You can also make it do a ton more searches than gpt3.5 since i’m assuming copilot works something like this

You enter query
Asks model what to search
Queries are searched
Asks model if it wants to ask or search more

It’s like a feedback loop

Whereas gpt3.5 will just

Search once based on prompt
Give model the context and ask for answer

[wide stratus]

Oh so Perplexity feeds queries to OpenAI, is that how it works?

[sleek vortex]

Yeah

And also the Anthropic API for paid users who use claude 2, or if you upload a huge file

[wide stratus]

Copilot didn't ask me a single question yet in the 2 queries I sent using copilot. How come? 😳

[sleek vortex]

Depends on the type of query, it doesn’t always ask questions

It’s like teaching a student, sometimes they might ask questions but it’s down to them

[wide stratus]

hopefully GPT-3.5 will become open-source one day. Would be amazing

Damn, going on wiki still is more environmentally friendly than a GPT-3.5 query... 😆 time to use wiki as often as possible, I guess. 😬

1567% more, whoaaaa

Hoping that GPT 3.5 is running on carbon-neutral electricity. 🤔

[formal parcel]

GPT-3.5 is 175B, it is just a fine tuned version of GPT-3

[formal parcel]

Sometimes Carbon Neutral are just lies

Tesla: Saving the planet one lithium battery at a time

[sleek vortex]

To a chat dataset?

[sleek vortex]

Probably not…

[formal parcel]

Yes

[sleek vortex]

Right

[last girder]

Anyone know if I can chat with pdfs on the perplexity app?

[sleek vortex]

Not yet, I believe

[formal parcel]

Yes but you need Pro I think

[sleek vortex]

Oh what

[formal parcel]

On the app?

Ohh

Not Yet

[sleek vortex]

You can chat with pdfs free

It’s probablt using claude or got3 depending on size of file

hence why there’s a cap on it

[odd lava]

I would love perplexity cut the prices of pro subscription in Indonesia fr

Two hundreds thousands is too much 🙂

[digital siren]

Is ChatGPT or other ai more affordable in Indonesia? Perplexity free is still pretty amazing.

[strong tiger]

Just out of curiosity but is there any publicly available information on the Perplexity model? I read something in an article mentioning BirdSQL formed from ingesting the Twitter corpus, but I'm always curious to learn more about the advancements people are making with LLMs.

[odd lava]

Bing and Google Bard are free here but its so dumb (not accurate) and slow while getting informations, I think Perplexity is the only one I can rely on and none of them compete

[feral gust]

Which AI Model do you guys use to get the latest news summaries?

[rugged pasture]

what if you have a perplexity subscription and you accidently delete the account?

[halcyon coral]

It is best to contact [e-mail], they can help you better in such cases.

[halcyon coral]

The free version uses FT GPT-3.5, but an internet search is also used to get the latest data.

[next berry]

I just love perplexity, its so handy

[muted kelp]

What does perplexity do with the profile data you give it? Could you use that to have it take on a certain type of persona when answering questions? How much does it stick to that?

[balmy condor]

Please save it for a different server

[halcyon coral]

Hey @muted kelp!
The AI Profile is used to personalise the output, for example you can set a response structure or influence the tone.
How much Perplexity adheres to it depends on the exact wording, there are some examples in #1124060874785685554

[brave vigil]

i need help writing prompts for maths topic summary's anyone have a good format

[supple trail]

Hi everyone. I wonder if perplexity AI have limited access to journal publications. I've never seen top tier journal papers cited when I trying to find some theoretical arguments

[rose hemlock]

Hello guys, is perplexity down once again or it's just me plz ?

[halcyon coral]

It should work, can you please try again?

[rose hemlock]

It works now thank you 😁

[tranquil flare]

I sadly find this to be true but I sure hope that can change later on

[halcyon coral]

Please don't spam around with flags here, also I would like to point out that political topics are not tolerated here @tall girder.
Thank you!

[muted kelp]

Thanks! I was wondering how to best use that section.

[muted kelp]

Hmm, can we not change the AI profile in the mobile app?

[wraith crow]

Thats right, it's not possible yet.

[muted kelp]

Aw cans… Thanks.

[muted anvil]

Perplexity Pro + GPT-4/Claude-2 is so powerful

[wicked blade]

Having any issues with the Perplexity servers today?

[muted anvil]

this?

[glad tapir]

yes, having the same issue

[wicked blade]

Yes.

Apart from the long waiting time.

[muted anvil]

now it's fine

[halcyon coral]

Performance is currently affected, in most cases it helps to wait a few minutes.

[maiden elm]

Yes I'm also having this message but randomly

[tame current]

I get this "Something went wrong..." message when querying

[rough haven]

I think perplexity is now dealing with the same struggles chatgpt has. Scaling…

Only perplexity should be able to fix it faster as long as the api scaling of openai and anthropic can keep up

[tame current]

perhaps openai and anthropic prefer better access to their own subscribers?

[rough haven]

I doubt it. Openai makes losses with chatgpt so I’d assume they gladly rent out their api

For anthropic I assume they don’t care. Why would they. Their product is used and sold. Nothing more required for them.

[digital siren]

Is perplexity down again? ChatGPT and Perplexity have been incredibly slow and error laden. I’m trying to just use the perplexity model too. The quality of service using OpenAI api products is going way down hill in terms of throughput.

[tame current]

even Anthropic has a problem... so Perplexity has a big problem!

[maiden elm]

Rumor says they're scaling

[digital siren]

Yes lol but just curious if anyone else is experiencing this!

[tame current]

can anyone at perplexity answer?

[digital siren]

It would be wonderful if perplexity used llama 2 70B as 4 model option.

It’s much faster and the output is usually very high quality.

[maiden elm]

Also llama is open source if I'm not mistaken

[digital siren]

Llama could be a solution. As a business perplexity’s biggest threat and weakness is depending on apis from two companies that could shut them off at anytime. I’m sure they are thinking about this long term. Llama would be a wonderful start to diversifying their products though.

[zenith atlas]

With perplexity labs and api based on llamas I think it’s just a matter of time before we see fine tuned llama on main application

[wicked blade]

sim, tendo o mesmo problema

Here it continues the same way.

[heavy igloo]

is perplexity offline? i'm getting the same Try again later message

[night bluff]

for me

[devout geyser]

I've been getting this error a lot today as well, I assume it's a capacity issue of some kind (e.g. Anthropic's or OpenAI's side)

I make sure to copy the question before sending it in case it does it 😂

[night bluff]

it works for me

my bad

[muted linden]

Maybe we already do? In the announcement blog of the API, they claim that "Our API is already powering one of Perplexity’s core product features. Just switching a single feature from an external API to pplx-api resulted in cost savings of $0.62M/year, approximately a 4x reduction in costs. We ran A/B tests and monitored infrastructure metrics to ensure no quality degradation. Over the course of 2 weeks, we observed no statistically significant difference in the A/B test."
So I was wondering but maybe I got it wrong 😉

[zenith atlas]

We dont know exactly what this core features are, but they definetly are not the main responces of the models or copilot and Im waiting exactly for that - llama copilot and llama base model for answering

[muted linden]

Honestly, I wouldn't be surprised if the perplexity model and the copilot is silently being updated to use llama instead of GPT-3.5 FT. People noticed it getting better in the last couple of weeks 😁

[boreal pasture]

whats the difference between 'gpt-4' and 'perplexity' in the settings thing

[halcyon coral]

Perplexity is a fine-tuned GPT-3.5 model while GPT-4 is the normal GPT-4 model.

[boreal pasture]

which one is usually considered better

[halcyon coral]

GPT-4 is better than GPT-3.5, but GPT-3.5 is faster.

[boreal pasture]

ye but what was the nature of the fine tuning

[halcyon coral]

Fine-tuning is generally used to improve the quality of responses by adapting the model to the specific use case.

[boreal pasture]

im in llm research, no need to hold back

i want all the info u got

[marsh frost]

#💬│general сколько весит планета земля?

сколько весит планета земля?

[boreal pasture]

bayraktar

[halcyon coral]

Much more information was not published 🙂

[boreal pasture]

oh got it, thanks

@neat elk what did yall do for finetuning the 3.5

can you do ppo with the api?

[night bluff]

But the co-pilot mode uses Chat GPT-4, right ?

[drowsy sparrow]

Hi guys! How many tokens or characters/words can Perplexity Pro with GPT-4 remember? I know Claude can remember more than GPT, but does Perplexity overcome that?

[tame current]

you really expect that? 🤣

[pliant ridge]

If I buy pro, is there any way to skip the cloudflare interstitial? Super annoying when I want to do a quick search

[drowsy sparrow]

so far, no

[wide stratus]

Why not?? OpenAI already is on track to generate more than $1 billion in annual revenue... 😝

Hey guys, why is there such a time-consuming verification occuring everytime you visit the perplexity website?

[halcyon coral]

Normally, this does not always occur, but there are some causes that lead to the verifications occurring more frequently or always:

• Browsers that send too little data and so Cloudflare assumes that it is a bot.
• flagged IP addresses (often with VPNs)
• Certain browser settings or extensions

If you use a browser with content blockers or a VPN please deactivate them, also "privacy" settings like for example with Brave can lead to problems, furthermore it might be worth changing the browser or deactivating extensions.

@pliant ridge

[pliant ridge]

yeah but I think the sensitivity is jacked up, I almost never see these popup modals on any other websites as I have a residential ip

[halcyon coral]

Yes, but that is important because there are persistent bots that otherwise cannot be stopped. Ultimately, the pro users benefit from this, as Perplexity does not lose money pointlessly.
Have you already tried all the troubleshooting options?

[wide stratus]

If you use a browser with content blockers
you want adblock users to deactivate adblock for Perplexity? Why, does Perplexity feature ads on the website? 😳 😖

[halcyon coral]

I didn't say that, however content blockers can also cause problems with Cloudflare or generally functions on websites.

[thorn moss]

Perplexity android app not showing answers again

On website works

@halcyon coral

Anyone experiencing this at the moment of just me?

[halcyon coral]

Could you please create a #1140622008086970420 with the app version, your operating system and a public permalink to the thread? Thanks in advance! 🙏

[wide stratus]

Ok. Because having the Cloudflare protection makes it appear as if Perplexity is worried about DDoS attacks. (since Cloudflare is known to have one of the best DDoS protections)

[thorn moss]

Done but in website it shows answer, on android not

I only put my phone model but version i think latest. Please check thanks

[fallen zenith]

Hi all

I feel chatgpt gpt4 is more accurate than perplexity gpt and claude

What are your opinion

I use it for coding

[fleet flume]

If you need more text for your research, Claude can provide more words related to your topic.

If you use it for coding, GPT is the best choice.

[zenith atlas]

As a pro user I dont see a need to use perplexity gpt as the speed is not important to me, and in all other aspects gpt-4 is better. Claude have different wording and writing style then gpt4 and some people prefer, but I personally use gpt 4 for everything but file upload or high-length threads where I prefer claude for its higher context length

[cold plover]

Hey 👋 I'm making a post on using perplexity for University papers and I feel like my post is surface level. I would like to hear how you guys use perplexity. If you want to see the post I put a link to it in the #😎│cool-projects channel.

[sweet jasper]

Bots querying perplexity with long prompts can cause lodes of damage

Remember that even anon users can query, and they are technically using gpt3.5 api

[junior hemlock]

Hey folks...I'm curious if anyone knows how Perplexity references YouTube videos as sources - does it use/embed the video transcripts?

[lunar anchor]

Why did selecting code within Code Blocks break 😦

[halcyon coral]

Yes, the transcript is used.

[devout geyser]

I thought it was just me, some kind of extension I'm using. I guess not. To workaround it I have an extension that allows me to bypass those sites which don't allow right click context menu. Conveniently seems to work with this problem too

[velvet anvil]

Could you let me know which extension this is? Would be fantastic for using password managers in sites with messed up form fields that don't allow right clicks.

[devout geyser]

For Firefox, it's called "Happy Right Click" by oldginger. I'm not sure if links are allowed here so I won't paste the direct URL to the addon at addons.mozilla.org. When installed and active, it appears as a mouse icon in the address bar. Clicking on that icon will alter the current webpage's right click ability if it's hindered (or even left click as in the case of previously not being able to select certain bits of code on Perplexity's output)

[velvet anvil]

Thank you Ixel! I am on Firefox as well, so this is perfect for me.

[short gale]

https://status.openai.com/ fyi elevated API errors in case anyone is currently experiencing issues on Perplexity

[white sundial]

Cool, I read through the post and it was really good, personally I use perplexity primarily for coding, as well as reasearch when in school.

[wraith crow]

hi @thorny fog please create a thread in #1161804761247526912 with your issue

[cold plover]

Thanks! I was using Chat GPT for coding but I'm curious to see how Perplexity is. Do you have any experience with both?

[wide stratus]

and they are technically using gpt3.5 api
and that means what?

[wide stratus]

Afaik YouTube search itself even works that way for a few years now. When you search for something (on YouTube directly) that was mentioned in a video, which is neither mentioned in the video title nor in the description, it will still find it, as it is caught in the transcript. 😃

[simple folio]

it means they're using OpenAI's GPT-3.5 model, which is a bold move given that they may will have to pay for usage of the API (could be wrong), and could also get easily rate limited

[opaque lotus]

Hey, (if I'm allowed to ask here), could I use someone's referral code? About to sign up for Pro beyond the trial :)

[wraith crow]

Hi Lynicine, here is a referral code you could use https://perplexity.ai/pro?referral_code=M0YTQBNL

[opaque lotus]

Hi! Thank you! I tried using it, but it says the code is invalid. Do I have to wait until the trial period is fully over to use it?

[wraith crow]

These referral codes are only usable for new subscribers. Since you already used the trial I don't think it's possible anymore.

[opaque lotus]

Ah okay, that's unfortunate. Thank you in any case :)

[crystal perch]

Use phind for coding

[wide stratus]

Well what model should they use instead in your opinion, if not GPT 3.5? 😆

And how did you find out that it is GPT 3.5? 🙂 (just curious)

[simple folio]

honestly i don’t care, as long as it works well - i’m not big on ai models

[simple folio]

says on the website hahaha

[wide stratus]

There is no alternative... only Bing AI and Google Bard (still in testing afaik) are true GPT 3.5 competitors afaik... And afaik Microsoft doesn't offer an API (yet) for their Bing AI

[zenith atlas]

bing is powered by gpt4

There are currently openAIs GPT models, Claude 2, Pi, Bard, and different small local hosted llms (but most of them are using LLAMA or LLAMA 2 as their foundation)

[wide stratus]

I know. I just mentioned the only other few companies that offer GPT 3.5-4.0 for "free".

(unless Bard isn't using GPT? In that case sorry!)

[zenith atlas]

bard AFAIK is some frankenshtein from google deepmind

[wide stratus]

Ah damn, yep, Bard doesn't use GPT, but sth. strange called "PaLM 2" instead. Sorry!!!

In that case I'm gonna claim that Bing AI is the only competitor to Perplexity-GPT 3.5 as it uses GPT 4.0 (and is sort of free, apart from the ugly Edge-browser enforcement)

*edited

[simple folio]

i mean there are way more ai models (e.g., meta’s model, forgot what it’s called) but they’re not as sufficient

[wide stratus]

Meta? You mean Facebook basically? Is that model free to use? Or can one sign up for testing it?

[simple folio]

free; there’s a gh repo with a lot of models

i’ll have to find it

[halcyon coral]

LLaMA, you can try at https://labs.perplexity.ai

[wide stratus]

lol, small Perplexity copilot fail: "According to the search results, there is no such thing as GPT-3.5."

[wide stratus]

But by "meta" you mean the Facebook "Meta", right?

[zenith atlas]

Its called LLama, it is open sourced model which means anyone can download and tinker it, perplexity have it own labs to test this models, and there is a reddit called LocalLLama which is about using this models locally or on a private servers. But for now they are way less reliable and are not that user friendly when used locally

and yeah its developed by Meta former Facebook

[wide stratus]

LLaMA, or LLaMA 2? Apparently, LLaMA 2 already exists aswell

[zenith atlas]

yeah, there are a ton of different models that use both of them as foundation but then train them on different data-sets to perform different tasks, there are models trained to write prose, code, chat and anything else

for example on perplexity labs there is model called mistral which is, IIRC, based on llama2

[wide stratus]

Are there plans to integrate it into the Perplexity Android app? I'd love to use a more energy-efficient language model than GPT-3.5 with nearly as good results as GPT 3.5!!! 😃 Let's do sth. good for the environment and save energy! 😃

[wide stratus]

Ok. And LLama 2 can also access the internet in realtime and not only up to a certain date like GPT 3.5 used to, right?

[zenith atlas]

There are plugins that can enable internet search but they require more technical knowledge then what I have, most of the models still dont know everything up to date as they use different data sets to train them. As it is mostly open source community its hard to say something definitive about all of the thousands of models that already exist. For example I know about a model that was trained on 4chan political discussions and can replicate them, and there are models that are trained on specific languages, it all vary a lot.

[halcyon coral]

I am not aware of any such plans.

[silver mantle]

Has anyone else experienced issues with Perplexity not remembering previous instructions? Despite having the pro version, it seems to forget specific rules that were established, such as bolding certain words and providing short answers. While it initially understood these instructions, it fails to apply them consistently to subsequent sentences.

[zenith atlas]

The models have limited context length and in longer conversations will start to forget the topic or instructions that were stated in a thread, if your use case is not requiring logic or math then you may try claude-2 model as it have bigger context length than GPT4

[silver mantle]

Yeah, I notice that when I changed to the Claude 2 model, I noticed a difference. However, when I used ChatGPT, it seemed to remember the context despite being a free version and having a very long context.

[sweet jasper]

Are you using writing mode?

Remember that pulling sources actually consumes context length, as you are injecting the sources into the prompt

[cold plover]

I have a school project right now so I'll check it out

[ivory bolt]

Yes

[wicked blade]

Is the query source limit always six? As in the image.

[low warren]

I hadn't run across this one before. Looks awesome, thanks for bringing to my attention.

[granite ocean]

I have a question, can pro users use perplexity bot with gpt4 access??

I am a pro user and I want to interact it with gpt 4

[granite ocean]

It is ok for me

[tawny elbow]

Down on iOS for me

[halcyon coral]

Please close the app and try again.

[halcyon coral]

No, this is not supported.

[prime ridge]

Hey everyone, today perplexity with copilot (I have pro) seems way faster but also wildly inaccurate, and the answers are very short. Is anyone else having trouble with it? In the settings, the model is also set to 'Perplexity'

[tawny elbow]

Did that but still doesn’t work

Not sure if it’s the beta app that’s bugging

[halcyon coral]

which version are you currently using?

[halcyon coral]

Please use version 2.4.4 from the App Store

[tawny elbow]

Gotcha. Thanks

Are they no longer pushing updates through TestFlight?

Seems odd the stable version is ahead of the beta

[halcyon coral]

Yes, right now all the bug fixes are happening quickly via the App Store without the beta.
But the beta will probably be used in the future 🙂

[tawny elbow]

[granite ocean]

Is there any possibility of it being added soon?

[halcyon coral]

My bad, I read wrong, is not planned so far.

[granite ocean]

Oh ok

:(

I wish it will be added

The bot can just check if a user has the pro role and then give them access to settings where the users can choose of they want gpt4 or claude2

I know it's not as easy as I said, its a bit more complicated task. But I believe perplexity team can do it

[velvet anvil]

Out of curiosity, why would you want to run queries with GPT4 and Claude 2 on Discord instead of the website?

[tame current]

has the site crashed again?

[ionic solstice]

I am on the fence about getting pro with perplexity, it seems good for research and programming? BUT I already have claude pro and chat gpt plus, so is it any different really?

[pure zephyr]

I'm currently using GPT-4 on Raycast and am trialing Perplexity Pro. I would say that Perplexity still has the best integration with online sources I've seen so far. For $20/month, it allows access to both Claude2 and GPT-4. When using the "Writing" focus mode, you're essentially using the Raw LLM APIs.

Downsides:

• No access to ChatGPT plugins, vision/voice capabilities, or Dall-E.
• Not as tightly integrated with macOS (compared to Raycast), making it a slight barrier to usage.

[ionic solstice]

whats raycast?

[pure zephyr]

It's a spotlight replacement for macOS. Regarding OS integration, I mean stuff like being able to read my clipboard and being accessible a hotkey away. In this example, I'm trying to fix some grammar in my text, and it's just this easy. However, Perplexity's internet search and citation gathering are still the best in class I've seen so far.

[ionic solstice]

I am looking for something thats best at pdf and online searching but also academically

[pure zephyr]

In that case, perplexity fits the bill. I'm sure that Claude has the PDF support as well, and you can look into elicit.ai for academic stuff. But frankly, Perplexity's academic focus mode works just as fine.

[ionic solstice]

academic focus mode?

[ionic solstice]

thanks!

[ionic solstice]

do you know if they fine tune gpt4 and claude 2 as well

[wraith crow]

fine tuning is not available yet for gpt4

[ionic solstice]

no I mean the way they implement gpt and claude its using there API

[ionic solstice]

explainpaper.com vs perplexity I am this point lol

[wraith crow]

Awesome project @fickle pond! Feel free to share it in the #😎│cool-projects channel.

[fickle pond]

ops sorry to post here

[halcyon coral]

Hey @ripe mist, please refrain from such messages!

[granite ocean]

This is really useful for me because you don't need to open chrome everytime then

[sweet jasper]

But your queries aren't private

[halcyon coral]

You can send a DM to the bot.

[granite ocean]

That's the main reason I want it like that

I saw some other bots with that feature

And wanted perplexity bot to implement it too

[sullen jay]

Is it offline?

[halcyon coral]

https://www.perplexity.ai/ ?
Everything should work, please try again!

[wicked blade]

Can anyone answer?

[eternal kestrel]

When was the last time y’all have committed loitering?

[halcyon coral]

Hey, yes the limit is six, with Copilot 25.

[tame current]

What are you talking about, man?

[wicked blade]

Why don't you consider expanding this research context from six to a larger number?

[ionic solstice]

does perplexity pro allow you to search wikipedia and such as well?

in the focus part

does that change?

[wraith crow]

the focus part doesn’t change, you can however add site:wikipedia.org to only fetch results from that website

[ionic solstice]

ah did'nt know that, thats not something you can do on ChatGPT right?

even with browser enabled

I have chatgpt plus and claude pro

so am on the fence with the subscription

[wraith crow]

No clue if that is possible with chatgpt. Last time I subscribed there gpt4 wasnt even released.

But with perplexity you can use almost all search operators.

[wraith crow]

https://moz.com/learn/seo/search-operators on this website you can find a list of search operators to further narrow down your searches.

[ionic solstice]

thanks

[rose shale]

Found a story somewhere about someone using perplexity.ai instead of google, so i thought i'd try it out..
I went to the main page, and well.. if one of its user cases is as a replacement to google, then it might want to reset the search after a short while.. having to click the "new thread" button in top left every time i have a new question to ask seems less intuitive than with google.
How about maybe, when you do a search, you are taken to a new page, like google, with a search bar above the restults it found.. That search bar could be linked to the original question.. and by reloading the page using the reload button, it would reset back to the default empty search page.. or something like that.. idunno..

Then again, i'm not certain you're going for the "google replacement" at all because the layout kinda reminds me of chatgpt in a way.. with the "new thread" button etc etc etc.. So it might just be that i went into this with the wrong pretences.. and you never intended for this to be a "the google replacer" or whatever..

All i'm saying is.. if it's meant to be a google replacer, it needs better useability when it comes to the possibility of asking a completely new question without having to reload to the main page each time you want to ask something else. 🙂

[tame current]

bro what is api website of perplexity

[robust belfry]

Hi everyone

I noticed a decrease in perplexity performance

I used to have clean lists with specific analysis when for example, I asked for studies related to the effect of a specific element on humans, and now I get this block of text with not much

[prime ridge]

I feel the same! If you have pro, switching to GPT-4 helps (even though it does not resolve this completely).
I opened a bug report here: https://discord.com/channels/1047197230748151888/1165273167334494318 I think that you could share some examples too, if you want, in order to help them better understand the problem 🙂

[robust belfry]

Yes

Do you have frequent non sense questions from the copilot ?

For example you ask " Could you list me the types of products that are available on topic Y ?"

Copilot then asks "what types of product would you want me to search about ?"

[prime ridge]

Yes, it happens to me too

[night bluff]

Is it just me or is Perplexity's Co-pilot model less effective or not really answering questions as well as it did 1-2 weeks ago ?

The answers are less clear and readable. Before, the answers were well-structured with spaces and paragraphs, now it's just a block that is thrown out

[random shell]

@night bluff too abstract, dude. Sounds like "Раньше трава была зеленее"

@night bluff from the perspective of my own exp — as of your/team skill in PE grows -- quality of the RAG inferred grows too. The only way exist on which you can track and measure the objective quality / performance of the concrete approach/llm/prompt -- is using the tools aimed to measure the NN (LLMs especially) metrics inferred during experiments you run over. Like WANDB and so on. This is the way. Peace.

[heady herald]

I am an iOS user, and when I access the web version of Perplexity using Safari, everything works fine. However, when I use the Perplexity app, it always gets stuck in a loading loop and doesn't display any content. Can you please explain what might be causing this issue?

[carmine linden]

If you're stuck on Cloudflare check, please try this:

• go to https://labs.pplx.ai
• solve the captcha
• go back to https://perplexity.ai/

[ionic solstice]

can someone explain the perplexity API to me as a pro user please?

pls @ me if response

[halcyon coral]

You can get more information about pplx-api here:

• https://docs.perplexity.ai/
• https://blog.perplexity.ai/blog/introducing-pplx-api

[ionic solstice]

is the api free?

[night bluff]

Here we have the comparison between a response from Chat GPT-4 and Perplexity. We notice that Perplexity Co-Pilot no longer cites its sources as it used to, which gave much more structured sentences. This was clearly not the case a few weeks or even days ago. Moreover, this is reflected in the quality of the response which is no longer as precise or does not exactly answer the question.

[halcyon coral]

Currently for Pro users, yes.

[night bluff]

I don't know what happened or maybe it's just me, but the responses from the Perplexity Co-Pilot model were much better structured and more precise than using the Chat GPT4 model directly. Today, it seems to be the opposite.

[ionic solstice]

ah cool

[lean plover]

I definitely liked it better with citations with the number next to the sentence instead of all on top.

[lean plover]

I hope they give pro users an option or change it back. It really takes away from the quality of perplexity’s search and ability to verify information.

[versed nova]

When I upload a set of documents in a given thread, are the documents only visible to Perplexity within that thread, or are they used in the responses to all of the threads in my account? I assume it's only within the original thread, but need to be sure that I'm avoiding unwanted cross pollination. Thanks!

[halcyon coral]

Hey, they are only visible within the original thread 🙂

[versed nova]

Thank you!

[haughty minnow]

Just saw the pplx-70b-chat model on labs.perplexity.ai. Is there anywhere I can read more about it? Will it be available in the API?

[viral snow]

why is this ai free like bro ( im not complaining, im rather confused and happy) this and bing ai are the best things to ever be free

im so glad and grateful

🙏

[stoic seal]

For some reason my collections are now empty. Is this a bug or did I loose all my collections?

[velvet anvil]

Most likely need to clear cache and login again. Hopefully not a bug!

[stoic seal]

Ok , checking

Looks like a bug

I tried it too

[tropic radish]

saw some 500s sneak through

[carmine linden]

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

[hallow niche]

Could anyone possibly explain the difference between GPT-4 and Claude 2 as model options? I understand the differences in each standalone model but it’s confusing how those differences work with perplexity’s API to answer questions. Any response is much appreciated

[halcyon coral]

Hey @hallow niche,
GPT-4 often gives more accurate answers and can usually solve logical problems better, but Claude 2 is popular for its writing ability and has a larger context.

[hallow niche]

Ok gotcha. Perplexity’s API basically does all the searches and than the differences in the models becomes more apparent for the answer?

[muted kelp]

Is Perplexity not able to read and compare two files? I'm trying to have it compare two .json files, but it keeps saying ot doesn't have the second file even though it has both of them. I even tried converting them to text files, but it's still not looking at both.

Tried this with Perplexity and GPT-4, but no dice.

[wide stratus]

Why is llama 2-13b still available on https://labs.perplexity.ai if there already is the more advanced llama 2-70b to chose from? Is 2-13b the very first iteration of llama 2 that was published?

[night bluff]

Claude-2 is very good for analyzing PDFs, especially research articles. You can ask it any question like "summarize this" or "describe Figure 1 to me". This model is mind-blowing.

[late oracle]

Those aren't version numbers and each model is trained differently. the 13b = 13 billion widgets, and 70b = 70 billion widgets. neither one is more advanced than the other because their training is always being improved.

[ionic solstice]

is there a way to make perplexity search within a date range?

I am trying to search through data between 1700 to late 1800's

pls @ me if response

[wraith crow]

You can actually compare 2 files with perplexity. https://www.perplexity.ai/search/What-are-the-Ubg6FETsSqu0RSUOcj3B0w?s=c
Maybe your json file it too large?

[muted kelp]

Hm maybe. One of them is around 5 MB.

[wraith crow]

Also make sure your focus is on writing mode.

[wraith crow]

You could also try to use Claude 2 model for a bigger context size if you're a pro member, that might help.

[wide stratus]

"widgets"? Hm. So what does the difference in "widgets" mean regarding its' responses?

[idle hull]

Is ChatGPT-4 no longer serving its purpose? I feel like I'm asking questions to the ChatBot in Perplexity and it's not serving the purpose I'm looking for.

[grand seal]

I had to stop using GPT-4 the past 4-5 days. Only using Claude now, because the quality of GPT-4’s responses has, like you’ve said, seemed to reach Perplexity/3.5 level. Copilot exacerbates it. Waiting to see if it gets better

[tame current]

I use it on PC. Personally it seems fine, but maybe your experience is different.

[muted kelp]

You know I didn't even think about that!

[grand seal]

Mac browser here for the past few months. Perhaps it’s only copilot getting in the way but I just stopped using GPT altogether because I haven’t had time to experiment/share and figure it out.

[tame current]

Fair point.

[proud patio]

Same.

[tame current]

@proud patio @grand seal MaxAI.me can scan Youtube videos now on the free version, it's pretty dang cool. Have you guys indulged in similar apps like this? https://www.summarize.tech/ is another one, but I think MaxAI may do it better from experience.

[sullen jay]

Same... Chatgpt keeps forgetting things now and my previous input...

[digital siren]

The temperature for the writing focus is way too low. TBH perplexity needs a complete overhaul of its writing/chat focus.

[pliant ridge]

congrats on the raise :)

great product, glad to see some solid recognition

https://www.theinformation.com/articles/ivp-leads-investment-in-ai-search-startup-perplexity-at-500-million-valuation

approx 13k people paying for ppl pro

interesting

[rugged apex]

Hello, I keep getting stuck on the verification page ~ is there a solution?

now I'm stuck at https://labs.pplx.ai/

[halcyon coral]

Hey @rugged apex!
Which browser and which extensions do you use? Do you use a VPN?

[rugged apex]

My edge browser went in.... But my Google Chrome still doesn't work

[halcyon coral]

Any extensions?

[rugged apex]

no

[halcyon coral]

Please delete cookies, website data,... for the website. Maybe it will work then.

[rugged apex]

Also, I would like to ask, if I want to set perplexity's copilot as the default search engine, what characters should I enter?

That's what I just tried but it didn't work, causing me to double verify

[halcyon coral]

So just to be sure, the problem is only in Chrome and you don't use any extension in Chrome (content blocker, dark reader,...)?
Are you using the latest version of Chrome?
Can you please also send a screenshot of the Perplexity page?
Does it work in an incognito tab?

[woven tundra]

Guys I have a question, can Perplexity access to a given link and do stuff like summarizing,... I tried with Youtube link and the result is super inconsistent

[rugged apex]

It's okay! It's uBlock Origin.

[calm mirage]

@latent wharf

[carmine linden]

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

[maiden elm]

Yes that's the main purpose of Perplexity, accessing links and doing stuff with the data (you tell it what to do). For videos it will read the transcript. If the transcript is automatic then it may not be good hence Perplexity may give inconsistent results. These are my opinions I'm not 100% sure

[austere hill]

Huge congrats Perplexity team on your latest funding. Keep up the great work!

[sharp flame]

You need better vpn add me

[sharp flame]

Phind is better imo

[obsidian ridge]

Which combination of options is the optimal one if one wants the best experience of conversing with someone with lots of follow-up questions (as opposed to perplexitys superior search experience)?

[maiden elm]

I didn't knew Phind, thanks for sharing. It seems to me that in the long run each AI service is going to specialize in one aspect. I have the feeling that Phind is focusing in programming. Perplexity seems to be general for now, but it will have to choose at some point. I say that because you said that one was better, I think they're just different.

[maiden elm]

IMO it depends on what are your criteria for the best. Also when you say conversing with someone, you want the AI to be an intermediary? Check this out https://discord.com/channels/1047197230748151888/1163535658602733679

[tame current]

can you share the full article with me? there is a paywall

[obsidian ridge]

I meant as in a direct discussion partner who behaves as close to a human as possible and who does not overly try and highlight a single answer - and who remembers history as much as possible. So basically «conversation mode». But will look into that other thing you linked, looks interesting 🙏

[tame current]

Any recommendation for an app (iOS) that can turn sketches/drawings into images? My son is interested in converting his sketches into AI-generated images but based on his drawings.

[west mica]

Did anyone here ever have a ChatGPT plus subscription? Did you cancel it in favour of Perplexity Pro?

Also, is the GPT-4 in Perplexity the same as ChatGPT 4?

[ancient elk]

hi, i'm looking for a way to limit perplexity's search results to the last 12 months? I tried all the obvious ways...

[wraith crow]

Have you tried the after search operator?

For example: question here after:2022/10/26

[ancient elk]

I tried it but I don't think that this actually does something in that direction

it doesn't seem to affect the search results reliably like it does in google

[wraith crow]

It should do the same as in google. Here is an example.
Without date: https://www.perplexity.ai/search/Reviews-about-the-qnJiXaK3ShKZeqgWITN8WQ?s=c
With date: https://www.perplexity.ai/search/Reviews-about-the-daW2HfbeSuSB9gzlPa50gg?s=c

As you can see it will use sources that are more up to date. Please make sure the date is in the following order YYYY/MM/DD.

[ancient elk]

is there any documentation for this operator and others like it?

[jaunty delta]

will i get access to gpt via perplexity api?

[obsidian ridge]

I did. More value for money as you also get Claude, GPT, and Perplexity proprietary algo.

And faster with much better search results if replacing google

[obsidian ridge]

I am wondering the same. As in: if i choose GPT as the algoritm within settings, does perplexity then behave identical to ChatGPT?

[zenith atlas]

I was using both subscriptions for a few months. For now I actually decided to stay with ChatGPT only, because I find that gpt4 there can do better in writing texts than gpt4 in writing mode + because internet search is not the most important feature for me. I’ll probably subscribe to pro again when the file upload will stop summarising files before feeding it to the model itself.

[zenith atlas]

No, there is definitely different pre-prompt and there probably a different version of gpt4 itself because open ai love to delay API releases, for example for now we don’t have dalle 3 and vision.

[obsidian ridge]

Allright, so we’re saying there is «feature lag» in Perplexity in regards to GPT4 and possibly some prompt difference - but that responses and conversation dynamic for the most part will be experienced as the same, right? Sorry for silly questions, I just need to know so I can make an informed decision on subscribing

[zenith atlas]

But there are also things in which perplexity is ahead of chatgpt - context length. IIRC chatgpt have only 8k or 16k while perplexity have 32K context length activated when needed

[obsidian ridge]

I take it Vision will make it to perplexity soon

[zenith atlas]

Yeah, probably in beginning of November

[median pine]

Hello sir, i have problem when i want to generate python code

Before is like this

now like this

there is no color in import codes and the others

[wraith crow]

You can find a list of search operators and what they do on this website: https://ahrefs.com/blog/google-advanced-search-operators/

[ancient elk]

so it more trial an error, there are no official perplexity docs for this? site: and filetype: seem to work too

[wicked blade]

Good morning. Where can I make suggestions and offer feature ideas or changes to Perplexity?

Right here?

I would like to suggest including the "Rewrite" option in the Android application, which already exists in the Web version.

In order to allow the AI ​​to rewrite the response with GPT-4, Claude-2 or Copilot.

In addition to the other operations also available in the Web version, such as editing the message, etc.

[thorn moss]

i suggested this one before bro. maybe its hard to update that feature to android cause i suggested it like 3 times and until now it was not updated. maybe it will take time

[halcyon coral]

Hey, you can suggest features in #💭│ideas, but the web features will be added to the apps over time!

[hallow niche]

Is it possible to copy/export an output WITHOUT the citations?

[halcyon coral]

You can write after the answer in a follow-up question: "give the answer without citations"

[slow python]

hello

[heavy oriole]

Hey, when I use Claude for the summaries of pages that are in English web (my AI Profile is in Spanish) its response is English and should be in Spanish. It does not happen with GPT-4.

[digital siren]

Does anyone know what the pplx-70b-chat-v.01 model on labs is? I haven’t seen this announced anywhere. Are there any metrics on it?

[thorn moss]

sorry for the question but is perplexity website or desktop version down? not functioning. only android app and in kiwi browser

[rugged ocean]

I have the same issue, sometimes, it answers in english even though I selected another language. Really weird...

[tribal flint]

no, it is illegal to do such a thing

[maiden elm]

What is?

[ionic rock]

what is it??

[sharp flame]

@maiden elm definitely use case dependent. Perplexity is good I just need to try with api. Phind is slow but I like the result it produces.

[maiden elm]

We have red crosses for asking. Is asking wrong?

[halcyon coral]

I would like it if we could have a topic-related exchange and not talk about illegal activities 🙂

[maiden elm]

Fine 😊 i was asking in order to know what not to do, maybe I'm doing the illegal stuff that was mentioned ? 🤔

Unless it was obviously illegal in which case I'm obviously not doing it

[halcyon coral]

You have done nothing wrong. You can also contact us at any time with questions about the community rules 😉

[prisma crypt]

Hey all, new to Perplexity. Can anyone point me to Perplexities security docs? I may need to put some customer data thought it and need to understand the privacy and security before we implement.

[halcyon coral]

https://blog.perplexity.ai/legal/privacy-policy

[prisma crypt]

Thanks, I did see that. But I am looking for more indepth information. Need to understand if my interactions are private or do they train LLMs. Are they encrypted? Does Perplexities any SOC documentation?

[halcyon coral]

Regarding the training, you can opt out of having your data used for training.

[prisma crypt]

Thanks! I actually did see that and forgot I enabled it.

[tribal flint]

Why Can I not change the token size for GPT-4?

I want to increase it

[halcyon coral]

Are you talking about the input length?
You can simply copy large texts and paste them into the input field, these will then be used as paste.txt

[tribal flint]

the output sorry

say if I want a detailed answer

[halcyon coral]

Does the AI reply break off at any point or do you generally get short replies?

[tribal flint]

no just wanted to know if there's a way to change them

you can do it in open AI playground

[halcyon coral]

I see, there is no setting for this, but you can tell the AI that you want a detailed answer.

[tribal flint]

thanks

[ionic solstice]

is there a way to make Perplexity only search in the UK?

[graceful grove]

Hi. I would like to ask Is Preplexity GPT-4 the same as OpenAI GPT-4 or Preplexity GPT4 is trained tuned seperatly?

[fluid mauve]

Im guessing Perplexity is pretty light on the standard security policies. Its a pretty new compant so I am not sure if they even have a SOC report. Mind you there must be something as they have raised a bunch of Capital investments.

[prisma crypt]

That's a grat point! I created a support ticket. Hopfully they will be able to give a more definative answer. Thasnks

[fluid mauve]

No worries. Having been in a company CyberSecurity team and going through an IPO we were generating a lot of policies as we went.

[prisma crypt]

All this LLM is new to a lot of clients, including me, I just want to make sure we are covering ourselves and getting the right insurance policies should something happen.

[fluid mauve]

Yeah. thats is a big ask for insurance companies. There's not a lot of data to make good risk-based decisions. Just from copyright and IP issues, it's a nightmare. Try being in a Financial market. How do you explain what the model is doing, and how it is doing it? What was the training data? is it biased,, and how is it biased? the list goes on.

[digital siren]

It makes me feel good that I’m spending my money with Perplexity, knowing that their founder supports the right of people to have access to AI.

[ionic rock]

tanking about money... any plans to have a promo again, currently not able to afford it 🥺

[digital siren]

Here’s $10 your first month: https://perplexity.ai/pro?referral_code=8JCPU08Y

[digital siren]

The free version of perplexity is honestly very good too. The perplexity model is a fine tuned version of gpt-3.5 that can match gpt-4 at certain tasks. It’s great because it’s much faster than gpt-4.

[digital siren]

Forefront.ai has Claude models with internet connection for free. Otherwise gpt-4 for free internet browsing is best done with bing

[digital siren]

Also you.com is $7/mo for students if that applies to you

[frosty galleon]

Hi guys, has anyone managed to use autogen with the perplexy API?

[last mason]

hey folks. are both these chrome extensions by perplexity?
https://chrome.google.com/webstore/detail/perplexity-ai-search/bnaffjbjpgiagpondjlnneblepbdchol
https://chrome.google.com/webstore/detail/perplexity-ai-companion/hlgbcneanomplepojfcnclggenpcoldo

[halcyon coral]

Yes, both are from Perplexity

[royal trout]

Based on the available search results, there is no version of GPT-4 currently available. Therefore, it is not possible to compare Vicuna-13B and GPT-4. However, preliminary evaluation using GPT-4 as a judge shows that Vicuna-13B achieves more than 90% quality of OpenAI ChatGPT and Google Bard while outperforming other models like LLaMA and Stanford Alpaca in more than 90% of cases

I am deletiing this

The thing is drunk af

Based on the available search results, MemGPT is a system that helps large language models (LLMs) handle longer conversations by cleverly managing different memory tiers
5
6
. MemGPT uses function calling to manage the flow of data between its main context and external context, allowing it to work with a virtually unlimited context
6
. MemGPT addresses the limitation of fixed-length context in language models, which is a significant advantage when dealing with long conversations or analyzing extensive documents
5
6
. In contrast, GPT-4 is a newer and more advanced version of GPT-3.5, and it is better able to comprehend context and recognize patterns
1
6
. Therefore, it is difficult to compare MemGPT and GPT-4 directly, as they have different strengths and capabilities. However, MemGPT's context expansion capabilities make it a promising tool for revolutionizing conversational AI
6
.

then in the next que smfh

[midnight needle]

bros got a website as his username 🫣

[midnight needle]

wait so the free version utilises "gpt-4"?

[sweet jasper]

Nah

[midnight needle]

what do you get for the pro version besides the unlimited file upload ?

[sweet jasper]

It uses a finetuned version, some llms train based on gpt4 responses

[midnight needle]

right. but that's on the free version

[royal trout]

the big deal here is memgpt

[midnight needle]

tf is memgpt

[sweet jasper]

600 gpt4 api calls

[midnight needle]

[sweet jasper]

And claude 2

[royal trout]

there is 2 gpt4s btw

[sweet jasper]

Per day

[royal trout]

thhere is a free gpt4

[midnight needle]

Thank you @sweet jasper

[royal trout]

buut is 13b vs 170b

and also you can just reverse proxy the other gpt4 so yes they are both frfee

[sweet jasper]

A finetuned model trained on gpt4 responses is different from gpt4

You can gain some performance from training your smaller llms on state of the art llm responses, but that doesn't mean it's reasoning becomes equal or stronger than state of the art for less parameters

[royal trout]

vicuna is 90% as good as 3.5

and better thban bard

but the thing is bard is like michael vic

[halcyon coral]

AI may give wrong answers, you can report such answers by clicking the flag icon below the answer.
Thanks in advance!

[sweet jasper]

Orca, llama2 and Falcon currently surpass vicuna I believe

As open llmw

?

This is just based on hugging face leaderboard

You can check it yourself

https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard

[halcyon coral]

Please watch your language, thank you!

[royal trout]

Im not watching my launguage

[sweet jasper]

But you do have a point with memgpt that it could potentially help llms

[zenith atlas]

with pro you have access to GPT 4 and Claude 2 models and not only gpt 3.5FT, although for now there is beta of the perplexity API which is free for pro users (but it probably will change on release). GPT 4, Claude 2 and copilot are sharing 600 uses per day, and as a little extra there are some channels in this discord only for pro users with frequent discussions.

[zenith atlas]

In recent month I tried to understand which open source llms are better in different tasks and as of now there is no simple answer to it, most of the llm community on reddit are hyping about orca and mistral, but there are also other models that people say better at writing or RP such as nous-hermes and lots of other models.

Overall for now there are only a few models that are close enough to gpt 3.5, but they are way slower and demand a pc with at least 64gb of RAM

[carmine linden]

Hey @upbeat quiver! For AI art you can use the ⁠#🎨│dream channel and generate it with the /dream command

[ionic solstice]

#💬│general message just following up this question

does anyone know?

thanks

[tribal flint]

שָׁלוֹם

[noble stirrup]

Hmmmm

[prime spruce]

https://tenor.com/view/shall-we-play-a-game-typing-type-can-we-play-a-game-do-you-want-to-play-gif-17643701

[digital siren]

Right, but having access to a fine tuned version of gpt-3.5 optimized to perform better than the original version of 3.5 for free is wonderful.

[tame current]

will prplxty have chat with pdf?

[halcyon coral]

You can upload a file to Perplexity and ask questions regarding the content of the file.

[tame current]

ye bu it isn't the best

[ionic solstice]

just following this up as still struggling with this: #💬│general message thanks

sorry to keep spamming but the question is bugging me

[untold hazel]

Is there any promo codes available for discount?

[halcyon coral]

There is no setting to limit the search results to a country and to my knowledge no working search operator for it.

[halcyon coral]

With a refferal link you get $10 discount on your first month, otherwise there are no other promo actions at the moment.

[long cypress]

is there a model i can use in prplxty api to generate accurate links? tried all 4 (llama-2-70b-chat, llama-2-13b-chat, codellama-34b-instruct, and mistral-7b-instruct) but they gave fake links

example prompt: find the best video or educational content that teaches (give only 1 and provide the link): Introduction to Generative AI

it gives this link - https://www.youtube.com/watch?v=9XQ_5RN_28M (which does not exist)

[halcyon coral]

#💻│coding message

[ionic solstice]

thank you, usually I put U.K. and sometimes it works

but I was wondering if something more specific

[ionic solstice]

ask perplexity 😂 it will find

[dark solstice]

perplexity sucks at math..

i put in math equations and sometimes it just gives me the answer and it does not show me any steps.

[stark lichen]

i am testing afforai for pdf stuff it's pretty good. i hope preplexity can improve to be on the same level.

[whole vault]

Hi. In perplexity.ai is there a keyboard shortcut to send the message. Enter key creates a newline.
Tried Cmd+Enter but it doesn't seem to work.

[tame current]

how is affora?

[noble stirrup]

I hope perplexity will be able to supports images or videos one day

Which is the best ai right now

For programming

[wraith crow]

Make sure to select the Wolfram Alpha mode

[noble stirrup]

Wolfram alphamode you mean?

[wraith crow]

aah yes my autocorrect did something weird i see

[noble stirrup]

Anyways can yoou answer my question please lol lol

[wraith crow]

For programming I advice you using tools that are integrated into the IDE you're using like copilot. But for me Perplexity does the job very well. Be sure to use writing mode. It's very good for programming questions

[halcyon coral]

Are you talking about something like GPT-V? #💭│ideas message

[noble stirrup]

It actually does, it researches the web and does not give incorrect information

[noble stirrup]

Is that even a thing?

But one thing perplexity is really bad at is designing

[wraith crow]

Writing mode shouldn't use web sources

[noble stirrup]

Even if you give a clear image it will not create a replica of the front end desgin

[noble stirrup]

Excluding programming

Any questions i ask it, it’s perfect

Anyways

Perplexity cant front end design as well as GPT-4 Can

[wraith crow]

could you give an example of what you're trying to achieve with perplexity? and what your prompt is

[noble stirrup]

Example: “Create a calculator similar to apple’s calculator IOS using html, css, JavaScript” Sorry i’ not that good at prompts

Thats just an example

[dark solstice]

but what is the difference?

[lament imp]

Selecting "Wolfram Alpha" will utilize resources on Wolfram Alpha rather than results it finds on the website. This is significant because Wolfram Alpha is a computations website, meaning it is engineered to solve mathematical equations and/or display solutions to mathematical problems.

TLDR: It gives more accurate solutions for math problems

[dark solstice]

interesting ok

[lament imp]

"Perplexity" is GPT-3.5 that is enhanced by the Perplexity Team. So that's why it says OpenAI: It was developed by them but enhanced/changed by Perplexity, thus giving it a different name

[steady meteor]

why did perplexity remove the search terms part of the result?

i really liked this feature

https://discord.com/channels/1047197230748151888/1168324904358248468

[dark solstice]

l;,;\

[digital siren]

Does anyone the context window for the gpt-3.5 fine tuned perplexity model?

[ionic rock]

thanks 😊

[gusty jolt]

yes, i have... #💗│sharing message

[sweet jasper]

I think this is the wrong discord for those things 🙂

[dreamy gale]

ios perplexity ai cannot be accessed how to solve it

[vernal moon]

Hi, can you tell a little more about the problem you’re experiencing? Thanks

[royal trout]

I just deleted off my computer I can't stand looking at it anymore it makes me so mad I really liked this for a while but it's really actually literally useless at this point in time I don't know how technology seems to be devolving But almost every bot is worse than when it was when it came out and perplexity was the only one that was holding off for a while that I knew of and now you guys have just let it go

[midnight needle]

#👋│introductions message

two sides to every tale haha

[tame current]

What is the difference between Perplexity and Perplexity Chat? Chat has focus on natural convesation?

[lunar anchor]

I mean there is this:
https://futurism.com/the-byte/stanford-chatgpt-getting-dumber
Paper about it
https://arxiv.org/pdf/2307.09009.pdf

[maiden elm]

Perplexity is the whole thing, Perplexity chat (gpt) is Perplexity using the gpt model. Tell me if i understood your question correctly.

Unless you mean writing focus which is Perplexity without web search.

[maiden elm]

Perplexity does not offer chatbots, even if you get the illusion that you're discussing with somebody, you're actually generating text automatically... Other AI online services use AI to intermediate the chat between two persons.

[noble stirrup]

So that means its just a fine tuned ai?

Not a trained one

???

[maiden elm]

I don't know that

[noble stirrup]

If yes then it’s not worth buying pro

[maiden elm]

Interesting, can you explain why?

I hope it's a trained one

Even so, i think it did a good job

[noble stirrup]

Since it’s just chatgpt you can finetune your own ai

No need to waste money

[steady mason]

What is the sequence of operations taking place in the background when I prompt Perplexity with v. without Copilot? IE, I'm trying to figure out what Copilot is actually doing differently to what Perplexity is already doing--searching sources, reviewing those documents, then providing a response. Sometime Copilot will ask a follow-up question, which is helpful sometimes, but other times it doesn't; yet I will still get different results.

[noble stirrup]

Exactly

[steady mason]

You mean "Writing"?

[noble stirrup]

Pretty sure it just briefly expands the explanation

[steady mason]

How you mean?

[noble stirrup]

Not even sure my self bro 💀💀💀

Which ai model is the best for code writing

Claude, GPT-4, Perplexity

[maiden elm]

Ask Perplexity 😂. Imo it's gpt but i don't do programming

[noble stirrup]

Waiaiaiaiwa

So

GPT-4 is just a fine tuned GPT-3.5 eight?

Same with claude

Honestly kinda ass

Also

@maiden elm

Why is bing so restricted?

It literally cant do shit

[maiden elm]

It is crap

It's Microsoft

Old business model

[noble stirrup]

No, it does surf the internet and provide actual correct information

But its mad restricted

You cant ask it to do shit or it will just respond: “Lets move to another conversation”

Goofy

[maiden elm]

Maybe afraid of bad ussges, it is legally bonded after all

[noble stirrup]

There was a bing jailbreak

I used it

Created malware

But got patched shortly after

[maiden elm]

Interesting

[noble stirrup]

Can i ask a question

[maiden elm]

I have no idea how a patch is done

[noble stirrup]

Whats the different between copilot and no copilot turned on

Has the same answers

Just copilot costs money

[maiden elm]

Aaah yes you ask that before nobody answer you

You want to know the answer?

[noble stirrup]

Yes….

[maiden elm]

Pilot will ask you more details that will help perplexity find more accurate information

[noble stirrup]

Fair enough

[maiden elm]

It is not perfect and it must be used in certain contexts

[noble stirrup]

It should be obvious

[maiden elm]

I'm also kinda new though but that's my pro experience so far

[noble stirrup]

Honestly

Ive been using it for lua

(I already know how to script lua)

It’s medium

Its not that bad but its not that good

It can script pretty decent stuff

Unlike chatgpt it fixes errors first try

[maiden elm]

Lua rings a bell

Like js?

My knowledge is nothing compared to the door that Perplexity opens. The ability to access updated information is being underestimated, that's the main feature of Perplexity.

GitHub copilote should be "better" for your needs imo, have you tried it?

[noble stirrup]

Search ho lua

[noble stirrup]

I haven’t, i will soon

Im just busy with school and muay thai trainings

I literally have 0 time

( i know you didnt ask but still)

Anyways

Thanks for the suggestions

Im trying to find the craziest smartest ai

But, remember

Ai isnt magic

[maiden elm]

No problem, good luck with school

[noble stirrup]

It’s easy

Dont worry lom

Lol

[maiden elm]

😀

[noble stirrup]

Wait

Is it night or day

For you?

[maiden elm]

22:13 in Brussels

[noble stirrup]

Ohhhh you’re in europe

Good night then

[maiden elm]

Yes the capital 👍🏼

[noble stirrup]

Yep

I know lol

[maiden elm]

Where are you?

[noble stirrup]

Im in albania

Balkan

To be exact kosovo

[maiden elm]

Wow

[noble stirrup]

Small country

But still

[maiden elm]

Nice man

[noble stirrup]

Thanks bro

[maiden elm]

Very international place this is

[noble stirrup]

Looool yea

[maiden elm]

😊👍🏼

[noble stirrup]

Damn

Anyways

Good night, i hope to see you again later.

[maiden elm]

Same bro, take care

[kindred folio]

hi

[still bison]

a

[regal frigate]

May I ask if there is a UI that supports uploading files or sending screenshots, and allows customization of the interface and key?

[midnight needle]

https://discord.com/channels/1047197230748151888/1047202784090538054

[tame current]

i love using perplexity

just cuz

the ui is flippin sick

[granite quartz]

I'm loving perplexity. I have one major issue though - is anyone aware of any way to ensure perplexity uses British sites? Currently everything is returning in dollars. I need to get results in pounds. Sites always seem to default to the US version which is very frustrating for me as someone based in England

[fathom tulip]

perplexity only analyse readables documents or i can drop any kind of pdf file ?

hi btw haha sorry

[midnight needle]

hey 👋

there was a question from another user about this the other day - #⚡│ask-community message

I don't know if he received a satisfactory answer though - when you say readable documents do you mean other formats besides .PDF?

[fathom tulip]

i mean there are some pdf that can't be read, like it's not text, it's images

And i wanted to know if perplexity is using an ocr method or if it's only reading readable files such as pdf (text), word, .txt...

[midnight needle]

gotcha, yeah I'd like more clarification on that too

[halcyon coral]

You can write in your AI profile in which measurement system you want to have the values.

[granite quartz]

Thanks yeah this is useful, but unfortunately it doesn't work great. For instance I saw a price on a site for $99 on the US site and it then converted that to something like £73, but the GB version of the site wasn't doing a direct conversion rate so it was actually £79.
I even asked the model to use the GB version of the site, but it wouldn't

[halcyon coral]

Have you ever tried to limit the results to UK websites (site:*.uk)?

[granite quartz]

In fact I realised I have actually already added this to my profile:

I am from the UK, so only refer to UK prices and UK services.

I don't think it helped much.

Great idea also, but a lot of sites use /en-GB in the URL, or just detect automatically and don't have anything extra in the URL, so it's too simplified I think. Also it will filter out .com sites that are UK only. It would be great to allow to switch the location of the crawler. I don't believe that would be too hard

[south mantle]

Hello
Perplexity shows some sites that require a subscription as the source. However, when I click on the source, I can't see the entire article because I don't have a subscription. How can perplexity see it?

[halcyon bronze]

Does anyone know which LLM the free tier of Perplexity uses?

Does it say anywhere online by any chance? and what does ft mean?

[south mantle]

.

fine tuned

[forest summit]

Is it possible to get LLAMA2 with better than a 4k context on perplexity?... min 8k 32k ideally...

[halcyon coral]

There will be models with larger context. https://docs.perplexity.ai/docs/feature-roadmap

[noble stirrup]

Wait

How did perplexity get gpt-4 and claude 2

[wraith crow]

Pro members can choose the model on their settings page

[noble stirrup]

Yes but

Did perplexity pay openai to get their models into perplexity?

[wraith crow]

using their api

[noble stirrup]

Damn ok

It doesnt give the responsrs like gpt-4 or claude 2 tho

I doubt its using their apis

And one thing it doesnt do is process images

[halcyon coral]

The API's from OAI and Antrophic are used, currently no images can be analyzed, because OAI has not yet enabled this feature via the API, also image generation is not possible.

[tame current]

I was wondering if anyone can helpo

I am using perplexity to do some research on our products, I want to gather this information and then write a blog on it. It occurs to me that perplexity isn't great for this

is there anything I can use or get better results from perplexity in this regard?

[forest summit]

they either stole it or someone got a big favor ;)😀

[copper lark]

A user interface comment that might help Perplexity developers, or other users: I misread and mistakenly chose the wrong of two adjacent buttons to delete ALL ALL ALL of my threads everywhere, when I only meant to delete the top (most recent, uncategorized) thread. GONE GONE GONE are months of effort and dozens of long running project threads full of code and details that I was NOT NOT NOT finished with. Now I'm sad. 😦 but, otherwise, Perplexity ROCKS for my use case.

[jade valley]

Twitter says there is a pplx-70b-chat, but I can't find it.

[halcyon coral]

Please contact [e-mail], maybe the data can still be retrieved.

[noble stirrup]

Is there any other models except perplexity, gpt-4 or claude?

[halcyon coral]

Please have a look at https://labs.perplexity.ai/

[noble stirrup]

Well since perplexity has its own model why not enable image analysis

It’s not using a api is it?

[halcyon coral]

#💬│general message

[noble stirrup]

Why does bing have image analysis then?

It does use gpt-4’s api

[halcyon coral]

Microsoft has invested a lot in OAI, including funding a supercomputer. Accordingly, they have access to many more functions.

[noble stirrup]

Ohhh

Thank you

[digital siren]

Someone on Reddit said the free version of perplexity uses llama 2. I’m almost certain it uses gpt-3.5 fine tuned?

[sweet jasper]

Its gpt3.5 finetuned

[dark solstice]

i have a questions what does mistral-7b-instruct and pplx-70b-chat do?

just curious what can i do with those

[lethal bramble]

Hey, I wish to chat to a Perplexity team about hiring, how can I reach? 😁

[fluid hawk]

What model is behind the Copilot mode?

It seems like GPT-4, but then they have the GPT-4 model as a separate choice...

[waxen imp]

it's powered by OpenAI's Fine-Tuned GPT-3.5

[fluid hawk]

it seems a lot smarter than even fine tuned 3.5

[thorn moss]

Yeah cause they improved the FT 3.5 like last monday

[noble stirrup]

It is

[sweet jasper]

So I got pro right now, If I use copilot, will I use gpt4 plus gpt3.5 for fine tuning the prompt?

[halcyon coral]

Correct. Or optionally Claude if you set it in your preferences.

[thorn moss]

Yeah but with the process 3.5 FT do it first then the end part is GPT 4 will translate and write the result right?

[halcyon coral]

Yes the routing process is handled by FT GPT-3.5, the final response is then GPT-4/Claude.

[thorn moss]

I can say the improvement is really noticeable. Thank you team perplexity.

[thorn moss]

Got it. Thanks. 😊

[dark solstice]

Why can I delete questions on the perplexity app?

And why does the app look totally different from the website?

[noble stirrup]

Uhhhh

I know for a fact my card is not declined

When are you guys gonna add paypal

[neat elk]

we released a new model yesterday, so it should be much better. over next couple of weeks copilot will improve significantly, stay tuned

[thorn moss]

wow. Thats a great news. I am so hyped on this coming improvements of copilot. I cant wait right now. rawwrrrr

[noble stirrup]

Dude it’s gonna be so good

[thorn moss]

yeah it is so good and more good coming. My mind is going blank. yameruuuu

[noble stirrup]

Looool

Do you have the pro version?

[thorn moss]

yeah bro

[noble stirrup]

Ohhh

Mine just ran out

😭😭😭

[thorn moss]

copilot answering really good now

[noble stirrup]

It is

[thorn moss]

call your bank now dude

say you are mad.

[noble stirrup]

Looool nah its ok send me the thing again

Discount

Ill try to buy it again

There should be a 10 euros discount

[thorn moss]

discount? i dont know that..

[noble stirrup]

Oh

I bought the pro agajn

Again

[thorn moss]

its life bro. we need it!

[fluid hawk]

sounds like you are talking about drugs lol

[thorn moss]

No bro. I am not talking about drugs bro. Canabis will be legal soon in Germany bro!

[fluid hawk]

ik, just kidding

[full vector]

How does Perplexity's model work? Like does it function differently than Bard or GPT+Bing on trying to determine what the "facts" are?

[fluid hawk]

wondering that too

[neat elk]

yes, it is different, we have our own ranking and other ways to do a better job than bard and bingchat

[craggy prairie]

^ facts

anyone else having issues changing the app icon in the new update?

[dusty void]

Hi hello. I wonder if other iphone user has some problems with Perplexity app? Because when I want to search for a result it will be blank. Is it because of new update?

[full vector]

Are you able to expand on what that ranking entails? Is it some way of cross-fact checking a piece of information across other sources? Is there like a confidence score for information as it's gathered?

[midnight vessel]

I LOVE THIS APP

[dark solstice]

Any updates this month?

[supple gust]

hey I'm wondering what model is used in the free web version by default? I'm seeing in the FAQ and about pages that GPT-4, llama, bard are all used in some way shape or form, whether in the pro version, api and so on but cant find which is used in the free web version with copilot off

[maiden elm]

My understanding is that Perplexity is a model in his own based mainly on gpt3.5... but someone should confirm this...

[quasi grove]

lol use an AI to find out about another AI.

https://chat.openai.com/share/d03b8358-11fa-445f-8042-57019b643ab2

That’s what I was able to furnish.

[muted pine]

hi i have a question, Perplixity pro with Perplexity LLM it's better than Perplexity basic ?

Cause i try GPT4 and the tokken limitation it's a bit to much ..

[supple gust]

the link led to a 404

[neat elk]

no, it is the same

[muted pine]

Ok thanks, and claude 2 it's 100K ?

[neat elk]

yes, correct

[muted pine]

Thanks boss ! 👍

[muted pine]

BTW, do you have any plans to add Stable Diffusion or another Image LLM?

[neat elk]

yes, GPT-V and Dalle-3 is coming soon, stay tuned

[muted pine]

SALLLL !!!! Good work !

[sweet jasper]

Are anonymous threads use to train perplexity

[thorn moss]

When Openai release their GPT vision API, does it requires a lot of time like days for you guys to apply it to perplexity or you can do it on the same day of release?

[calm marten]

any date for GPT-V

[halcyon coral]

This is more of a question for the OpenAI server 😉

[calm marten]

You are right about that one lol!

[calm marten]

how are you mate?

[halcyon coral]

I'm fine 🙂 , how about you?

[calm marten]

Great man ;P but just found out that there is no date to GPT V yet

What is your task as a Community volunteer? Just helping people answer questions or xD

[halcyon coral]

Moderation and support in case of questions/problems.

[calm marten]

They don't have a AI for that or Xd

[halcyon coral]

Would probably be very unreliable 🤔

[calm marten]

Depending on the human factor in it tho

Auto moderation as a filter is always good to have idk if this server uses the Discord Moderation or a custom build one

[halcyon coral]

We (fortunately) don't have many problems with rude users here and if there are any, someone is usually there to fix the problem. 🙂

[calm marten]

Thats lovely, yea it depends on the community got a server of 81k my own but moderation and rude people come with certain age groups of people

Ai is more for 18+ that brings more professional people with it that upkeep the rules

[sweet jasper]

Most ai users tend to be teenagers and those in tech if anything

[fallen zenith]

Hi do you guys feel that openai api or gpt 4 on preplexity is not that better as compared to chatgpt

I use it for mostly coding

[thorn moss]

for me it gives better response than chatgpt, i dont do coding that is why cannot say. For me search engine by means of perplexity.ai is the best

try using writing mode

[neat elk]

early next week

[viral sapphire]

Is there a way to share my Perplexity account with my wife like I can with ChatSonic?

That's the one concern I have before signing up for pro.

[fallen zenith]

Yes it is awesome at doing searching but please if anyone using it for coding give your experience

[rough haven]

Perplexity is loading super slow for me, am I the only one experiencing this

[thorn moss]

you mean next week like monday or tuesday? My brain gonna explode if this is confirmed.

[calm marten]

What will it offer Denis or is it unknown?

[thorn moss]

It will be mind blowing if gpt vision can be used with internet. Oh my

[tribal flint]

i love perplexity

[thorn moss]

i am hyped for this question.

[dark solstice]

I have a question..

Is Academic mode more detailed and accurate then internet mode on perplexity

[halcyon coral]

An internet search is also used, but the sources are limited to scholarly articles and academic publications.The sources are usually more trustworthy and more scientific.

[dark solstice]

SO the sources are more trustworthy and scientific for the academic one?

[halcyon coral]

Yes, this does not mean that the normal internet search is not trustworthy, but the Academic focus only takes scholarly articles and academic publications, which are usually better elaborated.

[dark solstice]

Is it also more detailed on Academic?

[halcyon coral]

The level of detail does not actually depend on the focus, but on your prompt.
It makes a difference whether you write:
Write a text about artificial intelligence
or
Write a text about artificial intelligence, break it down into X, Y, and Z. Use technical language and explain in detail.

[boreal valley]

oeeeeeeee

[dark solstice]

Alright

[dark solstice]

True.

[digital siren]

I’m just curious what are people using the Perplexity API for? I built my own ChatBot, but I don’t know that it has any additional use beyond Perplexity labs. I’d love to hear some creative ways that people are using it.

[gusty jolt]

i have hooked it into a discord bot, and into autogen

[pliant ridge]

Why does perplexity have no chat like context when using successive queries

It's super annoying, almost defeats the point of a chat like ui

if i wanna make a new query I'll create one

[tardy wolf]

Hey guys, I can see that POE.AI have enabled a creator program, where people can create custom bots and earn money with them, have you thought of adding this to perplexity as well? I like perplexity more than poe, so I would love if I can just create content here and capitalize out of it

[vagrant dock]

uh

so

how do I make perplexity ignore certain words in its searches

I asked: "do holidays and weekends count when porting my number from a mobile operator to another" (in portuguese) and instead of finding search results to number ports between mobile operators It searched for salary portability

[sweet jasper]

Seems like gpt4 api got nerfed a few days ago

Hopefully after openai announcement it goes back to normal