💬│general

[sage lily]

Oh ye. I don't have the app. Just using web. That sucks it's been deactivated. No announcement about it either.

[tame current]

what does the perplexity team think of this?

any potential? the downfall of perplex!=ß?!

im geniunely curious

since google is the main thing

[fossil sentinel]

Google's SGE sucks

[short gale]

Yeah and that’s when it actually works which from my experience the past few months is almost never 😅

[warped marsh]

That is not your screenshot right?

[warped marsh]

If you use it yourself, you won't ask this question

[stark pelican]

any admins here?

billing issue

[grand parcel]

✨

Tap this emoji and when its discription pop up then hold this emoji to see magic.

[pine lodge]

Best to send an email to [e-mail] for questions re: billing

[tame current]

yeah iam from eu

[warped marsh]

VPN works btw

[warm sigil]

Experienced error multiple times today

Anyone has same issue?

[tame current]

are you using Wikipedia focus ?

if so, it has been disabled

[warm sigil]

I don't even know what that is .

I'm using Android app

[tame current]

oh nvm what I said, are you using a vpn

[warm sigil]

Uh huh, guess that's an issue w/ my proton VPN as I'm connected to a public wifi, will switch a different server and see

[tame current]

Hello there, is copilot only using GPT4?

[short creek]

I would think so, yes

[tame current]

So when using pro and enabling "gpt4" in settings

• using copilot

"600 uses" = 600 gpt4 uses?

[wraith crow]

yes the 600 uses are gpt-4 uses. so that are copilot uses and normal uses if gpt-4 is enabled in the pro settings

[tame current]

Oh okay amazing I just paid for pro

[short creek]

GPT-4 uses one Copilot token. It doesn't matter if you have Copilot enabled or not. If you have both enabled they will still consume one. Though right now GPT-3 consumes them too. (Due to a bug with a recent update we are assuming)

[wraith crow]

Be sure to rejoin the server with the invite link in the settings or dm Alex Romanov to get access to the pro channels

[tame current]

Yes thanks

Ok so if i disable copilot

and enable gpt4

[short creek]

You will still consume Copilot tokens.

[tame current]

what's the limit of gpt4 uses?

ohhh

600

[short creek]

600 a day

[tame current]

it's still amazing

thx

[short creek]

But GPT-3 will also use them till they fix it. But it's mostly impossible to reach 600 since they dinamically replenish

So you don't have to worry haha

[tame current]

i only use gpt4

dw

[wraith crow]

Oh you're right did you already make a bug report in #1140622008086970420 ?

[short creek]

Not yet. I told IceLavaMan, he said they have been informed now.

So I didn't know if I should post it as well.

[halcyon coral]

the devs know about it 😉

[tame current]

is perplexity using byte pair encoding?

for subword tokenization

[chilly forge]

Just a small doubt here

should not be going down with a regular search regardless of the model used (GPT 3.5 or GPT 4)
so we basically get unlimited regular searches with GPT4 as well ? I didnt knew this

[short creek]

No, that was an error on their end

I'll delete the message so it doesn't get more people confused.

[chilly forge]

yaa lmao

I got worried for a while

[rough haven]

Am I the only one that really dislikes the new theme? I even liked the old logo better 🥺

[chilly forge]

Change is hard

[tame current]

its not that deep

[rough haven]

Yeah I know, but that’s not it…
I love change.
I just really dislike it

[short creek]

I like the new theme to be honest

[rough haven]

Really?

[short creek]

Yeah

[chilly forge]

I was just so used to the old theme I am not that fond of new one tbh

[short creek]

I mean yes. That also happened to me haha, but overall I think it's a nice looking app.

[rough haven]

I felt so official when I used perplexity with old theming. Now I feel like a child…

[short creek]

I came from ChatGPT so...

[halcyon coral]

I hope for an option to be able to choose the theme

[rough haven]

Ikr! Me too

[chilly forge]

Well theres always the tamperMonkey script way if you're missing the old one

Spezi created a post or put it somewhere ig

[short creek]

Check this #💗│sharing message

[rough haven]

Hmm, yeah but that’s so annoying to deal with uk?

[short creek]

Yeah I know. But was just sharing his concept

[rough haven]

Then I rather have it as is. It would be a big feature for them to implement some kind of personal theming

[short creek]

Indeed.

I'm finding some features missing on the Desktop version unless you are using Responsive too.

Like you can't delete all threads with one click unless you are on Android / Responsive

That would also be nice to have as Desktop users

[rough haven]

Whats responsive?

[short creek]

When you access the website as if you were on a table / phone

But yeah some further features like theming would be much appreciated I think

[rigid solar]

does perplexity have enterprise/business rates (ideally no cap copilot)?

seems like it would be cool to use to replace bing/google for companies

[arctic spade]

forgot they were dropping the school discount this week

is the yearly worth buying?

[short creek]

oof I suscribed to the pro tier yesterday haha

[short creek]

I'd say that unless you really need Code Interpreter from ChatGPT for a very specific reason, it's definitely worh it

[uneven seal]

I have both. ChatGPT is much better at iterating on something in a chat than Perplexity is. For programming this is especially valuable.
ChatGPT to write code - Perplexity to learn about apis and parse documentation / forum posts.

Also Perplexity leans on summarizing results it found, while chatgpt draws knowledge from internal intuition. Sometimes that intuition is better than the found knowledge, but often it isn't - since perplexity can cite sources.

[short creek]

I guess it depends on the case scenario. I've been using the "writing" focus and the results are pretty much the same for me. I still have both.

But I agree that if you are only using the "all" focus then it's as you said.

Right now from what I've been testing the best way to use it is uploading a file using the "all" focus so it doesn't browse the internet unless you ask something using "copilot". Then you get the best from both worlds.

[uneven seal]

oh man I had no idea the focuses existed lol

[short creek]

Then you are about to blow your mind.

[junior lynx]

I signed up for the Pro version, does the "Pro Role" aspect take time to reflect since I joined the server prior to signing up for it.

[uneven seal]

lmao I can tell

[short creek]

I was exactly like you yesterday when I subbed. Then started doing what I said up there and man... It's crazy good.

[short creek]

I don't seem to have a pro badge or anything and I'm suscribed too

Wait no, I do have a badge

[junior lynx]

ah, interesting...have to wait I suppose

[short creek]

Have you tried joining from your Perplexity Profile?

[junior lynx]

Yes..redirects to this server but beforehand shows an invalid link

[short creek]

Should have something like that. Maybe if you join through the link in there.

[uneven seal]

You know what perplexity needs - an API

I'd love to be able to write a script that automates the results of Perplexity

[short creek]

@uneven seal Let me know if you try the little trick from previous messages

[uneven seal]

Yeah I may soon

[junior lynx]

@short creek - I'm going to try something to see if it updates and if it does will let you know

[uneven seal]

Tho one thing thats nice with chatgpt is that I can turn off message history and prevent any data collection

[short creek]

Sure 😄

[uneven seal]

but once I finish planning on my honeymoon with chatgpt I may cancel the sub

[short creek]

You can do the same here without disabling history

[uneven seal]

oh man thats great

Already have it off

but I think they still capture some data

idk

[short creek]

The data that they collect is mentioned in their FAQ. https://perplexity-ai.notion.site/Perplexity-FAQs-0a9141bdd1b94c76b955e892f73451ff

"We may collect IP address, device type, unique device identification numbers, browser-type, broad geographic location (e.g. country or city-level location), and other technical information. We may also collect information about how your device has interacted with our website, including the pages accessed and links clicked."

[junior lynx]

Tried to sign out and then back in via DC's web-browser with the invite URL but will wait it out some time for now

[wraith crow]

You'll have to leave the server and join with the invite link in the pro section on the settings page. Or you could send Alex Romanov a DM to get access to the channels

[junior lynx]

that was my next thought...I just boosted it too 😭

[wraith crow]

You can also send a dm to Alex Romanov he'll give you the role if he's active

[junior lynx]

will do again 🙂

Thanks @signal hamlet @short creek @wraith crow - Much appreciated (Apologies for 3 mentions in a single message).

[short creek]

Hey you got your badge!

[rigid solar]

I created an iOS shortcut so you can just say “hey siri perplexity search” and it will complete the search with perplexity instead of Siri/Google

https://www.icloud.com/shortcuts/d8fee8a9e7b446988b68fa2df7617c15

Workflow:

[vast pasture]

Hello there was a discount announcement for students

I think I have missed it it was in the announcements about 2 hours ago

[frank rivet]

Hi! it's back live again 🎉

[frank rivet]

fixed

[fossil sentinel]

🙆‍♂️

[tame current]

Could I get the discount? I am an existing annual* subscriber but I have a valid student email address.

[tame current]

any updates after the prompt leakage? asking for a friend

Why was the school discount announcement reposted

[fossil sentinel]

what are you talking about

[tame current]

Also, where's the Pro Discord?

[frank rivet]

re-worded for more precise clarity on the policies of the discount

[wraith crow]

If you have an ongoing pro subscription you should rejoin the server with the invite link in the pro section on the settings page or message Alex Romanov to gain access to the pro channels

[tame current]

@signal hamlet could I interest you in a short parlay over this?

Welp clearly the invite link doesn't work for me anymore.

are you using the bing api?

[gray dove]

Perplexity wants to position itself as a prime research tool? I asked Perplexity, "What are the latest heppenings in Niger".

It tells me, "As of August 20, 2023, the latest happenings in Niger are not clear from the search results. The most recent news articles are from May and December 2023, and the other search results do not provide any specific information on recent events in Niger. It is possible that there have been no major developments in Niger since December 2023, or that news sources have not reported on them."

Really?

1. The happening are very clear and Perplexity actually referenced news articles in the response stating the happenings.
2. You can't have recent news from December 2023 in August of 2023.

[warped marsh]

This is with copilot

[gray dove]

If I hadn't provided evidence, your post would solve everything.

[jagged ivy]

maybe you confused Niger with Nigeria?

or i confused

For me it worked, free perplexity, no copilot

[gray dove]

Look at the screenshots. See a difference in country names? (Hint: the answer is no.) 😊

[jagged ivy]

It's my bad, the revolution was in Niger

[gray dove]

And my friend, we have not reached December 2023

[fossil sentinel]

That's why you should use Copilot for more in-depth/complex tasks.

If you expect GPT-3 technology to be perfect, you can only be disappointed tbh

Your prompting could also be better

In non-Copilot mode, I took your same prompt and added "Include dates." Still not a good prompt, and the response isn't perfect, but you can already see how much better the response is for that tiny addition in the prompt...

[scenic grail]

"whats going on in niger"

worked perfectly for me

lmao

[gray dove]

So you're telling me that

1. Perplexity news can't be trusted unless it's from copilot
2. Foreseeing future news is just something Perplexity users must live with unless using copilot.

Essentially, pay or it's useless?

I agree somewhat with the gpt-3.5 pot shot. It's pretty frightening though that we're looking to be a premier research tool and these are the sort of things people might get. Who then would pay for such a service?

Then prompt was fine.

My friend. Someone else prompted worst than me and got the correct answer...lol. Let's not diverge from reality.

worse not worst since some may nit pick on grammar. lol..

I congratulate you on getting a correct response. It doesn't solve the problem(s) though. See my initial post.

Glad perplexity is working on fixing things. Just wish my screenshot proved I could trust it.

[scenic grail]

wtf

[fossil sentinel]

Wtf

[scenic grail]

how did you do that

[tame current]

prolly bad internet

[scenic grail]

o

I think it's a similar situation to using any language model, really. like using chatgpt, I don't just take its word all the time, if I want to learn something from it I take the time to see if its bullshittin me or not. i think the same applies here

[gray dove]

Ok at least perplexity should be aware of this. That's why I posted. I'll resign and leave this to the experts

[fossil sentinel]

It's not that bad prompts can't get the answer you want, but better prompting allows for more consistency and accuracy. Also, with the nature of LLMs, you shouldn't just blindly trust it due to the risk of hallucination. But what Perplexity does amazingly is its ability to provide relevant sources. It does it better than other Internet-access AI I've seen so far

I think people are on the wrong track when expecting LLMs currently to be perfectly reliable

[gray dove]

Akuma. Perplexity telling me that there is no news on Niger since Dec 2023, in August 2023... Has nothing to do with prompting.

[fossil sentinel]

Yes, that's what I mean by you are focusing on the wrong thing

[gray dove]

Akuma, that's absurd. Explain how focusing on Perplexity throwing in imaginary dates is right?

[fossil sentinel]

Perplexity is based on GPT-3/4

It's not a Perplexity issue

[scenic grail]

you're quite literally complaining about a service that takes advantages of another service that isn't even close to perfect.

[fossil sentinel]

Complain to OpenAI if that is your concern

[scenic grail]

do you think this is chatgpt 5?

[gray dove]

The just use the other service?

[scenic grail]

um

chatgpt doesnt have access to the web unless you pay

so

if you want?

who knows if its more or less reliable than perplexity

[halcyon coral]

LLMs sometimes make mistakes, GPT-4 less than GPT-3.5.
However, good prompts can reduce the error rate.

[fossil sentinel]

It utilizes GPT technology to do what they do better than others

[gray dove]

I post a perplexity error and now I'm trolling

[fossil sentinel]

But you can't expect the fundamental limitations to be resolved by Perplexity themselves

[scenic grail]

this guy must be trolling

[halcyon coral]

?

[fossil sentinel]

Hallucinations can happen anytime, but do you think bad prompts don't make LLMs hallucinate more often?

[scenic grail]

keeps going on about it even though it's already been explained

lol

[gray dove]

Perplexity shot out bogus future dates.

Bye

[fossil sentinel]

[scenic grail]

troll

bye

[halcyon coral]

Inaccurate prompts can definitely lead to more hallucinations.

[fossil sentinel]

Exactly

[gray dove]

Deal with the problem guys and gals. Paying should not be the option to correct information from perplexity

Bye

[fossil sentinel]

👋

I feel like I wasted my time for no reason

I take that back, I did waste my time

[scenic grail]

[pine lodge]

Wow I step away for a moment and things happened.

That said if hallucinations do happen and you see it in your response, please always also include a link to the query and we can look into it (even better, make a post in #1140622008086970420 )

[tame current]

Are you using bing api?

[scarlet zephyr]

Is the pro subscription worth it? How does it compare to using Bing Chat?

[pine lodge]

we use a variety of sources to get more accurate results!

[tame current]

Such as?

[pine lodge]

can't spill too much info unfortunately since the space we're operating in is crowded

[tame current]

I get it, wish you could be more transparent :)

[fossil sentinel]

Well personally I've found bing chat to be quite lacking compared to OpenAI's GPT-4

[scarlet zephyr]

Bing chat uses GPT-4 in creative mode

[fossil sentinel]

I'm aware, it's just less capable

[scarlet zephyr]

So how so? Are you comparing to what you get from Perplexity?

[fossil sentinel]

Or how should I say it, I've experienced a lot more inaccuracies

[scarlet zephyr]

I am specifically interested in a comparison with Perplexity Copilot for instance

[fossil sentinel]

Yes I've tried a few times comparing bing to openai's gpt-4 and perplexity

and I've found Perplexity's copilot to be better than openai

[scarlet zephyr]

Thanks

[hardy minnow]

Hey does perplexity pro use apis like gpt 4?

[short creek]

It's GPT-4 based when using Copilot / Enabling GPT-4 (Pro users only)

[hardy minnow]

Right i meant like for API's like wolfram and such

[short creek]

I would assume they all use GPT-4 when enabled. It's just focus swapping.

But if you want me to ask anything to it we can test

[neat canopy]

Hello

[short creek]

Hello

[cyan mulch]

Any current post grad student here who have used GPT4 for their studies and research but have now switched to perplexity ?? No regrets?

[short creek]

I graduated 2 months ago. (Web Development) Been using ChatGPT Plus for 5 months so far. Switched to Perplexity two days ago. Definitely no regrets on my end.

[gleaming drum]

Hello

[slow kite]

is there any coupon for low purchasing power countries?

[grizzled canyon]

SCHOOL25

[slow kite]

is this for everyone? it looks like only for academic person..

[tame current]

#📢│announcements message

students, educators, and new subscribers

I guess it is

[stoic crag]

I'm trying to purchase the Perplexity Pro plan, but when I click on "Get Started", it doesn't redirect me to the Stripe purchase page; instead, it just reloads the current page.

[livid fable]

I am on the monthly Pro subscription which I got using the anniversary promo voucher last month and went to upgrade to annual, added the new promo code (SCHOOL25) and got the message: This code is valid, but your order doesn't qualify. Am I not eligible?

[devout geyser]

New subscribers only, not existing or previous ones I imagine. As is often the case with most businesses and promos

[livid fable]

Wish it could be otherwise - then they get more annual subscribers 🙂

[arctic dove]

Hi guys, does anyone know the context size of Perplexity pro? Is it the same as GPT4? M thinking of buying the Pro plan but still weighing between that and Claude (currently using Perplexity pro for web search and using Claude for reading long text)

[halcyon coral]

It´s 32k

[silver laurel]

Should i buy the pro? What are your thoughts?

[wraith crow]

The update soon will also add claude 2 and will support 100k tokens

[ebon rose]

i just want to say thank you devs for developing Perplexity it helped a lot in getting resources and research fast appreciate it

[hardy spade]

Is there an official who can help about pro membership purchase?

[chilly forge]

You can DM anyone among this ones

[hardy spade]

thanks

[modern crown]

When we are using GPT-4 searches only does that count against our copilot uses because it goes down every time I do a search.

maybe I have been confused, but I thought that was included of being a PRO member. I thought the use case stuff was only for copilot.

I thought gpt-4 searches were unlimited

[wraith crow]

As a pro subscriber you have 600 gpt-4 uses a day. That includes copilot and normal gpt-4 (if 4gpt model is enabled in settings)

[modern crown]

Okay that makes sense

Do you know when it specifically resets? Is it like midnight or something?

[wraith crow]

Each use reset after 24 hours. So If you ask a question at 7 am it will reset that 1 use the next day at 7am.

[modern crown]

Ahhhhh

That makes sense! Thank you!

Do they ever think they will increase this?

[wraith crow]

I don't have an answer to that, unfortunately. But don't worry, you will probably never reach the 600 daily limit. I use this tool very often and have never reached the limit. They put this limit mainly to prevent abuse.

[short creek]

I know it may feel like you are going to run out of them at some point but it's literally impossible @modern crown hahah

I felt like that when I first suscribed and I'm a heavy user

[modern crown]

Thank you so much!!

I find that perplexity is EXTREMELY accurate haha

I use it with other tools but it is now my primary tool and it does a good job coding too /)

My guess is they are going to add new features down the road for more supported file types?

[short creek]

Indeed

[short creek]

And yes I think there is some talking about that in the FAQ.

[wraith crow]

Yes the team is also working on supporting more files to upload. Also new features are being added soon, like claude 2.

[short creek]

Spezi do you know about it in depth?

Claude 2 I mean

How is it going to be implemented and so on I mean

Is it going to be on labs like LLAMA?

Or directly implemented on Perplexity

[halcyon coral]

In Perplexity (file upload)

[short creek]

So when it gets implemented it will be able to handle up to 100k tokens?

[chilly forge]

Damn I think Claude Integration is the most awaited feature till now

[chilly forge]

It is Suppose to

[wraith crow]

Yes, I'm guessing it will be some kind of toggle users can enable #👑│pro-lounge message

[short creek]

That's really nice

Does it also mean the model would be able to give longer answers?

[halcyon coral]

It was not said whether the output size will also increase.

[chilly forge]

I think the 100k token size is combined right ? (Input + Output)

[short creek]

I guess we'll have to test when it comes out 🙂

[chilly forge]

Still unclear ig

[halcyon coral]

Yeah, it´s the context size

[short creek]

That's crazy though

Right now it's 32k Input + Output?

[halcyon coral]

Yes, in file upload it´s 16k (input context) #👑│pro-lounge message

[modern crown]

@halcyon coral you are awesome 🙂

I am absolutely a huge fan of Perplexity

[tame current]

wtf isnt the max length 4096 tokens?

[thorn moss]

yeah i thought also 4k input. was it changed to 32k? is this legit? can someone please clarify?

[halcyon coral]

According to Denis Yarats, the input context is up to 16k tokens. #👑│pro-lounge message

[thorn moss]

then what was 32k?

[chilly forge]

Its the combined (Input+Output) Tokens

[thorn moss]

is this for real? im having a brain damage!!!!

[short creek]

hahaha

[thorn moss]

perplexity is totally beating openai's butt!

nah. according to alex its 2k input. 😂

[narrow silo]

@signal hamlet teach Perplexity to pin chosen posts

those weekly ones

I see this isnt done, and it would be good to have list of those

maybe even in separate channel

¯_(ツ)_/¯

[tame current]

more like 16k characters maybe?

[chilly forge]

2k input tokens max ?

[halcyon coral]

Text question

[halcyon coral]

Here you have to ask Denis...

[halcyon coral]

What is the input token limit for search prompts and AI profile?
Search - 2000 tokens | ~8000 characters
AI profile - 1500 characters

[chilly forge]

Cool ! Thanks 🙂

[cunning sky]

when perplexity vscode extension?

[halcyon coral]

There is no fixed date, but an extension for vscode has already been requested by other users.

[dark mortar]

I'm in pro, is there a way to make 'copilot' the default mode?

[short creek]

If your last prompt had "Copilot" checked

Next time you open the tab it will remain checked

[dark mortar]

Not happening for me, click 'copilot', enter a query, open another tab, 'copilot' is not checked

[short creek]

Finish your query. Close that tab. Or just close the whole browser.

Open again. It should remain checked.

[rigid badge]

Hi, is the server down ?

It counted one copilot answer each time I've tried again and it showed this error message

[fossil sentinel]

Seems like writing mode works tho

so a search issue?

[jovial crater]

Hello, yea everything works fine when I disable Copilot

[short creek]

Works fine for me

Wait no

Copilot is broken right now

Works on android though

[short creek]

It's working now.

[sage lily]

What the heck is up with the way it formats it.

Also this.

🤔

[short creek]

Lol that's kinda weird

[sage lily]

It is indeed.

[short creek]

Mind sharing the prompt so I can try to replicate?

Using wolfram I assume

[sage lily]

Yep. Hold up. Nope. Writing mode/focus.

[short creek]

Damn

[sage lily]

The minimal number of kings needed to occupy or attack every square on an n×n chessboard is given by a formula. Work it out.

That's the prompt I used.

[short creek]

Were you using GPT-4 or GPT-3

[sage lily]

Gpt 4/co pilot.

[short creek]

https://www.perplexity.ai/search/c0c761bf-3314-4f43-9427-b93f7fe396f9?s=c (GPT-4 no Copilot)

Seems like it went nuts in your conversation for some reason

[sage lily]

Lol.

[short creek]

xD

Usually I don't get those formulas unless I'm using Wolfram

[sage lily]

I see I see.

[short creek]

Otherwise it's usually working with plain text

[sage lily]

I don't use Wolfram much.

Mostly just writing.

[short creek]

Me neither. I'm mostly using "All" or "Writing" for the most part.

[sage lily]

What kind of questions do you like giving it. I like asking logic and mathematical problems.

[short creek]

For me it's usually searching new tendencies on web development when I use the search focus or coding related questions when I use writing. Besides that I tend to be using random questions to see the differences on it's behaviour when I use GPT-3.5 or GPT-4

[sage lily]

Nice

[short creek]

But it's mostly for coding purposes

I've came from ChatGPT + Bing for searches but I'm finding myself using it as Google Search too haha

[sage lily]

Haha.

[short creek]

It's quite cool tbh

[sage lily]

It really is.

I also have a question. The models that perplexity uses. Do you know if the devs tinker them at all? Since they're Openai's.

Is that the right word? Tinker.

[short creek]

You mean tweak them?

[sage lily]

Yes. Tweak. On their end.

[short creek]

I would say they have been slightly tweaked for better. I'm getting better results here than what I used to get in ChatGPT.

[halcyon coral]

The devs have created a system prompt, which is a kind of optimisation.
But it was not yet possible to optimise advanced models.
#💗│sharing message

[short creek]

The biggest difference is that here you can play around with "focus" to "fine tune" GPT for specific tasks

[short creek]

Exactly

[fossil sentinel]

Happens occasionally

Perplexity's latex formatting is not as consistent in my experience

[sage lily]

Oh. I was like what are you talking about lmao

[short creek]

Haha

I think he was referring to your thread up there where you got the weird bug

[fossil sentinel]

Lol my screen was scrolled up earlier in the convo, didn't realize

[sage lily]

Lmao

[short creek]

rofl

[fossil sentinel]

I believe they use GPT and their own in-house models

[short creek]

I think they are not using their own models yet

[fossil sentinel]

I remember them mentioning it while back

[sage lily]

You know what would be mint. Being able to upload files that contain images and ask questions about the images.

[halcyon coral]

As far as I know, they don't have their own model yet.

[sage lily]

But can't upload files that contain images on perplexity which kinda sucks tbh.

[short creek]

https://youtu.be/jksGQhMtXjo?t=879

It should open at 14:39

They talk about that topic specifically

[short creek]

Bing does that thing, you can upload images there and ask questions about it. It's not just OCR recognition. It's crazy to be honest.

[sage lily]

How many questions can you ask bing before you have to wait for who knows how long?

[short creek]

"So, right now the plan is to build our models to work alongside OpenAI's. I expect OpenAI to have the best model for at least 2 to 3 more years. Nobody knows the future after that".

That's contradictory though

haha

[fossil sentinel]

They mentioned as far back as May that they use their own in-house alongside gpt

[short creek]

Yeah this video is from last month so it's weird haha

18th July

Then he said something like "... We are happy to work alongside their API's"

So maybe you are right

Anyway thanks for sharing the info @fossil sentinel

[sage lily]

I'm a dumb a** but what's OCR recognition?

[short creek]

It's Optical Character Recognition

[fossil sentinel]

My experience with Bing's OCR recognition is very very poor to say the least

[short creek]

It's not OCR

[fossil sentinel]

Bard can though so that's pretty cool

Since it's basically google lens lol

[short creek]

OCR (Optical Character Recognition) is a technology that converts different types of documents, such as scanned paper documents, PDF files, or images captured by a digital camera, into editable and searchable data.

[short creek]

I tested both and had the opposite results tbh

[sage lily]

Oh ye

[fossil sentinel]

It's the only use I have for Bard

[short creek]

So this is kind of what old Google Lens have been doing this whole time.

[fossil sentinel]

well it is google lens

[short creek]

And now as Akuma pointed out they are incorporating AI alongside so it can also recognize objects and many other things.

[fossil sentinel]

they just integrated into bard

[short creek]

Yeah I know but only in the English website

The spanish one is kinda behind

[fossil sentinel]

Rip

[short creek]

I can access the english url though so it's not that bad haha

I tend to go there to check on new features

[fossil sentinel]

I'm just waiting for Google to release Gemini

[short creek]

Multimodal AI?

[fossil sentinel]

Ye

[short creek]

Don't know much about Gemini though

[fossil sentinel]

I think they plan to utilize their AlphaGo method

It seems to be hinted that it'll be much more capable than GPT-4 from OpenAI

But it remains to be seen

[short creek]

I mean so far I consider Bard the worst out of all the tools that are out here

So I really need to see a big improvement from Google before I trust any of what they say

[fossil sentinel]

I mean

DeepMind is part of the Gemini project

[short creek]

Wasn't it also a part of Bard?

[fossil sentinel]

Idk

[chilly forge]

Bard would diminish once Gemini is out

[chilly forge]

No

[fossil sentinel]

It'd be very surprising if it wasn't bad considering how it's a rushed product

[short creek]

True

[fossil sentinel]

https://www.wired.com/story/google-deepmind-demis-hassabis-chatgpt/#:~:text=But Hassabis says his team will combine that,as planning or the ability to solve problems.

capable of planning and solving problems 👀

[tame current]

Does anyone have suggestsions for LLM that cater more towards excel/google sheets?

[short creek]

If you really care about Excel, Word, Powerpoint integration you can always try Microsoft 365

It has various AI integrations in the whole Office suite

[tame current]

Is there any way to add the query to Perplexity URL? Trying to query a search from Alfred App

[sacred trellis]

I use this as a custom search engine in Arc (Chrome based browser)
https://www.perplexity.ai/search?q=%s&copilot=false&focus=internet

[narrow silo]

Can't wait for AI to plan problems

[sacred trellis]

💀

[limpid bough]

Does anyone know how to use llama 2 on Perplexity? Appreciate

[wraith crow]

You can use llama here: https://labs.perplexity.ai/

[floral merlin]

guys, any tips on how to make the ai profile have consistent impact while letting the model use internet?

[chilly forge]

^Same Question indeed

Currently writing mode provides a comprehensive integration with AI profile

[halcyon coral]

Works best with copilot.

[floral merlin]

I dont see any difference with or without copilot

it still struggles to adhere to my profile

[tame current]

Thank you this works! https://www.perplexity.ai/search?q={query}&copilot=false&focus=internet

Is the writing mode a paid feature ?

[floral merlin]

while with writing mode on -> work 10/10 times

[modern crown]

I like the writing focus, but what if I want to change the focus on follow-up replies is that a feature that might be coming in the future?

[halcyon coral]

Requested often, so probably.

[modern crown]

Good to know 🙂

[frank rivet]

definitely in the list of most requested features 🙂

[narrow silo]

Will I understand GPT-4 if I haven’t seen the first 3 yet?

[short creek]

Hahaha

[sage lily]

🤔👀

I don't know how but it changed focus on the follow up answer/generation.

[sage lily]

I'm shocked as well.

[chilly forge]

Is it somehow related to https://discord.com/channels/1047197230748151888/1142182845205463051 ?

[sage lily]

Well kinda but tbh it looked like her second generation was in writing because it didn't say it was searching for sources.

That face when people want to switch between focuses manually in between each generation but Perplexity does it automatically but randomly lmao.

[floral merlin]

guys, I'm really frustrated on how the web results negatively impact the model response's quality

it struggles to explain concepts by spitting verbose responses but none of them is actually explaining the questions

I can use the writing mode and it will get to the exact point without any hallucinations

but ofc if that information is before the cut off date

I'm recently working on newer features/apis, so I am forced to use web search 🤦🏻

instead of getting precise answer, the time I spent double-checking those responses is taking more than if I just manually google them

I think mainly because the model keep being fed with unnecessary sources while the follow-up question really don't require any of them 😩

sorry for these rants, maybe I'm just so desperate for this change https://discord.com/channels/1047197230748151888/1135139914921545788 🥲

[chilly forge]

Dropped a suggestion in DM. Check it out

[idle dome]

When uploading a file, how many pages can the model read?

[devout sun]

Can I set my preferred sources of information?

[tame current]

you can use site: operator in perplexity

[chilly forge]

or you could try mentioning about them in the AI Profile ( but AI profile isnt properly incoorporated with WebSearch )

[stone current]

I hope the AI’s perplexity can decide on its own the details needed by the user unless the user requires other details. This could really speed up the response.

[last remnant]

Hi, do you delete threads after a certain amount of time? I can't find a past thread which I always would search for to reference back to.

[halcyon coral]

No, the threads will not be deleted.

[dusk vessel]

Does anyone know ho Perplexity can use data from Statista, Gratner (paying sites) - do they only take data from freemium reports? If so they must be webscraping it - which is not allowed under their rules.

[short gale]

Yeah it can be particularly bad with clickbait youtube videos & blog posts, the kind that are horribly verbose & generally unhelpful yet perfectly SEO’d to appear near/at the top of search results

Try modifying your AI profile to include what kind of sources you prefer for particular topics/searches; you could also experiment with the opposite approach of telling it what to avoid but research has shown LLMs tend to provide better outputs when told what they should do rather than a list of things it shouldn’t

[vast pasture]

Guys is there an API key I just bought pro

[short creek]

I don't think they have their own API.

https://perplexity-ai.notion.site/Perplexity-FAQs-0a9141bdd1b94c76b955e892f73451ff

[wind musk]

i absolutely LOVE perplextiy ai, the best free chat bot out there

[tropic crest]

how have yall found perplexity compared to chatgpt for coding?

and writing too?

[wind musk]

well, actually same but it could provide more up-to-date stuff since it has access to the internet

[tropic crest]

yeah me too, because I'm debating on switching from gpt+ to perplexity pro

[halcyon coral]

I think Perplexity is better, because of the access to the internet Perplexity has unlimited knowledge, while ChatGPT only has the normal model with limited knowledge.

[tropic crest]

yeah, until I found perplexity I felt that GPT4 was the best at coding tasks (especially with interpreter) which is why I never swtiched to anything else

[short creek]

I have been coding with ChatGPT Plus for over 5 months, but I don't really miss it. Perplexity, on the other hand, seems way more versatile to me. With Perplexity, you can search the internet just like you do with Bing Chat, but your experience is personalized and searches are way faster. When it comes to writing, you have access to pretty much the same features, but with way more GPT-4 usage throughout the day.

[tropic crest]

agreed

especially since gpt+ caps at 50/3hrs

[short creek]

Plus there are some tricks you can do yo access the internet only on certain queries and use "writing" focus during the rest of the conversation

[short creek]

Yeah, ChatGPT would only do 400 prompts a day compared to 600 from Perplexity. However, with ChatGPT, there were times when you needed to prompt more than 50 in a short period, and you were not able to do it. In those cases, you would have to switch to Bing for a while if you still wanted to use GPT-4.

[tropic crest]

also underrated feature of perplexity is that its able to do much better file synthesis

chatgpt is so bad at it

[short creek]

Well I disagree with that.

Code Interpreter is unbeatable atm. You can give it anything you want.

Perplexity shines way more when it's just PDFs

[tropic crest]

hmm true

Yeah I was referring to mainly summarizing pdfs

[short creek]

Depends on the use case scenario

[tropic crest]

I havent used code interpreter for other stuff for file upload

[short creek]

Yeah then I can agree with you

Also I found GPT-3.5 works way better in Perplexity for some reason. System prompt must be 10/10

[tropic crest]

Yeah agreed

[short creek]

Also you know those situations where you have to ask it something that you don't really know how to ask? (coding related questions I mean)

Try copilot when that happens

Does a great job understanding the whole thing

[tropic crest]

Just bought pro and noticed they turned off gpt4 by default

interesting

[signal hamlet]

Hi @tropic crest! As a Pro user you can keep GPT-4 always on.

[tropic crest]

Also is there a reason that wikipedia mode is only available on android and not web mode?

[halcyon coral]

The Wikipedia focus has been globally suspended, even in the Android version the focus no longer works.

[tropic crest]

ah

[halcyon coral]

Wait...yesterday the Wikipedia focus in the app didn't work, today it works again.... Should that be the case, @signal hamlet?

[signal hamlet]

Thanks for the feedback! It's ok, we are working on updates and improvements to the search focus section. Sorry if it was confusing.

[short creek]

You can enable it from your profile and it will use GPT-4.

[split plume]

for future reference if you plan to ever use something for reference (as an avid researcher) you should always make a copy of the resource on your local drive (unless you anticipate the resource is updated often)

[modern crown]

Major shout out to the team of Perplexity and how amazing the community is here ❤️

You have a forever customer with me!

[pallid trout]

https://www.reddit.com/r/perplexity_ai/comments/15zjdfm/am_i_wrong_or_has_perplexity_been_nerfed/?utm_source=share&utm_medium=android_app&utm_name=androidcss&utm_term=1&utm_content=share_button

[keen walrus]

will perplexity be given access to chatGPT plugins in the future? one thing i'm really missing is the ability to transcribe and summarize youtube videos

[ashen kettle]

Is there a context length limit on the code llama lab?

[tropic crest]

Do you have any suggestions on how to get most value out of copilot?

[tropic crest]

The Code Llama models provide stable generations with up to 100,000 tokens of context. All models are trained on sequences of 16,000 tokens and show improvements on inputs with up to 100,000 tokens.

https://ai.meta.com/blog/code-llama-large-language-model-coding/

[ashen kettle]

That's the model, I'm asking about the perplexity labs implementation

[lunar anchor]

Might be worth including a little tooltip about when to use Copilot, when highlighting over it, instead of the number of useages left for the day (or maybe in addition to?) 🙂

"Use this when you don't know exactly how to ask your question, but you want Perplexity to guide you towards the answer you're looking for"

[tropic crest]

does anyone have any examples of how they were able to use copilot to help them on a problem that regular questions didnt cut it?

[short creek]

Mmmmm

Let me try and come up with an example.

This could be an example. Kind of a generic question

But as you can see it can narrow the results for specific areas.

Instead of giving you a big chunk of text that you might not want to deal with

[tropic crest]

ah I see

that makes a lot of sense. I appreciate it!

[short creek]

At least that's one case scenario that I just thought of

You can definitely give it more uses

I personally use it for coding so when I want some extra reasoning and fresh ideas I might give it a try

But narrowing information is a really nice thing you can do with it

[floral merlin]

imho, I highly suggest you stick with the the basic mode, and if the answer doesnt satisfy you, then switch to writing mode

i've recently trying to use copilot after a while

and I cant find any use for it

[short creek]

I mostly use writing when I'm coding.

But there's a very specific use scenario for it @floral merlin You can access the internet through copilot on a specific query.

[floral merlin]

you mean that i can access internet in writing mode with copilot enabled?

do you have a specific example?

[short creek]

For instance you can upload a file to trigger "writing" focus. That way it won't access the internet but if you need to search for something on a specific moment during the conversation you can just enable Copilot.

Uploading files trigger "writing" so it doesn't search the web

But then you just enable copilot for questions you want to get out the internet

It's the best of both worlds

[floral merlin]

no it's not, for me

i specifically ask it to search, and i always get "As an AI model" type of answer

[short creek]

Hold on making an example for you

[floral merlin]

but writing mode is supposed to cut the model from using web results, right?

[short creek]

You don't set the focus to "writing".

you just upload a file in "all" focus

Then the conversation won't search on the internet

Unless you enable copilot

Try adding a PDF to a conversation

Then ask it to saerch the internet. It will come up with the usual "As an AI language model..."

Then Enable Copilot and ask it for some current data

[tropic crest]

So I messed around with copilot and I realize that I get better results when I describe the problem rather than asking it to do ssomething (how I usually interact with LLMs)

[short creek]

That's nice!

[short creek]

@floral merlin https://discord.com/channels/1047197230748151888/1136772120672747642

@tropic crest So, was it useful then?

[tropic crest]

Yeah

very useful

[short creek]

I'm glad to hear

[tropic crest]

I tried again by describing my problem and its giving better solutions

[short creek]

Yeah but it's hard to describe when to use it to another person

[floral merlin]

did you try the same question + detailed descriptions as a whole without copilot

then compare the quality

I often get noticeably worse answers on the copilot side when I do the comparison

[short creek]

I've tried both things and it usually depends. Mostly getting better results with Copilot rather than just using GPT-4 in "writing" mode.

[floral merlin]

okay so in other words, it depends on other factors

[short creek]

If you input nice prompts Copilot can give you really good answers

I still need to test it out more. I'm trying to find good case scenarios where I can use it more often

[floral merlin]

but I think if we can switch focus within convo, both copilot + basic mode can & will have more consistent answers

[short creek]

That would be way too cool

There are some threads talking about that specific topic around there

[hazy abyss]

I don't fully get the SCHOOL25 offer. Is the price $200/year in the annoucement pre- or post-discount?

[halcyon coral]

$200 without discount

[hazy abyss]

I've asked the support via email to help me apply the discount while I am still on trial (ends at some time today). They seem to have applied it, but the Stripe invoice hasn't changed it is still $200. I don't want to be billed the amount without discount. Should I cancel the trial before it is too late?

[halcyon coral]

Has support confirmed that the discount has been applied? Then it should be visible on Stripe.

If not, you should cancel the subscription to be on the safe side and contact support again.

[hazy abyss]

The discount didn't transfer over to Stripe. I am trying to contact support, so far no response, but we will see.

[limpid bough]

Hi! Does anyone know how to select Llama? I can't... Maybe not available in Spain? Appreciate

[halcyon coral]

Llama isn't implemented in PerplexityAI, you can use Llama @ labs.perplexity.ai

[halcyon coral]

I mean, did you get a reply saying that the discount was applied?

[limpid bough]

Amazing! Many thanks! BTW it isn't possible to train it with personal docs right? Even if possible, they wouldn't keep secret I guess....

[halcyon coral]

Yeah, it's not possible.

[hazy abyss]

No, all I've got is an automated reply that said the support ticket was closed.

At the same time, the billing tab on the Perplexity website has changed to "Discount applied". That didn't transfer over to Stripe, though.

A few minutes ago, Stripe attempted to charge me $200 following the expiration of trial periodn, which I declined on my bank's side. Now Stripe considers me an existing customer and won't let me use the discount coupon.

[halcyon coral]

All right, sorry for the inconvenience, please email [e-mail] again and describe your case.

[hazy abyss]

Done, thanks

[tame current]

why does this not work 😭😭
https://www.perplexity.ai/search/0552fa5a-eedb-45bf-8c04-19d7f5eb5e32?s=m

[slim mural]

because wrong promt

[tame current]

what prompt do I use ?

[slim mural]

like this is your request
best pc games search only for these tlds: fr, ru, io, me, org

even i can't at first understand what do you want from AI

this what u want from it i guess

[tame current]

how does ur discord look like that

[wraith crow]

✨

[wraith crow]

click on the emoji and hold the emoji in the popup

[tame current]

new ui?

i havent been on my mobile for a while

[cunning sky]

cute ui

[blazing dust]

same

[lusty basin]

Hi, I am wondering: are perplexity threads publicly viewable/discoverable by default? What if I submit a PDF document, is that also publicly viewable?

[jagged ivy]

Hello
Try to ask this question with a pro subscription
Because with free pplx it answers isn't correct
They sold a Cadillac for 5000 marks, bought it for 6000, and sold it for 6500, how much they earned

[lament imp]

This was GPT-4 + Copilot

[jagged ivy]

In my case it is gpt 3.5

[lament imp]

Gotcha, I thought you were asking someone to show you what the answer would be on GPT-4

[jagged ivy]

I can't test free copilot because I ran out 5 uses
I remember that there was the thread #1136991521401471029 and I want to check this now

[frank rivet]

Hi! threads are not discoverable by default, only when you share the link. Your uploaded document isn't either; even if you shared the thread with someone they wouldn't be able to view the document, just the response.

[lusty basin]

Thank you. I just subscribed to pro since this was my one reservation about using it.

How does the regular mode work without copilot enabled? Is it still using AI to synthesize the answer? I am trying to understand the difference with copilot enabled vs disabled. Is it mainly that it asks follow up questions with copilot enabled?

[tame current]

What does YouTube Focus mode do?

[lusty basin]

I believe it would limit web search results to that source only

[frank rivet]

Perplexity always uses AI to deliver results, even with regular search. Copilot is an interactive search assistant that asks you clarifying questions to guide the AI search process and find the best answer for you, we updated it to be even sharper recently, so it's quite fast. It's great for in depth research, whereas a quick search is also very good but will give you a concise summary compared to Copilot.

[tropic crest]

Does anyone have a good prompt for editing writing? I generally just want mild editing to make the text more readable, clear, and consice but I find that gpt often makes it sounds robotic and uses jargon

[sage lily]

Co pilot usually asks questions but not always. It usually asks questions when it needs a better understanding of the prompt or of what you want. Usually the answers from Co pilot are better than the ones without Co pilot being used.

Sometimes though, the questions from Co pilot are what you want the Ai to answer or the questions are too hard to answer so you can skip them.

[lusty basin]

I have noticed that

[lusty basin]

Does it consider more web sources as well?

[short creek]

Yes, up to 25 sources

And if it asks you more questions it may search again and add even more sources

Then provide an useful response

[lament imp]

Is it only me or has been Perplexity been substantially faster today in beginning its responses?

[sage lily]

Also obviously it won't consider any web sources if you're on writing mode/focus.

[short creek]

What's wrong haha

[chilly forge]

Yes it was ig the Free Co-pilot & Pro Co-pilot acting differently in answers for the exact same question
PS- No idea about the current situation though please try yourself once to confirm

[chilly forge]

naah its true

[warped marsh]

What does this mean? https://twitter.com/perplexity_ai/status/1695102995325710484?s=20

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

485

▶ Play video

Twitter · Aug 25, 2023, 3:57 PM

From now on perplexity copilot will not use gpt 4? If it won't use gpt 4, then why does the web ui still say "Summarizing: GPT-4"

and so if I ask coding questions with copilot on and in writing mode, then I will get answers from their "fine tuned model of gpt 3.5" and not gpt 4?

[pallid elbow]

Should I boost this server for a week

[pallid elbow]

No they’ve kinda made gpt 3.5 to gpt 3.9 or smth like that

[halcyon coral]

Whether you have selected GPT-4 or not, GPT-4 is always used for the answer.
However, the interactive process of understanding the question, asking questions and building search queries is now done by GPT-3.5 FT.
After that, however, the sources are used by GPT-4 to generate the final answer.

[warped marsh]

Where are the benchmarks then?

[pallid elbow]

Uh like a bit worse than gpt 4 I think

[warped marsh]

Oh

[halcyon coral]

There is no GPT-3.9, just a fine tuned GPT-3.5

[pallid elbow]

Ik

[warped marsh]

Won't gpt 4 understand the question better though?

[pallid elbow]

It’s like they turned it to gpt 4 but still using 3.5
It’s very confusing lmao

[devout sun]

Why has the available quantity for search decreased to 300 per day, while the price remains unchanged?

[halcyon coral]

The fine tuned model is specialised in this tasks.

[warped marsh]

Is there any benchmark for it?

[warped marsh]

So if I am on writing mode then 3.5 will only be used to understand the question and for nothing else. Correct?

[halcyon coral]

No, I don't even know if a comparison of GPT-3.5 FT and GPT-4 would be useful because, the model is specialised in understanding questions and forming search queries.

[warped marsh]

For building search query, is 3.5 also being used in the normal perplexity pro search?

[halcyon coral]

In normal mode with GPT-4, only GPT-4 should be used.

[warped marsh]

Even for building up the search queries?

Alright. Thanks @halcyon coral

[halcyon coral]

NP, btw here is a speed comparison: https://twitter.com/perplexity_ai/status/1695103002246340854/mediaviewer

[devout geyser]

It has?? 😮

[halcyon coral]

I still have 600

[warped marsh]

Same. Had 600 in the beginning

[devout geyser]

Same

[pallid elbow]

Well won’t make a difference

You won’t search more than 300 times

A day

[sage lily]

Unless.... 👀

You're a massive power user lmao

[sage lily]

What model was being used when understanding the prompt and deciding what questions to ask the user before this change was made?

[halcyon coral]

GPT-4

[sage lily]

I see.... So they switched to Gpt 3.5 but it should be better at understanding the prompt and asking questions because of the fine-tune material/way it was fine-tuned.

Interesting.

Does this mean less dumb questions or what lmao.

[halcyon coral]

In any case, faster responses.

[sage lily]

I didnt really mind the speed before. It wasn't too slow.

Lower costs is an interesting thing though.

I wonder how much lower. And if that means they could give people more Co pilot uses per day even though no one really needs more lol.

[pallid elbow]

Lmao it is enough for me

[sage lily]

Same tbh.

[devout geyser]

indeed, the most I've used in a day I guess is close to 200, most days I use 50-100

[halcyon coral]

https://twitter.com/perplexity_ai/status/1695103004746100830/photo/1

Perplexity (@perplexity_ai)

The transition to the fine-tuned GPT-3.5 model also reduces inference costs. This efficiency enables us to continue investing in enhancements, making your Perplexity experience even better. Try Perplexity Copilot now and see the difference for yourself!
https://t.co/RySTpcMaph

Twitter · Aug 25, 2023, 3:57 PM

[tame current]

does perplexity use Google's cache for accessing search results ?

hm I guess it does, that would explain the speed

[jagged ivy]

If it is cheaper there will be more copilot uses?

[halcyon coral]

I rather think that there will be more functions. I think that the number of Copilot uses is sufficient.

[tame current]

for free users ?

[halcyon coral]

What for free users?

[tame current]

I mean is it sufficient for free users

it's only 5 per 4 hours

they could increase it ig

[halcyon coral]

Yes, I know, I only meant the Pro subscription.

[tame current]

I know what you meant, I'm just pointing out

[wind musk]

when perrplextiy is typing my whole screen goes like flashing like crazy, that looks like garbage and epilepsy people gonna get crazy

fix that

[tame current]

Hello, anyone has an idea when will this option become available? Thanks

[halcyon coral]

The devs are constantly working on enhancements to the existing functions, but there are no precise timelines for when a function will be released.

[short creek]

I have 600

[chilly forge]

^Facts

[short creek]

Most of the times I do.

I would not like them to make it half tbh

[jagged ivy]

I meant free pplx, because now there is only 5 in 4 hours

[jagged ivy]

Hm, it isn't the correct answer
Because they lost 1000 marks, and earned 500, so they lost 500 marks

[warped marsh]

He can understand better than anyone

[jagged ivy]

Promocode for students works only for new subscribers?

[lament imp]

What

[jagged ivy]

If you sell a car for 5000, buy it for 6000 and sell it for 6500 then the revenue will be -500 marks

[lament imp]

You sell it for $5000, you're up $5000.
You buy it back for $6000, you're down $1000. (5000 - 6000 = -1000)
You re-sell it for $6500, making $6500. Add that to the -$1000 you're currently at, you have $5500. (-1000 + 6500 = 5500)

Unless I am misunderstanding

[pallid elbow]

What

[jagged ivy]

If I am in the red for 1000 I need to sell a car for 1000 marks more than I bought it, that is, for 7000 marks

[lament imp]

To do what?

[jagged ivy]

What?

[lament imp]

Nevermind

[jagged ivy]

Okay

[wind musk]

is there an api for perplextiy

[carmine linden]

Hey @wind musk! Perplexity AI does not currently provide an API.

[wind musk]

oh ok

that was so useful, how to setup my bot like this lol

[candid heron]

Anyone know what the heart icon does in the UI?

[plain trench]

Perplexity is an incredible BOT it keeps getting better and better, no hallucinations just, I can't do that! Its ChatGPT only PluggedIn🥹👌🤝

[strong narwhal]

GPT 4 limit is 600/24 hours what is co pilots limit?

[short creek]

Copilot is GPT-4 itself.

So it doesn't matter if you use Copilot or GPT-4. Both of them will consume 1 token on use.

[strong narwhal]

so 600 combined limit for both?

[short creek]

Copilot spends 1 token. GPT-4 spends one token.

And they will replenish dynamically throughout the day.

[short creek]

If you use GPT-4 or Copilot at 12:30 AM on tuesday and spend 1 token, It will come back wednesday at 12:30 AM.

[strong narwhal]

Ok understood

Thanks

[tame current]

it just "like"

is there a way to get more results from perplexity?

[chilly forge]

Like button

[chilly forge]

As in ? Ik like the sources has been capped at 25 now are you talking about it ?

[tame current]

no, more results besides the ones it shows or rather I should say different results

like forums instead of just blog posts

[chilly forge]

More Results —-> More Sources right ?

[tame current]

yeah ig

[warped marsh]

25 is also a lot though, no?

[tame current]

not really

depends on what you want

[warped marsh]

I haven't come across any question or research prompt that would need more than 25 sources unless you just ask perplexity to search for your entire research paper in a single go.

[warped marsh]

In which cases do you need more than 25 sources?

[tame current]

any time I use it like search engine

rather than just for finding answers

[warped marsh]

like? If you are comfortable sharing then can you share any one prompt that in your opinion requires more than 25 sources?

[chilly forge]

Depends really on the topic ! If the topic is too vague and the sources itself individually doesn’t give a clear info then it would lead you to a more confused answer than earlier

[warped marsh]

I, personally, haven't come across any prompt like this

[chilly forge]

Assume there is a JEE question which many sources answer differently. Ik this isn’t the case nowadays since there are way too many EdTech sites but back in my days I didn’t got a correct answer for any JEE advanced question by Googling

[warped marsh]

in that case more number of sources would lead to more inaccurate answers

[tame current]

I mean that happens with 25 sources too, just ask it for suggestions for movies or something and the answer isn't even a list

I often just go straight to the sources, so having more helps

but I'm the only one who does that 🤔

[paper viper]

#💭│ideas message

[tame current]

Where can I ask questions?

[halcyon coral]

Here

[narrow cove]

Hi! Can Perplexity be used offline? Like on a PC without internet?

[zinc folio]

hello everyone

my name is bob

i do have a few concerns with perplexity

it does not give me exist searches

they are all generic as hell

[wraith crow]

You'll need internet connection to make use of Perplexity

[tame current]

I wanted to ask if perplexity also supports math. I mean, for example, displaying fractions, radicals, and exponents

[halcyon coral]

https://www.perplexity.ai/search/6f160081-033a-47ac-b41e-9da197255a5e?s=u

[tame current]

thanks

[shell terrace]

Hi all, I'm not sure if there's a better channel to ask this in. I want to understand if Perplexity retains context in threads and if so, how much context it retains. I asked this in a more detailed way on Reddit, so here is the link if helpful: https://www.reddit.com/r/perplexity_ai/comments/162uxqa/does_perplexity_remember_context_in_a_thread/

[median flame]

Is there any way to increase/adjust the font size in the iOS app? My eyes are getting a bit on in age. Thank you 😊

[paper viper]

It may change if you change your phone font size: https://support.apple.com/en-us/102453

That option is for the whole phone tho sadly

[chilly forge]

I think so there is a setting to change app specific fonts as well

Ref- https://www.perplexity.ai/search/f7c35b3c-4493-4494-946a-100bc8b217f3?s=u

[paper viper]

Ooo dope didn’t know that

[tame current]

aww Android doesn't have that

[still vessel]

Hi guys, small ui issue when in focus on my Huawei Pro 40 (in browser). Overlap header / logo.

[mossy basalt]

passaparola x un povero uomo di 50 anni senza donna,che cerca compagnia vicino prv di vr,per ritornare hai anni 2000

[tame current]

Hi @still vessel , which browser is this ?

[halcyon coral]

(Should be Huawei Browser)

[silver laurel]

Should i get the pro or just the the free version? The free one can help me with content structure and ideas generation. Anyone have reasons to upgrade?

[halcyon coral]

Almost unlimited GPT-4 is the big advantage. Plus unlimited file uploads.

[still vessel]

DuckDuckGo. It's fine on the default browser. But I mainly use Duckduckgo, which is faster ;-)

[halcyon coral]

oof sorry

[lament imp]

Huh I just tried it in my DuckDuckGo, so it may be on Huawei's end

If refreshing doesn't work that is. If it does then it was a loading error

[halcyon coral]

Has duckduckgo changed the layout to the huawei browser layout? This is what the default huawei browser looks like (2022):

[tame current]

Could you also share the browser version and the android version please ? Thanks!

[ancient geyser]

is there a way for perplexicity to be able to search specific websites databases and search bars?

[agile surge]

is there a comparison on the context can be held between chatGPT4 and perplexity?
Like I am using ChatGPT4 for more than 6 months with context about my new startup-
Can I do something in perplexity thread?

[rugged hemlock]

love the new widgets for iOS..!!!

[stiff solstice]

Export data maybe? I haven't tried it!

[agile surge]

doesn't work that great. I can just keep it but cannot utilize it directly

[stiff solstice]

What exactly are you trying to extract?

I copied the contents of the HTML and put it in a .txt, which perplexity is able to read

70k words though, so OCR can't scan it all in one go

you could take the HTML text and dump it into Claude 100k for context trimming

Just an idea though

[agile surge]

I want to retrain perplexity on the conversation I had on chatgpt4 in a thread for last 5 months. All of a sudden by ChatGPT4 went to default chatgpt3.5 because of that I switched to perplexity.

[hollow plover]

Hello. Where are you located?

[agile surge]

India

[tame current]

Perplex just recommended a piracy site to me :,)

[signal hamlet]

Hi @tame current! Thanks for reporting, please DM me or share the permalink with the issue in the bug-reports. We'll fix it.

[tame current]

Ah. I already cleared my history

I don’t really like clogging stuff

[signal hamlet]

Got it. Please send us a link, if you get something like that.🙏

[tame current]

No problem! Such things are impossible to prevent completely. I’m sure you all are working hard in this project

[pulsar pendant]

I started to use the file upload feature and chat with Perplexity about pdfs and it's amazing. If I could make one more suggestion, I would ask that you add a UI similar to chatpdf.com that shows the pdf itself, and the generated text links to each page number in the same way that Perplexity would link to sites when searching the web. Furthermore perhaps the file upload could be extended to docx, epub and other extensions. Thanks for reading!

[keen walrus]

will PAI be getting plugins at some point? might consider replacing my chatGPT sub with this if so

also does PAI pro do long form answers like chatGPT, or is it still short form of like 2-3 paragraphs?

[short creek]

I don't know about the plugins but you can definitely ask for longer answers and it till provide them.

To an extent.

[tame current]

just my opinion, but you shouldn't try too hard to prevent piracy, it affects legitimate users too... in some way, like it getting more restricted or less useful. not to mention it's not exactly you guys job, there are many people wanting to stop piracy but they've never been successful even the larger search engines (Google, Bing) haven't been able to. as long as perplexity doesn't intentionally promote piracy you're good to go and if you get a DMCA or something you can always remove those things later

just saying

[paper furnace]

how do i delete all threads? the button used to be there is gone!

[chilly forge]

You got one more reason to get iOS now

[chilly forge]

not sure about the "databases" but you can always try the site:

[tame current]

too expensive + locked down

[chilly forge]

True ! After using it I could say you pay for the experience ig

[tame current]

not worth it for me atleast

[chilly forge]

btw if you are in USA is its price is almost the same compared to other Andriod flagships now ?

[tame current]

not US

[chilly forge]

Ahh Ok

[ebon zodiac]

can i use Llama in perplexity in Android or iPadOs?

[chilly forge]

Using WebApp Yes - https://labs.perplexity.ai/

[ebon zodiac]

thanks!

Has it not been integrated into the app yet?

[tame current]

nope

[ebon zodiac]

i known

hasn't llama been integrated into the perplexity AI?

it's just a chat mode in the lab.

sadness

[tame current]

it hasn't yet

[narrow silo]

hey

Perplexity team

would you hire me?

my biggest strength is machine learning

https://i.imgur.com/jhCNHqk.jpeg

[wind musk]

will perplexity ever fix the flicking when message is being written

[tame current]

Don’t most AI lag a bit while writing large messages?

[warped marsh]

Flicking?

[chilly forge]

*Flickering

[warped marsh]

Aaah that glitch

That is annoying but doesn't happen a lot of time

[chilly forge]

yeah for me it has been months since last time I faced it

[tame current]

Claude 2?

[wraith crow]

Yes Claude 2 got added today.

[ancient geyser]

nice

im not sure if this is a bug

but perplexity has poor "memory"

I would like it to take into account the history of the thread when I make a new prompt

however it completely starts from scratch

has anyone managed to fix this issue themselves using prompts or a different method?

[wraith crow]

You can create a thread at #1140622008086970420. And also share the thread where this problem occurs. Then the staff will be able to help you better with this problem.

[ancient geyser]

I dont have the role @wraith crow

as in it doesnt let me

[wraith crow]

everyone is able to create a thread you just need to give it a tag at the bottom of the textarea

[wind musk]

yea, you know if it ever gets fixed or nah

[jagged ivy]

Can't wait for llama 2 because its speed

[chilly forge]

already there but on labs

[jagged ivy]

But it can't search in web

[chilly forge]

hmm

[jagged steppe]

HI everyone, I have been using claude.ai with a vpn, I first tried perplexity, but Im not sure if I used it correctly. Claude started putting limitations on how much I can ask. I have a couple of questions. What is the context length of a perplexity thread? Also, is it worth paying for the paid version? does the free version have any access to gpt 4 or claude 2? Can you choose which model you want to interact with? The co pilot feature is secondary to me because I primarily want it for help writing content.

Also, is it able to read handwritten content?

[halcyon coral]

Context length is up to 32k with GPT-4 or 100k with Claude.
You can choose in the Pro version which model you want to interact with (Perplexity, GPT-4, Claude)
You can´t use GPT-4 or Claude in the free version.
Yes it is worth paying, you get 600 GPT-4 questions/24h

[jagged steppe]

Omg, that is amazing

Is it able to read handwritten content? sometimes I need help reading my student's handwriting

[chilly forge]

Its not multimodal so No

GPT4 is but not in Perplexity as of this date

[pulsar pendant]

No, but you can ask Perplexity if there is an online OCR that is able to read handwriting.

[jagged steppe]

does anyone perhaps have a tool I can use in conjunction with perplexity so that I can read handwritten content and then paste the results into perplexity

[jagged steppe]

Great idea

[arctic dove]

I thought the 600-use limit only applies to Copilot right?

[arctic dove]

Bing has this feature afaik

[halcyon coral]

GPT-4/Claude/Copilot

[chilly forge]

@signal hamlet

[halcyon coral]

👀

[devout geyser]

Congrats on the Claude 2 integration 👍

[signal hamlet]

Thanks for reporting. That guy left the server 🔨

[halcyon coral]

Left or got a ban 😁

[hollow osprey]

Hey guys! What does actually Copilot does???

Got it!

[hollow osprey]

I wonder it is using ASR or using transcriptions directly to build answers.

[hollow flare]

Oh my gosh. I've been off pro for a few months, gonna have to get back on to see how the Claude-2 integration is. I've been using it over ChatGPT about half the time anymore. Great work guys.

[vernal dirge]

The gpt 4 usage is unlimited it's actually the copilot that has the usage limit of 600

[halcyon coral]

No, sorry.
600 uses = Copilot, GPT-4, Claude

[vernal dirge]

Wait I remember using perplexity more than 600 times once and there was no issue?

[halcyon coral]

I can't say anything about that, I've never reached the limit, you probably used GPT-3.5 afterwards.

[uneven sapphire]

• what's the appeal of perprelxity over chatgpt?
• which model are we talking to when we select perplexity? gpt3.5?
• copilot on 3.5 is just a good gpt3.5?
• does copilot mean gpt 4 when activated?
• what doe copilot mean when activated for claude?

[vernal dirge]

It was clearly showing that it was using gpt 4

[vernal dirge]

Copilot always uses gpt 4. Perplexity is better than chatgpt due its reliable internet accessing feature and is very good at research. Copilot is like gpt 4 on steroids.

[modern crown]

Is the perplexity model in the pro settings gpt 3.5?

[vernal dirge]

Yes by default I think it is you have to change it in the settings.

[modern crown]

Cool, I was just wondering that

[halcyon coral]

It is GPT-3.5 FT

[uneven sapphire]

i feel like it can't hold context in a conversation between two questions and i give up everytime and go to chatgpt

[modern crown]

So if we turn the model to Claude 2 and use copilot will it use gpt-4?

[halcyon coral]

It will use Claude 2

[modern crown]

Makes sense

Thank you!

Has anyone thought Claude 2 is better than GPT?

Or use it a lot?

[hollow flare]

Claude is definitely better than GPT 3.5, arguable on 4 - they're just different

[vernal dirge]

It can actually happen sometimes there can be issues that's why there is a support channel to help you out ☺️

[vernal dirge]

Yeah they were made for different purposes and different focus

[keen fog]

Hello everyone. What process does perplexity follow when handling large PDFs, web pages, and YouTube videos in sources? Does it search for specific content? summarize all the content? or use a search function similar to Google? Does it only analyze a small part of the webpage and ignores the whole page?

[modern crown]

I am super excited about Claud 2 being on here because we then have internet access for it!

I have used Claude 2 a bit

[shut gyro]

How do you get Claude 2 to only reference the uploaded file and not do an online search?

[halcyon coral]

You can use the Writing focus.

[shut gyro]

Every specific question I ask Claude about the PDF I uploaded is fabricated and can't be found in the original document.

[thorn moss]

Question, not file upload. in search internet and writing how many input words can we write to claude 2?

[tropic crest]

ask it to quote from the origindal doc

[shut gyro]

I did, it made things up

[last remnant]

I couldnt find the #📘│faq but is there a link that provides the difference between these models?

[halcyon coral]

• https://www.linkedin.com/pulse/claude-2-vs-gpt-4-2023-comparing-top-ai-models-kim-garst
• https://medium.com/@Soke.AI/comparing-claude-2-and-gpt-4-a-technical-deep-dive-121ffc5bd255
• Perplexity is just fine tuned GPT-3.5

[last remnant]

Thanks!

[thorn moss]

up @halcyon coral

Congrats perplexity you are killing openai and i cannot escape this subscription forever. 😢

[halcyon coral]

2k input (tokens)

[thorn moss]

Thanks man

[neat flame]

How can I use the Claude model on the android app?

[rotund geyser]

For now you will have to change it via the web interface as far as I know

[gusty blaze]

Just wanted to say I’m loving the new iOS app update. The addition of images was a much needed one

[edgy fiber]

Apologies if this has been answered previously (a quick search didn't turn it up) but what is powering the search portion of Perplexity? i.e. when it responds with specific cited sources, etc. Is it using Bing or DuckDuckGo as the back end?

[barren lily]

hello

[eternal saddle]

I'm also interested

[jagged ivy]

Where to report Claude 2 bugs?

[wraith crow]

You can just use #1140622008086970420 and add the web tag, you could add Claude in the title

[jagged ivy]

Oh I forgot about this forum, thanks

[karmic charm]

I'm sorry, but I am new to this hole AI-business, so, what is Claude?

[wraith crow]

claude is a llm that can have a context size up to 100k tokens, it’s basically the same as gpt but developed by another company

[jaunty ether]

hi, does anyone have a tips or a link to a guide on how to better configure the profile? on my threads I seem to have to keep repeating information that's pretty explicitly stated there...

[signal hamlet]

Hi @jaunty ether! We have the forum - #1124060874785685554, where you can see examples of AI Profile settings that other users tried and how it worked for them. Please check that and let us know if that helps.

[signal hamlet]

Also, if some of the Copilot clarifying questions are totally irrelevant, feel free to share the faulty threads in the #1140622008086970420 channel, we'll check.

[finite spoke]

Thanks for brining in Claude2 . I have been using the model over a month and was better than GPT4 for my use cases. Is there a way to context switch easily rather than going to settings to update the model. Or else notify the user which model is he using (rather than looking at settings)

[chilly forge]

Nothing as of now

[stone radish]

What is the advantage of pro version over chatgpt plus? Do I still need gpt4 key?

[chilly forge]

#📘│faq

[ancient hemlock]

btw was the winner of the sunday’a pro subscription lottery announced?

[chilly forge]

#💗│sharing message

[idle zodiac]

This is what i need in my life !

[uneven seal]

Hey with Claude is there any way for it to unpack a zip file and read a bunch of files from a codebase?

[idle zodiac]

👍

[uneven seal]

Interesting I’m going to need to experiment with this

[chilly forge]

Please give an update here if thats possible since I think its currently possible only with Advanced Data Analysis (formerly known as Code Interpreter) in ChatGPT Plus

[uneven seal]

Yeah that’s my assumption too

I’m not around this week to test this but I’ll be diving into Claude next week

[idle zodiac]

Im made my own perplexity workflow stack and already have things tuned. This just 3x'd my workflow I usually made it read a pdf but it still took it at the full 100k tokens I understand what you mean.

Also anthropic are testing a lot of things and your account could be hindered to less tokens if the inputs are complex. - So having multiple account in layers could help. Kinda sucks but ...

[sage lily]

Hello everyone.

Nice to see that Claude is available on Perplexity.

I thought that Claude would only be available to use with file upload but I guess not.

Also I have a question. For copilot with Claude. I saw that questions wise and clarifying the prompt it's still Gpt 3.5 fine-tuned. But when giving the final answer, Co pilot does use Claude aye?

[solemn trench]

how much better is Claude, what kinds of tests have you guys done? just curious

[chilly forge]

#1145321651022610462 message

[grizzled canyon]

it's going to depend on use case but claude can be more conversational and approachable where gpt-4 is better for technical applications

your mileage may vary as I found their temperament and approach to responses can be equally as valuable but different

depends which kind of insight or conversation youre seeking

[sage lily]

Yes I know that but what about the summary of the answer lol. Or the "last generation" it does before you can send follow up questions.

[solemn trench]

i see. interesting. ill stick with GPT4 then. thanks

[lusty bear]

I'm interested as well

[tame current]

@lusty bear @eternal saddle it's Google

it can just be deduced by using it, for example bing has a search operator feed: which finds RSS feeds and it does not work on perplexity

[chilly forge]

@pine lodge

Theres been so much spammy accounts lately xd

[pine lodge]

on it

🔨 bonked

[tame current]

happens whenever a server gets slightly popular

[edgy fiber]

Huh, interesting. I didn't think Google even had an API for its search product. Also, the responses and citations come back pretty immediately, which suggests that Perplexity is pre-storing a huge chunk of Google's results in a vector DB beforehand, is that right?

[carmine linden]

Hey @edgy fiber! Perplexity AI does not currently provide an API.

[chilly forge]

“Slightly”

[tame current]

Google does have an API and I know how perplexity is so fast

it uses google web cache

[chilly forge]

According to arvindh they have some external indexes

https://youtube.com/watch?v=jksGQhMtXjo&si=fQE0vYTPJ9DPU0kC

[tame current]

because one time I searched for price of a product on Amazon and the price was wrong then I checked google web cache and it matched up with perplexity's answer

[tame current]

timestamp ?

also, the fact that perplexity can access sites behind cloudflare is proof enough 🙂

[chilly forge]

https://www.youtube.com/watch?v=jksGQhMtXjo
9:38

Quoting his words-
So every time you enter a query in Perplexity, we understand your query, we reformulate it, and we send it to a search engine that is very traditional, multiple search indexes, not ours and external indexes. Pull up the relevant links...

[tame current]

oh actually, there are also many unofficial APIs that scrape Google

[edgy fiber]

Yeah I meant an official API. I've tried using one of the unofficial ones and it was extremely unreliable even at low volume, so I'd have to assume Perplexity isn't doing that

[edgy fiber]

This part is interesting. So they're doing some kind of translation between your query ("we reformulate it") and a subsequent Google search, without having to vector-index...the whole Internet or something