💬│general

[zinc folio]

ARF ARF ARF

WHAT U SAID TO ME LIL BRO

chat watch this kid he boutta lose this battle

[sleek vortex]

not really

well

yeah

maybe

for perplexity at least

well it’s not like it’s coming any time soon

still a month away

and it’s only going to be so good anyway

you’re fine on perplexity tbh

i can imagine a “x searches per day” limit too

or even per month… saw them testing that wirh chatgpt.com

just stick to one model then

4o or opus or something

I mean you’re not missing out on much

Most of the information is handled by other models anyway

The final model is just handling the conclusion

very true, yeah

sonnet 3.5 is the best all rounder for now id say

and best at coding out of that selection

however 405b llama might be more uncensored in some cases

[surreal marsh]

i'm sorry

it's so over for perplexity

https://openai.com/index/searchgpt-prototype/

[fading moth]

Pfft

Please

That's been leaked for months now and no one has cared. GPT may be a household name but it has a problem. It is only GPT.

[south kindle]

Ideally SearchGPT does well, and PPLX is forced to adopt needed changes finally.

[fading moth]

GPT is falling behind compared to other LLMs, and with LLM usage/popularity growing... More options are more important.

GPT search is going to just be what Bing is to Google.

[surreal marsh]

eh, GPT-4o is still ahead in most respects

[fading moth]

I disagree wholeheartedly

[surreal marsh]

OpenAI is Google and Perplexity is Bing

[pale compass]

While I'd say I prefer 3.5 sonnet to 4o, I don't know if I'd say GTP is falling behind (other than perhaps in coding with 3.5 sonnet) . It seems as if other models are now catching up, with Open AI now having a head start on the next generation.

[surreal marsh]

try having an actual back and forth with either

there's a reason 4o is higher ranked

[fading moth]

3.5 kicks 4o out of the water

[surreal marsh]

in what respect

maybe coding

[fading moth]

Writing, for one. I'd say 4o is better at coding/math, but the rest I'd give to 3.5.

Even then... Deepseek I've had better results than 4o for coding.

[surreal marsh]

users generally seem to prefer 4o

look at lmsys

even 4o mini

[fading moth]

Lmsys is flawed due to human evaluations. We know there was a bias due to 4o using markdown in responses to be more visually pleasant vs Sonnet.

[pale compass]

Honestly 4o mini being ahead of 2.5 is absurd and more than anything takes makes me take taht leaderboard less seriously now.

sorry, 3.5 not 2.5

[surreal marsh]

it just means people prefer it

[surreal marsh]

ok but that is actually important

[fading moth]

I don't care if the response has bold text if the code is subpar

[pale compass]

4o also has issues with repeating information when asked follow up questions in my expierence, something 4-turbo never had an issue with, and something people on that leaderboard likely wouldn't pick up on if they only use it for a one off prompt.

[fading moth]

Benchmarks have just been all over the place no matter where you look. I've seen benchmarks where 405b beats everything, mini beats everything, Sonnet beats everything.

[vital fossil]

Can we just wait to see how SearchGPT performs before we predict doom and gloom for Perplexity? Competition is a good thing anyways, gives us consumers better options. Let the AI wars begin 😉

[tame current]

that's a dumb take

[surreal marsh]

we're not the average user though

[fading moth]

The average user is getting more exposed to different LLMs daily. The release of Llama3.1 and the marketing that Google is putting into Gemini is putting GPT into a space of forced comparison.

Then we have sites like Perplexity that offer 4o and other models. Average users have options to test what they like more instead of being locked into one option.

Ultimately, all this is conjecture, and you could be absolutely right. Time will tell.

[harsh verge]

Competition drives innovation 🎈

[vital fossil]

Apparently that's a dumb take 😉 Or so I've just been told.

[tame current]

referring to the AI wars part

LLMs are cool and all but this isn't what most people imagined AI to be

I'd be interested in LAM (a true LAM not just API bindings) rather than yet another LLM

[sleek vortex]

apple had some interesting starts with like a model that can look at user screens

it was oss

but idk

apple intelligence is prob using that, but mainly siri intents

[tame current]

I see

@vital fossil sorry if my comment came off as rude

[vital fossil]

No worries, it’s not my first day on the internet nor was my comment that serious.

[fiery dust]

OpenAI testing a Perplexity clone was not on my bucket list

[mental bone]

It was only a matter of time

[placid coyote]

just why...

[torn sierra]

It looks much better than what we have in Perplexity

Visual quality seems to be really good. Images are also well organized. Ofc it doesnt mean it’s gonna be better overall

[spiral dirge]

hello guys

[shut plaza]

i signed up for the waitlist for SearchGPT

[torn sierra]

Hope so 🙂

[tame current]

i want ask a question, how can i know that the chagpt4-o that in perplexity pro is the orginal one that openAi create it ?

[jaunty dove]

It’s not like OpenAI is the only one adding search, Anthropic is also working on adding web search support to Claude later in the year.

[fading moth]

Because Perplexity has to use OpenAI's API to use 4o. Perplexity doesn't run 4o locally.

[tame current]

ok i don't ask abou search

[tame current]

What is the evidence on that ?

[jaunty dove]

That wasn’t to you, we were talking about searchgpt from OpenAI in the channel earlier.

[tame current]

ok

[fading moth]

Because OpenAI is closed source and doesn't let anyone run it locally.

:/

Same with Gemini and Claude.

[tame current]

To give you an example: He can use LLAMA2 instead of buying GPT4-o and write to you on the page that you are using GPT4-o chat and you are actually using LLAMA2

in another way: How can you trust that they are not deceiving you?

[fading moth]

Llama is open-source, for one

[shut plaza]

that would be a very strange thing for Perplexity to lie about

[tame current]

Why ?!

[fading moth]

Because it would be extremely illegal

[tame current]

I Know that LLAMA is open source

[shut plaza]

also, you can try it yourself and compare Perplexity's GPT-4o responses to the responses that OpenAI's API produces

[fading moth]

Also, if you ask the LLM what it is it will tell you what model it is.

[shut plaza]

that's not necessarily reliable tbf

[fading moth]

I guess

It will tell you what company tho

[tame current]

he can train LLAMA 2 to give u this answer, i thnk, but iam not good at AI, so i am not sure he can do this or not

[shut plaza]

also

why do you think Perplexity isn't using GPT-4o anyways

[tame current]

because perplexity offer to you:
1- Chatgpt4-o

[shut plaza]

what

[tame current]

and Claude 3

[fading moth]

3.5*

[shut plaza]

and your point is?

[tame current]

and another AIs, and each one of them cost 20$

[shut plaza]

it's called an API

[tame current]

i forgot that they added it

[fading moth]

Poe does this, you.com does this

Phind does this

[tame current]

ok, i know

Who told you that they are all so honest?

[shut plaza]

Perplexity isn't being charged $20/month like a regular customer

they're being charged by usage through the API

[fading moth]

They also have way too many collabs with big companies to just be "llama 2"

[shut plaza]

because you've provided absolutely zero evidence that Perplexity isn't using GPT-4o or Claude 3.5?

[tame current]

But this is not enough reason to believe them

[fading moth]

Your argument is that they are guilty until proven innocent but you have no proof.

[tame current]

they should firsts give me an evidence that they use those AIs then i will subscribe !!!!

[tame current]

that's true

[fading moth]

As someone who has used Claude on Claude.ai and GPT on their direct site... And someone who has run llama. Llama 2 is terrible by comparison and no amount of tuning could create the quality/logic difference.

No matter how much you tune a Volkswagen beetle it cannot match a Ferrari.

[tame current]

ok, but i can't believe on them, must first give me an evidence

[shut plaza]

what evidence do you even want

[fading moth]

It's impossible for it to be llama 2 because the answers the models give are different from each other and far beyond the capabilities of llama 2.

[tame current]

that they are using the orginal API From Openai and Claude 3.5 and the other Large Models

[shut plaza]

how exactly do you expect anyone to provide evidence of that

[fading moth]

They save cost by reducing the context window down from 200k.

They also have their own llama model called Sonar...

[tame current]

IDK

[fading moth]

How do you know your ice cream has milk in it? Just because it says on the ingredients label that it has milk doesn't mean there is milk in it. <-- Your argument

[vital fossil]

Thing is whatever proof they provide, you could still question it. It’s no win situation so you either get pro or you don’t.

[shut plaza]

yes

[vital fossil]

Some people will never be convinced. 🤷‍♂️

[tame current]

no that's not right, they didn't give me any thing, where is good the evidence that they gave us ?!!!!!

where is it ?

nothing

[fading moth]

Bruv, you only have to use Claude AI on their site and GPT on GPT's site to see the responses are the hecking same when called from Perplexity.

[tame current]

convince with what?

• are u give me anything to convince with it?

[zealous kayak]

I can't wait for searchGPT to come out "over the next few weeks" (next year)

[vital fossil]

Waste of time and energy this I fear.

[fading moth]

OpenAI has been operating on Valve timescale (is this still a relevant joke)

[zealous kayak]

no because I'm pretty sure OAI does plan on releasing something eventually

[fading moth]

Got me there

[zealous kayak]

tbh I barely care about the search stuff

I use perplexity mostly because they give you access to multiple models under one sub

so it's not like I ever plan on making the jump

[fading moth]

Rather 4.5 than search

[shut plaza]

i'll probably stick with perplexity myself

[zealous kayak]

yeah

[shut plaza]

but i don't mind giving searchgpt a shot if it's free

[zealous kayak]

to be honest, I won't

simple reason

I don't trust an AI with giving me factual information like, ever

[shut plaza]

i'll trust AI with trivial stuff

but i'll absolutely fact-check anything remotely important

[fading moth]

Critical thinking required

[shut plaza]

openai single handedly halving the critical thinking in students after releasing chatgpt to the public

[fading moth]

Yeeeaaaah, GPT kinda messing up education in humans low-key

[vital fossil]

I miss the option to organize stuff in ChatGPT. Perplexity has that going for it, aside from multiple models.

[shut plaza]

it'd be amazing if Perplexity could violate the deal between Reddit and Google

that's a deal that deserves to be violated

[fading moth]

Do we know if searchGPT is even going to be free or locked behind sub?

[shut plaza]

idk

[tame current]

prolly be free since competitors are free maybe

[fading moth]

That would be the smart thing

But

[shut plaza]

mayhaps

[fading moth]

Sensible deduction

[zealous kayak]

tbh I hard disagree

I think GPT exposed how terrible a lot of education is

if your course can be passed by memorizing and asking an AI as dumb as GPT

you're doing something wrong

but that's me and my bias towards open-book exams and the like

[fading moth]

Yes, reliance on memorization and regurgitation of information is a large flaw in the education system. But there are a lot of flaws in the education system; many due to teacher shortages and underfunding.

[lament spoke]

Searchgpt = meth addicted perplexity

[steel crater]

@signal hamlet Hi my brother, can you tell me why i can't use discount code for 50% off on perplexity but that same discount code works like a charm on my friends account?

[tame current]

What does this even mean

[hushed beacon]

hey all

quick question: is perplexity really good at coding? compared to chatgpt or claude?

[tame current]

It can use those models so that's a moot question

But for free, chatgpt or Claude sites are better

[hushed beacon]

i dont think its a moot question. using those models does not mean that perplexity as good as claude. claude may not provide all coding capabilities. i already researched it and could not find any sufficient information around that

[hoary tangle]

Hi

[tame current]

To your point, I'm still sticking with what I said because it depends on your used case. If you're coding problem is one that you think a lot of other people have run into, the results from Perplexity might assist in the result. But, if the problem is more nuance and requires raw critical thinking, the search results might clutter up the context window and the meaning of the prompt that you gave.

[livid aurora]

https://x.com/AravSrinivas/status/1816531378990932148

Aravind Srinivas (@AravSrinivas) on X

Someone made a cool "Let Me Perplexity That For You" LMPTFY tool: https://t.co/8gstKhOGGY

Twitter · Jul 25, 2024, 5:50 PM

[tame current]

https://github.com/Any-VM/AnyGPT/tree/experimental

Gpt.anyvm.tech/v1/models

Dm me for key and endpoint

[pulsar sail]

Hi guys, do you prefer Claude 3.5 sonnet over Opus?

I still find Opus superior in my opinion, but don't know if it's just my perception!

[tame current]

What does this do

[tame current]

Multiple ai models free within a single endpoint

I have pretty much all of them

[tame current]

Who is paying

[tame current]

Me

Why

I host anyvm and a multitude of other services

Not hard to pay

[rotund terrace]

anyone having issues with system?

[lament spoke]

Laughs in perplexity 😆
https://www.reddit.com/r/pcmasterrace/comments/1ec6i2n/google_is_now_the_only_search_engine_that_can/

[fickle otter]

hello friends, I love switching between the gpt 4o and Claude 3.5 models to diversify the answers. It is just a little bit of a hassle to switch between both. Is there a way to include in the prompt which model to use?

[worn zenith]

I mostly do writing/role play and opus is definitely superior in that department

[cosmic oracle]

Hello guys, do you know what happened to the "rewrite" bottom in the mobile app? (I'm using Ios)

[vague matrix]

I'm writing this because I am assuming you might be changing it from the settings: in the current version you can click in the bottom part after the answer, on rewrite, and switch the model. It will respond again with the model you choose. it's not what you asked I know but it's more convenient than going in the settings...

[fickle otter]

I used to change in the settings, but now I just click the "rewrite" link under the answer, it's the fastest way I found, but I would love to do something like:
"/1 describe this" or "/2 describe this" (where /1 = gpt, and /2= claude) So I could have the 2 answers simultaneously 🙂

[fading moth]

No, yeah... Having Sonnet answer half and 4o answers the other would require two calls at once.

[zenith wraith]

2024年母婴店的出路在哪

[royal trout]

Hello, I recently moved far from my family, and following my grandfather's passing, I realized I don't have many people left. After falling out with the last of my relatives, I found myself in a tough spot. Now, I'm in a new state where I don't know anyone, and it's quite intimidating. I'm facing the possibility of losing my apartment. I arrived a few months ago and have struggled to find a job. Being disabled, I urgently need to find remote work. I'm not seeking generic advice or suggestions for apps I could find online myself. Telling me to 'pull myself up by my bootstraps' and do something like Uber isn't helpful, as it doesn't fit my situation. My expertise lies in creating excellent bots I'm passionate about technology and excel in customer service and computer-related work. I've created some special bots that no one else has done I have a bot I have three very unique bots right now one detects lies the other 2 are quite special one actually searches the Internet every single output finds the exact right answer verifies it and then Outputs and then the very most special one is quite valuable it searches the Internet finds the best information on making the framework locates the best possible way to lay it down and does it in real time essentially it's a bot that makes bots in the most perfect way. I have a very special framework that I know once figured out will become the standard in writing bots. As I said I am gifted in that area or just super committed. im reaching out to my network as a last resort, having exhausted other avenues, and the situation is becoming quite dire. If you have any leads or know of opportunities, I would be immensely grateful.

[fading moth]

Wat... why is your username a website o_O

[royal trout]

thats my website yeah obviously its failing

but i launched it 100%alone 3 years ago

[fading moth]

And remote tech support jobs haven't panned out, eh

I am also disabled and I feel your pain, bruv. Wish there was something I could do~

[royal trout]

Well i would take tech support at this point i would take anything above 15 an hour

I realy only need to make 300 a week

So i dont lose my place

[tame socket]

anyone use SearchGPT?

[fading moth]

Yes, it's more aesthetically pleasing than Perplexity but ultimately it just feels like a shiny Perplexity minus the useful features.

Don't really get why they launched it instead of focusing on pumping out 4.5, or just giving GPT web access instead of developing a separate product.

Either I can't find the feature, or it doesn't exist, but not being able to create a customGPT for web searches, i.e., collections, is a huge problem.

[tame socket]

Because OpenAI's position has recently come under pressure, but the development of core new technologies is very challenging.

So, OpenAI may need some attention-grabbing initiatives or innovations to maintain its prominence.

[silver laurel]

The only feature i need now is able to regenerate the prompt query using different model given in perplexity. Going back and forth is not ideal

[sleek vortex]

elaborate?

what do you mean regenerate the prompt

[silver laurel]

After the prompt is entered and return result. I want be able to use another model to get a different answer to see which give the best context

Directly within the same page

POE does that quite nicely

[sleek vortex]

oh, so comparing

[tawdry chasm]

wait now if u turn off pro, u can still use the model without using ur daily number of messages?

[hoary tangle]

Just got 10$ referral code from friend. Subscribed to Pro. I hope it will help me in my coding journey

I work as Junior Software Developer but I study myself in outside work to improve

So I can work in better jobs in future

[craggy dust]

Guys can anyone tell me what is the context length of model offered by perplexity pro? Is it limited to 32k even for the models like 3.5 sonnet or is there any difference between perplexity 3.5 sonnet and anthropic sonnet

[sleek vortex]

no

shouldn’t be

[sleek vortex]

32k for all

[tawdry chasm]

can confirm it is

[sleek vortex]

probably a bug

[tawdry chasm]

probably

[sleek vortex]

they already aren’t making money that wouldn’t be not a good idea for them

[tired quartz]

What happened to pplx? My pro search doesn't do search at all now. Have tried several queries.

[tawdry chasm]

yea it doesn't search anymore

[tired quartz]

It works again. Strange sometimes pplx just doesn't do any search no matter how I refine the query and ask it rewrite. But over a while it starts to work again with the original query...

[rigid rock]

Wassup

[blissful plover]

Love the product.
I hope this is not harsh. I genuinely want the company to succeed.

However, I am curious about the company's response to the announcement of SearchGPT today.

[placid coyote]

I loved working with opus on pplx when it was "free" opus50. it was able to roleplay so well and still give okay code, comparable to turbo at the time. but now ... I am trying Cursor, there is 10 free opuses a day and for programming tasks there, it is not so good compared to Sonnet 3.5 😢 (though it could be because of nature of my project, or system prompts of cursor)

[fast meteor]

ask the android dev to implement this https://developer.android.com/media/media3/session/background-playback

[empty mist]

I keep coming back to this question. pro toggled on is supposed to "dive deeper" when searching the internet. So then is it really necessary when using writing mode since you're not using the Internet? some have said that with pro toggled off in writing mode the ai model acts more like itself.

[jagged ivy]

Is llama default ai model in perplexity?

[median summit]

It looks like OpenAI is trying to do something similar: https://openai.com/index/searchgpt-prototype/

[zinc birch]

This is exact copy of Perplexity

[placid coyote]

oh no, forgot to disable ai profile again and it incorporated the lesson inside the text 🤦‍♂️

you could try an approach from sonnet - https://www.perplexity.ai/search/could-you-analyze-and-summariz-ilkvNBDxTRm78rp5mk0Byw . it seems to work (https://www.perplexity.ai/search/please-respond-exactly-with-th-uQoe9hGxTwirIlMXuVdnfA). though I am not sure, if your text is much longer than 2000 words if you won't hit output limits. edit: not sure if it actually did any rearranging ._.

[fading moth]

Prompt enhancement bots are definitely a thing worth having in your pocket.

[fading loom]

anyone know if there's a published roadmap or anything of the sort for perplexity

[fast meteor]

only api team has one

[normal echo]

Mistral CEO on why infra startups are good now https://www.youtube.com/watch?v=Zim9BqRYC3E

[thorn comet]

Has anyone been encountering additional hallucinations today? I have had a bot generating daily reports via Perplexity's API for the last few months, and almost every report generated this afternoon was false - posting totally fake sources and information.

[hallow linden]

one reason I will probably stick with Perplexity Pro over ChatGPT Pro and the like is that others have caught up. Claude 3.5 Sonnet is better than 4o, and llama 3.1 405b is on par with it.

[tired quartz]

When there is Poe bot, there is GPTs. When there is Perplexity, there is SearchGPT...

[shut plaza]

what

[serene sand]

https://x.com/apostraphi/status/1816869114645807466?s=46&t=JsxhFTRLBknd8RUv1f73bA

Phi Hoang (@apostraphi) on X

Keep your eyes on your own path. That's where the magic happens.

Twitter · Jul 26, 2024, 4:12 PM

[tame current]

This server is strange

[silver laurel]

Just tried POE for code and they have a console to preview the code directly. Any plan for Perplexity

[fading moth]

Your guess is as good as anyone's

[tawdry chasm]

huh?

oh

#📢│announcements

[finite gorge]

if i upgrade to paid subscription is there unlimted gpt4 ?

[tawdry chasm]

it's not technically unlimited but I am sure u won't be able to hit the limit

[finite gorge]

is it the exact same conversation style and intellignece as using it on open Ai chat gpt? why do i feel like gpt 4 on open AI always just gets what im saying, and other i have to go around in circles

im tripping out

for coding, particualrly

[placid coyote]

should be gpt-4o (if you selected it), same as on ChatGPT. but there are differences - system prompt, how search results are passed to LLM (especially in pro search it could be so much context instructions get weakened or forgotten), also RAG (file uploads) is different (usually better on ChatGPT)

[tawdry chasm]

not quite, but u can make it the same

[finite gorge]

how can i make it the same

[placid coyote]

are you giving it code in prompt directly, or upload (eg long paste)? RAG on pplx can be funky (sometimes doesn't work at all)

[finite gorge]

its giving me code

[placid coyote]

the question is, does it seem to see your code?

[finite gorge]

i asked it what i want to create, perplexity just waffles on and doesnt listen , it gives me all this extra info i didnt ask for

[placid coyote]

if the model knows it, try use writing focus and disable pro search

[finite gorge]

nto sure what you mean

[placid coyote]

for example if you give few methods and ask it to add logging, isn't it trying to reinvent all the code of the methods?

[finite gorge]

also it forgets what i asked previously , so with exach updated code revision its not taking into accoutn the full convo, hence going around in circles

[placid coyote]

this can happen especially with pro search enabled - the data from search starts drowning "history"

[finite gorge]

im not tech savvy, i just ask it to make me a script for trading view trading

ok so best not to use pro if i want concise sharp answers

[placid coyote]

if it it is a new script, you are using common tech or APIs, try the writing focus without pro search

[placid coyote]

for customizing responses (eg concise, to the point), you can use ai profile and write it there. edit: it is in Settings -> Profile. then it will be applied to all new threads (except collections I think)

[finite gorge]

ok, so is their a point to suncribe to paid if im not using pro?

[placid coyote]

you are still getting pro models (gpt-4o, sonnet 3.5), a lot of uses. just not using the pro agentic search

for research the pro search is pretty good. but even if you don't use it often, you are getting 600 of omni+sonnet daily (with smaller context - around 32k, for many things still usable). that is quite good, but depends on your typical usage

[finite gorge]

great , thanks for your help

[tranquil orchid]

Hey @finite gorge!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[solar bison]

how long is the context window of Pplx llama 3.1?
super happy that we have it btw

[tawdry chasm]

I think it's also 32k

lmao there's a bug that if u turn pro off for Llama 3.1 405b u don't use ur daily number of messages

this is for Llama 3.1 405b only, not any other model

[ocean cipher]

its able to properly maintain context? Like I am only using prerplexity for searching using pro search.

But if its able to maintain this much context than its better for me? I have seen the perplexity doesnt properly main context.

[oblique trout]

Hello, i'm just wondering if it is possible to stop perplexity from appending sources to every request. AI gets confused by this a lot😭

[tame current]

use writing focus

[oblique trout]

ty but by sources i kinda meant something like images/documents u uploaded

like, if u upload one image for a specific request

perplex will keep appending the image to all subsequent requests

and AI gets confused a lot even if a request is totally irrelevant to the image

[tame current]

ah, yeah that's been an issue for a while

[tame current]

https://discord.com/channels/1047197230748151888/1237163951750385705

but I don't know if it'll work for images the same way

[oblique trout]

ah yeah ty. It works for images as well

it's just a bit of a hassle to remove sources every time

like u need to generate an incorrect response first

interrupt the AI

then remove the sources for the AI to generate the correct response

[torn sierra]

They haven't fixed it yet. Maybe they’ll never do. There's no easy way to uncheck what's uploaded to the sources. You have to edit the source from the previously requested message and remove what you don't want. It's irritating. Unfortunately, generating images isn't easy either. Although it's not an essential feature, it can be useful sometimes. In general, Perplexity is technically not as user-friendly and straightforward as other AI services

On the plus side, pplx gives you the access to many models with virtually no limits, and the search function is quite good. However, it's still worth double-checking sometimes because errors often occur in the responses based on search results

[oblique trout]

yeah that's why i've been using pplx for a while

[torn sierra]

I'm not sure if there's any better service out there. It's probably hard to find something cooler that would have such access to Sonnet and new models. I looked at You.com but I don't like it at all🥲

[oblique trout]

i tried you.com

their context mgmt is an issue

i used gpt4 some months ago on you.com

its context was too short from what i could tell

it seemed to forget a lot of things in my chat, which was not long at all

the response quality was also an issue, it was kinda different from what I could get from the official openai api

pplx's responses are generally good

in line with the api i'd say

not sure if it is better now but back then it was not as good as pplx

[granite grail]

According to their docs, their context length is 64k tokens, while perplexity’s is 32k.

Maybe it got updated since then

[torn sierra]

You can use writing focus, even better. The provided URL will work in writing mode 😉

[wraith crow]

There is not really a need for pro. You can indeed select writing (this might be better when you want to ask follow up questions without other sources involved) and ask your questions and provide an url.

[tawdry chasm]

the pro is for searching and executing code no? if u r not letting it search then what's the point of turning on pro, pro will literally not affect the results

[tame current]

Ok

[tawdry chasm]

sometimes u don't want it to work tho, gives unnecessary stuff to the model

[fast meteor]

i use the default focus. the internet acted as the additional knowledge base

[valid holly]

So far really good and learning to ask better questions

[tawdry chasm]

depends on ur use case

sometimes its better with the model not getting distracted

[quiet phoenix]

Hi guys,
is there any given way that i can set for perplexity that what kind of data range it can use?
Like "only 2024 data" for a specific trend search

[fast meteor]

pro search should ||will|| be intelligent enough to decide whether to use web search

[wraith rune]

Anyone notice the default model seems to be better than the Claude 3.5 or the 405b? Idk if it's just me

[random ocean]

Add ‘after:2024-01-01’ at end of your query.

What is the typical delay between when an article is published and when it becomes searchable in Perplexity's index? In other words, how current is Perplexity's database of indexed articles compared to real-time information?

[fast meteor]

popular news outlets are frequently updated, indie blogs are much less frequently updated

[tame current]

What large language model does the free version of Perplexity use? I asked it and it said it was made by open AI, and then I asked it what type of model from open AI, and then it went back on itself and said that it is made by anthropic

[bitter copper]

Hi everyone. I am rewriting articles for my blog with Perplexity. Does Perplexity Pro offer much assistance? I am thinking to subscribe Perplexity Pro.

[elfin ferry]

https://discord.com/channels/1047197230748151888/1225549032718602332

[tame current]

Which model does the free tier use?

[elfin ferry]

i dont think it's publicly stated, my guess is claude 3 haiku

[sleek vortex]

more than one

whatever they have free load for

haiku, own hosted llama

maybe 4o mini

[elfin ferry]

or that

true

[tame current]

ok

[zinc birch]

I’m new to perplexity pro, I want to generate some images but I am getting that is not possible by perplexity pro (I’m on iPhone)

https://www.perplexity.ai/search/generate-an-image-that-shows-c-kyD2BVGoQuaGlU1ysFz9KA

(This is the answer I got)

I apologize, but I do not have the capability to generate, create, edit, manipulate or produce images. I can only perceive and analyze existing images. I cannot generate a new image showing a "chill beach vibe" or any other type of image.

For image generation, you would need to use an AI image generation tool like DALL-E, Midjourney, or Stable Diffusion. These tools can create images based on text descriptions. Alternatively, you could search for existing stock photos or artwork that capture a relaxed beach atmosphere.

If you'd like, I can describe what elements might be included in an image conveying a "chill beach vibe", which could help you find or create such an image:

• A wide, sandy beach with gentle waves
• Palm trees swaying in a light breeze
• A colorful sunset or sunrise over the ocean
• Beach chairs or hammocks
• A beach bonfire
• People relaxing, reading books, or playing casual beach games
• Surfboards or paddle boards leaning against a beach shack
• Tropical drinks with little umbrellas

These elements could help create the relaxed, laid-back atmosphere associated with a "chill beach vibe".

Sources

[elfin ferry]

https://discord.com/channels/1047197230748151888/1225557012532695060

if you're on iphone it's not possible, only on web

i know, it's stupid

[torn sierra]

It's worth reacting so that the Perplexity team finally gets this sorted out properly and ensures that working with files is as it should be. This means having the ability to edit sources (files) on the spot, rather than messing around with removing files from sources after generating responses😎 https://discord.com/channels/1047197230748151888/1266708591763521626

[zinc birch]

Wow , a guide for this is outrageously stupid.

Why not make it a part of chat and give it to all users?

[torn sierra]

The answer is Perplexity is a search engine 😎 That’s what they say when you ask

[elfin ferry]

except if a new user doesn't know how to use an advertised feature you gotta do something

a guide in a discord server won't help most people

most people don't have a discord

or even look at #1224467082079371363

[zinc birch]

So the problem according to me is , I don’t have to go to search engine to understand some features are hidden away from mobile users right?
It should be Intuitive —

like let’s say I ask for generate image question- answer -which states I can’t do that on mobile and go to website on desktop instead

Any users must not go to guides to understand the processes. These guides should pop up as a response to questions

If that specific device / OS / version can’t do the task that was asked

[elfin ferry]

oh yes that's also important, at the VERY least make sure that generate image button is available on all platforms

but again, it's not ideal to have that button at all

[torn sierra]

What we have now is utterly moronic. And trust me, you're not breaking any news here. I've been around here for months now and despite countless complaints, they've done squat about it – same goes for the file handling. Add a file, and bam! It's stuck there like gum on a shoe, with the LLM constantly yapping about it. Want to fix it? Oh, you'll love this – you gotta edit the sources AFTER generating the response and yank that file out. Only then does the answer get a do-over 😅

[elfin ferry]

i know i cant stand that thing

you should also add this to #1111765634267742408

might have a better chance of being seen

[zinc birch]

Also I want to do some searches , I think let’s say
question 1, question 2 ,question 3, around 35 questions

Specific set of questions that I want to run on different companies.

But as the answer window or question window is limited I can’t put them in one question.

How do I go about it?
I just want these set of questions answered for each company name I want?

[fading moth]

35 questions at once?

1. For any LLM, that is a bad idea. You are better asking fewer questions per prompt than en masse. Part of this has to do with LLM reasoning and the other is output length limitations.
2. Depending on the complexity of the questions you might get subpar answers due to the LLM trying to squeeze in answers to each question within the allotted space for output.

[torn sierra]

I think Perplexity simply wasn't prepared for such major changes. They utilized LLM and RAG for information aggregation. However, search engines are now developing much better. For example, searchgpt seems to look visually much better, at least. Besides, there are many unresolved technical issues here with the UI and the application.

[fading moth]

To be clear, Perplexity's only LLM is Sonar. They do not run 4o or Sonnet.

Issues with RAG outside of Sonar are beyond the control of Perplexity. Perplexity is a frontend. They cannot fine-tune Claude or GPT. Claude is not designed for web searches and neither is 4o. There are bound to be issues trying to fit a square into a circle hole.

[torn sierra]

Let's hope searchgpt turns out to be a huge success and is superbly made. Maybe then it will force some concrete changes in Perplexity

[zinc birch]

I can ask 3 questions right, I want a way to ask these 3 questions separately on a collection and combine them in a parent collection

Hope I’m making sense

I don’t want to overwhelm with 35 questions at once, but I need these answers nonetheless

[fading moth]

Yes

[zinc birch]

So I am researching things for next investment I might wanna do and I know what I exactly want to know for different companies that i am researching for.

But I don’t want to ask these same questions 100 times as I am interested to ask these exact 35 questions against these companies that I’m researching.

So anyway that can be done is extremely useful. Collections does that but it’s kind of limited in my understanding

[fading moth]

Collections are meant to fine-tune the responses you want without having to repeat yourself.

If your goal is researching for investments etc. then creating a collection focused on the important factors when considering investments will help.

There is always the option of creating multiple collections with separate questions built into the collection.

Example:

You are an expert on investments and have <number of questions> to answer in order to generate accurate advice/data on investments.

How is <company> doing in the stock market?
What is the current market trend?
Etc. 

Another option would be to create a Collection focused on investments, what you're looking for in answers, specific things you're most interested in gleaning from data, what sites you want to gather the information from, etc. Then ask the questions in small bursts in a single thread with a singular"Expert Investor" collection.

There isn't really a best approach, and it will take a great deal of trial and error to set it up the way you want. The example I gave shouldn't be taken word-for-word, as it is just a random example.

Your task makes sense and is definitely doable, it will just take some clever prompt engineering and Collection creating to get there.

I definitely suggest browsing #⭐│starred and #1124060874785685554 for useful tools/prompts to give you jumping points.

Something worth looking into would be making a Collection for prompt enhancement. Essentially plugging in a prompt and having the collection reword/improve it for an LLM to interpret.

[torn sierra]

https://felo.ai/search

[torn sierra]

I think it’s great considering it’s free

[tame current]

Lol.

[shut plaza]

to me, this just seems like perplexity but worse

[torn sierra]

I haven't tested it extensively yet but if we compare the free option in Perplexity and the free option in Felo, it seems that Felo offers more comprehensive responses and is powered by gpt4o-mini which is a good model. In the free version of Perplexity, there are only a few pro searches available every few hours. I'm only talking about the free versions. Ofc, the pro subscription in Perplexity offers much more. 🙂

[shut plaza]

iirc the free model for perplexity is also GPT-4o Mini

[torn sierra]

I thought it was Claude 3 Haiku

I work mostly with Sonnet so I had no idea

[somber relic]

Why does the button to get the $5 free api credits not do anything? I've even waited 10 minutes

Also there is no differentiation in price for prompt vs completion token?

makes gpt 4 mini look ever so good

[sleek vortex]

My best guess is it’s a mix

Because they pay for different model capacities

Free will use whatever’s available, likely from a mix of these models:

• own hosted llama/“sonar” fine tune on their gpus (free capacity from labs)
• haiku (likely the main one as they probably pay for a ton of haiku capacity)
• 4o-mini (maybe? think i saw some users have it not sure)

also it may depend on region - they might have a seperate GPU cluster or aws/azure model allocation in say japan for lower latency

[torn sierra]

I played with it using the mobile version but when I opened it on my laptop, it really does look like the UI from Perplexity 😅 So I guess you’re right. It's probably the same template... 😅 Anyway, in the free version, it can provide quite a lot of sources (without pro search) and it's pretty decent. These are just my first impressions from playing with it 😉

[fading moth]

Doesn't matter if it can pull 300 sources if it isn't setup to properly communicate them to the LLM. Also, if they don't have Sonnet then count me out; 4o is inferior to 3.5 more often than not.

There are also diminishing returns on sources. Their content is skimmed, at best, and supplemented as information for the LLM. Given there will be an inevitable overlap of information in the sources; more doesn't mean better.

[arctic summit]

Hey

new to perplexity

[fading moth]

Welcome

[arctic summit]

is that the only interface to use models?

[arctic summit]

thanks 💙

[arctic summit]

so perplexity is just search engine using ai models ?

or i can use different llm models as solo

directly without the search engine feature

[pale compass]

@arctic summit Its main purpose is as a search engine, and that's where it excels. But you can change the "Focus" of a query before placing it to "writing". With Pro search off, this functions as a more typical chat environment for Perplexity.

Just keep in mind, all models are limited to a 32k context. So don't expect to be taking full advantage of Sonnet's 200k context window.

[arctic summit]

@pale compass Thanks 💙

[fading moth]

Yeah, LLMs (without web search access) have knowledge cutoffs for training them... Perplexity circumvents this weakness (sort of) by pulling relevant sources from the internet to give the LLM to supplement their response; information that might not be included in their training data or happened after the knowledge cutoff.

[arctic summit]

Okay .. last question .. which one is better approach , should i let perplexity find context for me and then i use this context in other llm to solve something

or should i use perplexity for both finding the context and solving my problem?

[fading moth]

If you're using 4o/Sonnet then you're fine with both. Perplexity (unless you're using Sonar) is just a frontend for LLMs.

If you're familiar with the game platform Steam... Steam doesn't own or run the games it provides (aside from Valve games). It just gives you a way to access them in a convenient manner. Same idea.

[pale compass]

I'd say it depends on your problem. If you're asking it to rewrite code for you. Well, The 32k context I had mentioned before is taken up by any files you upload, your prompt, and the search results, and the the return tokens for answers can be a bit limited in my experience. So If you're coming in hoping to use sonnet to spit out completed code for you, it may not be the best experience.

This also can be a limiting factor if your problem spans multiple queries. As previous queries can fall out of context quickly, again because of the limited context.

[fading moth]

There are other platforms that give this web access + LLM = improved response idea with a 200k context window, but they all run on an API you have to pay for; not a monthly rate.

Can't have web search + 200k at a monthly rate ~_~

[pale compass]

Oh yeah. I use perplexity a lot. I'm a Pro user and love the service. Just want to give realistic expectations.

[fading moth]

Ye

[tawdry chasm]

they will go broke lol

[placid coyote]

by any files you upload,
just tested it now (paste from clipboard) and while it seems to no longer use rag, now it looks like they cut off files. so you can't easily fill/use the promised context... https://www.perplexity.ai/search/please-respond-exactly-with-wh-yd1cfTcxTQmkMlOVUtkAaA

with normal file upload (not via paste), it seems to also see exactly only the starting 3.1k tokens of a file, but it can output rest of the file, in a horribly mangled formatting... 😞 not sure if it didn't broke the code, doesn't seem so on a first glance. probably rag for the second part https://www.perplexity.ai/search/please-respond-exactly-with-wh-3gIrXAlxSmCaySfyep1WdQ

[torn sierra]

It's not meant to replace anything in any way, I emphasized that I was referring to the free option. For a free version it's quite decent and might be sufficient for some people. There are no ads and yes, when I play around with it, it doesn't always exhaust all sources, but it simply skips those that didn't contain significant content (which is actually good). It just seems pretty cool for a free option, that’s all

[placid coyote]

okay, so my last test for now. you can make it to fully see (context, without primitive rag) your whole source file if it is over some arbitrary token count (around 3k) if you split it in multiple files... not very practical. in the split test it stopped at 3,095 tokens. https://www.perplexity.ai/search/please-respond-exactly-with-wh-FeTeHpErT4WDH_0Env9nVA

[pale compass]

Yeah, I'm not sure. Not sure how we can be without hearing directly from someone who knows their backend. But that would be a shame, because one of my primary uses of Perplexity is by uploading Chapters of my textbook to it and asking for clarifying questions. Would be a shame if it cuts the chapter off half way now and hallucinates. I'll have to mess with it some time when I'm less tired haha

[placid coyote]

natural languages, and especially english, are much more "token efficient", the approximation used to be 0.75 word = 1 token (for common natural english). but omni and sonnet use different tokenizers than what I was posting higher (that one is for gpt4+3.5, but the web tool is quick to use and looks good). so if you are doing chapter per file, it may work for you, roughly depending on your word count

[placid coyote]

tried asking pplx and it doesn't look very good. only the lowest range of common chapter lengths would fit (slightly above 2k words) https://www.perplexity.ai/search/do-some-napkin-math-confirm-as-HfKTtQjXSoyDW75d9yV6Yg

[tired quartz]

https://x.com/altryne/status/1817307423251546347

Alex Volkov (Thursd/AI) (@altryne) on X

SearchGPT (@OpenAI) vs @perplexity_ai

Ok I heard you guys, here's a teaser of a comparison I'm cooking up for @thursdai_pod (will be sent to subscribers first then to everyone, link in first comment)

TL;DR it's SUPER quick but basic and I really like it 👏

Twitter · Jul 27, 2024, 9:13 PM

[torn sierra]

felo.ai search is already better than Perplexity - at least in a free tier. Hope Perplexity team will implement some changes not to stay behind. Perplexity is good to have many LLMs like Sonnet but when it comes to search (its main focus) it may not be an option to choose

[tired quartz]

Don't worry.. SearchGpt will overwhelm them all.

[tired quartz]

100 pro per day is too less than on pplx.
Btw, can you modify your prompt there on felo?

[torn sierra]

tbh, hope so! 🙂

[torn sierra]

100 pro per day? Dunno what you are talking about 😅 In felo.ai, at least in a free tier and mobile version you can’t change your prompt - it seems so

[glacial anvil]

He's talking about the subscription.

promek means the free version of felo.ai search is better than the Perplexity.

[torn sierra]

Oh, got it. I didnt check on their paid version. felo.ai free tier can go through many sources and choose the relevant ones

[glacial anvil]

I think promek feels like felo search free version's answer seems provide the same quality to Perplexity Pro search answer?

[torn sierra]

Kind of. It looks that way tbh. I have Perplexity pro for several months now but mainly to use Claude LLMs like Sonnet 😉

[worthy heath]

I wanna thank yall for this app

it helped me alot these days

[tired quartz]

Is Genspark better?

[torn sierra]

Have no idea. Never heard of it

[tired quartz]

Someone told me Genspark is the best ai search, better than those mentioned above. But unfortunately its service is unavailable in my area.

[tired quartz]

AI搜索引擎（仅按字母排序）：

askan
arc
Bing
everypixel
genspark
gptgo
Globe
komo
perplexity
phind
rosebud
searcholic
songtell
scite
you

群友推荐补充：
秘塔AI
Perplexica
devv
360搜索（群友反馈脑图功能可以）
ChatGLM
天工开物
trackaianswers
maester
morphic
jeeves
bookabout
dorkgpt
miku
lepton
aisearch
flowith
thinkany

[torn sierra]

Just checked. It seems to be amazing

[fast meteor]

🥳

[tired quartz]

Just checked Metaso, it makes me shocked.

There are so many AI search engines today...

[tame current]

Genspark def the best

[tired quartz]

Perplexica - an open source pplx?

[tame current]

thanks to open-source RAG frameworks

[lone sedge]

Changed from perplexity to claude.ai. Perplexity was really bad :/

The context window is too small

[fickle sorrel]

once again, frustrated, i am respectfully asking how can i use the llm+scrapper at www.perplexity.ai on a pay per token basis? api doesnt work. thats just a offline model. even if i choose the 70b online sonar. thats all offline. on openrouter its all also offline 😭

[sleek vortex]

can’t forget uncovr!

[fickle sorrel]

._. help

@sleek vortex help plzzzzz i beg uuu sir miss plzzz

[fast meteor]

online models in api are online

[fickle sorrel]

then why do they not scrape data

they instantly answer.

[tame current]

scrape data from where

try asking something recent

[fickle sorrel]

from where? thats different with every prompt.

in labs.perplexty.ai the llms are all offline.

[tame current]

?

works fine for me

[fickle sorrel]

i dont know if that model came after llama 3

its also too fast.

[tame current]

but I do

[fickle sorrel]

its offline. its too fast for scraping

or they simply updated the model.

[tame current]

it definitely is online

[fickle sorrel]

why is much faste rthan their homepage then

[fast meteor]

they do not perform live scraping in general
the scraping takes place at some intervals
popular news outlets: multiple times a day
except this one: you could try the keyword site:example.com on pplx.ai

[sleek vortex]

online models doesn’t use the same system as the website

[tame current]

they also run the model themselves, the model is just fast

[sleek vortex]

likely they are just hooked up to a cache of their web index or a embedding search of their web index or something

it’s about as good or slightly worse than pplx free

[fickle sorrel]

and much worse than pro?

[tame current]

well, pro uses more expensive models

[fickle sorrel]

oO but like... on api it doesnt eve ntell me what is being browsed

[sleek vortex]

yes

[brisk quest]

the online API will not output the sources.

[fickle sorrel]

how useless?

[sleek vortex]

tldr quite useless

[brisk quest]

from their website: Access to citations and images via API is in closed beta. To request access to citations, fill out this form and send an email describing your use case to [e-mail].

[fickle sorrel]

guess ill need to sub to pro then.
im making a fiction novel about kaiju attacking germany and i want to be as true as possible with the how the bureaucracy will develop 😄

how does pp pro compare with search brave ai?

thank you

[fickle sorrel]

what is web beta

[fading moth]

Brave's AI search and Perplexity's search are both trying to make internet searches smarter, but they work a bit differently. Brave's search is like a regular search engine with some AI help, giving you a list of websites to look at. Perplexity, on the other hand, is more like having a smart assistant. It reads through web pages for you and gives you a summary of the information, with links to where it found the details.

Perplexity tries to answer your questions directly, often in full sentences, while Brave's search is more traditional but with some AI improvements. Perplexity might be easier if you want quick answers without clicking through lots of websites, but Brave could be better if you prefer to see and choose from a variety of sources yourself.

[fickle sorrel]

lol that account is made on 2023.. .and not banned

[fading moth]

@fast meteor do you have deleting permissions for scam like this?

[fickle sorrel]

u can win 100 ton XD 😄

100 ton of perplexity 😄

[fast meteor]

not in these 4 channels 😦

[fickle sorrel]

k then ping vova

[fading moth]

q_q

Yes

Ping @toxic moss

[fickle sorrel]

BAN RAS

BAN

[fading moth]

D:

[fast meteor]

i guess he/she got the account hacked

[fading moth]

It was a trap!

[fading moth]

Oof

[fast meteor]

i already timed out that account for a week

[fading moth]

[chilly nova]

my perplexity ai went crazy

it just writes stuff like this now

[halcyon coral]

It's better to ping a mod

[halcyon coral]

Please share the thread in #1140622008086970420

[tawdry chasm]

tbh I feel like that for the most part, a search engine should use a small and lightweight model, and then if the user needs, transfer it to a bigger model

most searches doesn't need a SOTA model to produce high quality responses

[fast meteor]

what is your default model? sonar?

[tawdry chasm]

been using L3.1 405b, but mostly feels like it really isn't necessary

Sonar is not good for the task tho

it's also not lightweight

[fading moth]

Sonar is really lightweight by comparison to Sonnet/4o

[fast meteor]

also compared to 405b, sonar 70b is lightweight

[south kindle]

Pretty sure this guys been at it since Easter at least when we got his main account kicked from the server and off the Easter egg hunt winner list. lol.

[fickle sorrel]

any way to improve this ? https://www.perplexity.ai/search/create-plan-how-germany-would-AZEtxrTwTeaYcG.yfGtP_w

[sleek vortex]

hmm

(i have no clue if that is a good response...lol)

[fickle sorrel]

that is not perplexity?

[sleek vortex]

uh yes, its not

[fickle sorrel]

uh then what is that?

[fast meteor]

that is another guy called crazy something something

[sleek vortex]

uncovr!

[fast meteor]

👀 https://discord.com/channels/1047197230748151888/1225545989989859409
👀👀 https://discord.com/channels/1047197230748151888/1267140551736557711
👀👀👀 https://www.perplexity.ai/search/to-response-to-query-create-pl-2MKinl_3TwihXocb73IUsw

[fickle sorrel]

not real X x x x

[fast meteor]

wut?

[sleek vortex]

wow lol

[fast meteor]

ideally at the end of each detailed paragraph, a tldr should be produced and prepend to the next query for more consistency

[fickle sorrel]

your username in pp. notreal

[glacial frost]

yo i'm looking for a mod for my project paying $600 a month

[fast meteor]

refined, more coherence https://www.perplexity.ai/search/to-response-to-query-create-pl-3mf5YSf3QDKpMosaFBBQ5Q

and characters descriptions
in fluent paragraphs (NO BULLET POINT STYLE) for a realistic fiction novel
make references to the previous replies

Dr. Armin Schuster, the BBK's president, in reply #4
"We've prepared for every possible scenario," boasted BBK President Armin Schuster in reply #7

[fickle sorrel]

pullet boint 💀

[fickle sorrel]

@fast meteorthx. do u know good llm controlling browser tools?

[jaunty dove]

Interesting feature coming soon: https://x.com/AravSrinivas/status/1817022053452755143

Aravind Srinivas (@AravSrinivas) on X

Query autocomplete is slowly rolling out to users with a Llama 3.1-based model.

Twitter · Jul 27, 2024, 2:19 AM

[torn sierra]

Have no idea what is interesting in this 🤣 I guess there are other issues to resolve

[fickle sorrel]

i mean sonnet 3.5 controlling a browser would be way better

[steel crater]

if i have 3 referrals is next month free or?

[lament spoke]

Perplexity next? :0
https://www.windowscentral.com/software-apps/openai-could-be-on-the-brink-of-bankruptcy-in-under-12-months-with-projections-of-dollar5-billion-in-losses

[fickle sorrel]

thats all irrelevant out of context taken news

[fleet lake]

Hello Everyone!
I hope everyone is doing well.
I'm new in this community and I recently purchased perplexity pro. But I have One question -
What are the limits in Claude 3 Opus AI Model?

[fading moth]

50 Opus messages per day.

[wispy summit]

They said Opus, Opus limit is 50 uses per 24 hours, 540 is for all other AI Models.

[fading moth]

My bad

[placid coyote]

the new big llama is so stubborn with asian languages... 😞 I tried very hard in this thread, but I gave up. it's just like talking to a real llama https://www.perplexity.ai/search/respond-in-very-pessimistic-an-JwtojmgNQWS6bxqh6LQyDQ

[sudden void]

hey hey, I've noticed that my threads history is always empty? is this some bug or a "feature"? 😄

my history worked properly and then suddenly I lost my threads

Also, I am using perplexity both on web and android, for example I start some thread on android and I want to switch to lap top web perplexity... but the thread from the android is not visible on the web app

[fleet lake]

Thank you for response!
You mean to say in Opus I can only do 50 massage per 24 hours and 540 is for all other AI Models or 540 massage for each AI Models?

[wispy summit]

540 for all other models total

[sleek vortex]

540 shared between all other models
opus 50

[austere kestrel]

isn't the index itself kinda already made up of cached versions of the pages its crawled? i think it is probably embedded / vectorised too, being part of a RAG system. combined, it is quite fast

[sleek vortex]

more or less yeah

well we don’t know specifically what but something like that

[austere kestrel]

I think the sonar online model model via the API vs via the web (without Pro enabled) are effectively the same; they are the two left here. They take about the same time, and the answers themselves are quite similar

The Pro Search I think just runs multiple queries against the index. Unless a URL is in the prompt, i don't think any crawling or scraping happens.
And with genspark (far right), you can see it doesn't have an index; instead, it is manually scraping and summarising each site - it takes much longer

[sleek vortex]

hmm… i think i’m quite fast for not having an index

[austere kestrel]

could try poe (if you're happy to paste in URLs manually) - it'll try to scrape each of them, and then pass the contents directly into the prompt as context

[austere kestrel]

yup i think so too :))

[sleek vortex]

there’s no cache on the scraping yet either

which is pretty crazy

later,i’m going to add an s3 cache and see if it’s faster or not but still its good right now

[sleek vortex]

some crawling might be happening still

especially for like document urls and etc

[austere kestrel]

yeah

thnk i said apart from when urls are given

i haven't seen it scrape and attach as files more than 4 provided URLs - but it's still a nice touch (esp in writing mode i'm finding)

[sleek vortex]

seems like it’s got stuck trying to scrape the llama paper or idk what

[austere kestrel]

is it meant to be able to handle PDFs?
(a large-ish one in this case too)

[sleek vortex]

not actually sure if i enabled pdf scraping

but the context length usually trims a source to 6k tokens or something, and the average query is around like 6-10k context?

might have to add an exception based on the number of sources, it would be nice if it could summarise this singular query

oh wow, i think we’ve hit around 120 “visits” in the last 24 hours

[austere kestrel]

yeah though i notice the URL is to download PDF; I think with Jina and other scrapers, it only works if the URL is for a PDF that is actually hosted/visible on a web page (which can be scraped) - if that makes senes

[sleek vortex]

well it reports 98 unique users

so give or take like a few repeat devices and browsers

90 users?!

not bad at all

[austere kestrel]

with poe, it fails (though it like >1min to process.. and 9mb is the actual size of the file when downloaded... so not sure actually)

[sleek vortex]

did we check if that link is IP locked

sometimes the case with CDNs

[austere kestrel]

yeah not sure. though that llama report as a PDF might just be too big... Here's a much smaller / lighter PDF, which is also a download link [https://nvidianews.nvidia.com/_gallery/download_pdf/669fc96b3d63326ffd76e5c1/]
seems to work

[sleek vortex]

i think i might be returning the PDF as raw utf8 instead of parsing it…

[alpine flicker]

anyone knows how to log in to the android app with manually providing code instead of clicking link?

[tame current]

I don't think you can

[alpine flicker]

That's bad. I cannot use the app.

[tame current]

which code do you mean

why not

[alpine flicker]

the one you need to provide during login.

when someone choose to use email address as login. perplexity always send the link to email that user needs to click. On Desktop you can enter it manually in webapp. In android app it is impossible.

[tame current]

oh that

hm

[sleek vortex]

add the website to homescreen then ig idk

[vivid forum]

is this good for coding?

like if i wanna learn coding

@sleek vortex

[sleek vortex]

well dont just use ai to learn coding

theres also lots of resources for any type of coding

youtube videos, etc

you can use ai to explain stuff you dont understand so on

but yeah its a good tool for that purpose

[meager hull]

are there any trials for pro, even just a 7 day?

[tame current]

(pro user) When is voice assistant coming to web its great its on mobile (android finally :P)

[vivid forum]

should i buy perplexity pro or claude pro?

[tame current]

with perplexity you get a lot more messages than claude and you get other LLMs too

[vivid forum]

but i dont think the models are the exact same maybe they dumbed it down?

[meager hull]

thats why im curious about a free trial even if its a 1 day trial tbh. i just want to compare the outputs

[alpine flicker]

It helps a bit but won't do work for you. At least not yet.

[dreamy prism]

Do PDF document upload have a page limit?

[placid coyote]

on pplx they have lower temperature (worse for creative writing) and smaller context (can "see" less). but they should not be dumber

[dreamy prism]

Or with pro can I upload let's say a 200 page document and ask questions about it?

[placid coyote]

you probably could upload it, but RAG will be used, so if it would be useful depends on your use case. was playing with it recently #💬│general message

[little tiger]

I have been with perplexity ai from last year and I like how detailed this AI is compared to just letting gpt-4o mini handle blogs alone and bought the pro subscription. Can this help with fact checking already written blogs and as as the new perplexity pages(if i can find it that is)?

[placid coyote]

the fact checking - probably. from my experience, it is quite good at verifying short texts (comments from social media). but it will depend on length of the blog posts (and probably complexity; there is a limit to searches even in pro search). you might have to split it into several threads or followups.
edit: tried giving it text of this article https://serokell.io/blog/haskell-in-production-fossa, this is the result https://www.perplexity.ai/search/analyze-and-verify-biggest-cla-lR2cAPV6QPuGSS2uPA44tg (it got converted to paste.txt)

[little tiger]

compared to 5 from the free version I have a whooping 600 when i paid up to cover more ground.

[placid coyote]

yep, a lot of uses per day. while the fact checking to be reliable would probably had to be done in a few parts (a bit of manual work), pplx is still quite powerful in this area in my opinion

I just wanted to set expectations - pplx may have issues with longer texts and for best results those texts should be split in parts. which is some work, though blog posts usually aren't that long

[tame current]

huh, I never thought of using it like that

[little tiger]

Compared to GPT 4o mini and gemini, honestly I needed Perplexity AI to backup all claims from the generated blog posts to avoid most of the AI hallucinations.

[tame current]

fwiw gpt4o (not mini) can search the web too but you have to prompt and it is not as good as perplexity

[placid coyote]

yeah, I really disliked that about chatgpt, that you have to prompt it or it will probably hallucinate or use outdated info. not sure how good the new searchgpt is, I guess it will be able to search more reliably, but I wonder if it can (or rather how well) also continue chatting in the thread

[tame current]

apparently it still hallucinates: https://www.theverge.com/2024/7/25/24206488/openais-searchgpt-demo-results-arent-actually-that-helpful

I didn't read the whole article though

[little tiger]

...I'll let those models stick to creativity from now on and NOT research/blog posts.

yeah I was guilty

[tame current]

perplexity is better, still worth double checking but it's pretty reliable

[little tiger]

Thanks to perplexity the time I have to fact check is a LOT less than before I can tell you that much.

[tame current]

yup, it also filters out those useless blogs

that don't actually contain what the title says

[nimble lance]

i find the default works the best in terms of getting most up to date code. is that what most folks use as well?

i use it to generate code utilzing the qwik framework. extremely greatful

it is able to figure things out that claude 3.5 seems a bit outdated on

[fast meteor]

#1240356137497530408 message here

tldr sonnet 3.5

[nimble lance]

Hero status. Many thanks

[winged moon]

Is anyone like me and sometimes want to have multiple versions of a thread so they can change up a prompt at the end or midway through?

[alpine flicker]

I chose perplexity for more scientific results. It is indeed a bit better when it comes to reliable data

hmm perplexity cannot plot a chart in the answer.

[little tiger]

My blog posts needs credibility and doing it alone with the other models will NOT help that cause as they hallucinate the answer. Thats why I can count on perplexity to check out the correctness of a blog post, and I must say its accurate in its answers.

[alpine flicker]

It still lacks a lot but yea for accuracy it is indeed nice.

it cannot plot charts for example 😄

but it gave me python code to plot the chart, better than nothing.

I wonder if it can give me code for jupyter

[tame current]

hey guys, anybody knows where I can send a mail to the Perplexity team?

Don't wanna tag anybody, is anyone from the team able to respond? 🙂

ah no matter, have a great week y'all

[fast meteor]

[e-mail]
[e-mail]
depends on your needs...

[fading moth]

Ye, depends on what/why you need~

[tame current]

I want to work for Perplexity AI lol

[wraith crow]

You can view open positions at Perplexity here: https://www.perplexity.ai/hub/careers

[fast meteor]

hmm, eu hq is in london, but not in barcelona/berlin/zurich 🤔

[shut plaza]

some of those positions can earn a ton of money jesus

well over $200k

[fast meteor]

expects crunch every week i guess 🤷‍♂️

[tame current]

yeah i need to be from SF, nah bruh. I'm a poor dev from a third world country

would've loved to make 200k a year for centering divs but oh welp 🤷‍♂️

[fast meteor]

for all those well paid jobs
i very like the Korean Language Customer Support (Remote Contract)

Remote 40 hour per week contract - $40/hr.
only 40 hour per week no crunch, can travel the world while doing the remote job, 228$ a day should cover the daily expenses
but i dont speak korean...
and it also requires
Based in Europe or North America 🤔

[fickle sorrel]

@fast meteormaster yoda, i beg your wiseness and experience, should i pick default for ai model ? which model is that even? i am subbed

[fast meteor]

#1240356137497530408 message
default might be haiku

[lapis spindle]

perplexity previous version was good we had Gemini pro and CHATGPT-4 and other amazing ai models but now it has CHATGPT-4o which is already free on CHATGPT free version

[tame current]

Yeah it's pretty alright

[little tiger]

Important; can perplexity AI deal with keyword research effectively or that's too much power for a simple task when prompting?

[sleek vortex]

give an example query

[little tiger]

for e.g. top docking stations. with the right prompting even GPT 4o gives a simlar but close query. For blog posts however the heavy power perplexity AI has for fact checking is justified.

[sleek vortex]

hmm

pplx gpt4o (right) prefaces it with some level of waffle, then lists like 7 docks

[little tiger]

I dunno the one on the left's website is but looks interesting and uses gpt 4o but it looks like both of them tackle the keyword research differently so it wont matter much

[sleek vortex]

left one is uncovr.app

but yeah seems like they both do okay

[little tiger]

right, then I stick to my original plan with perplexity AI pro and that's focused research and accurate blog posts.

[tame current]

Why does it look purple

[sleek vortex]

complexity extension

[tame current]

Can you link it I can't find it on Google

[sleek vortex]

#😎│cool-projects message

[sour valley]

@full gyro Wondering why we can't find the exact Claude 3 Opus daily limit usage on your FAQs official website.

I'm so confused as to what's the current official numbers for Pro, if someone can help out. I hear we get 600 Pro searches per day now (although their FAQ site says 300)? And as for Opus, it's 50 messages/credits per 24 hours (each credit restoring 24 hours after used)?

I appreciate y'all's help 🙏 !

https://www.perplexity.ai/hub/faq/what-is-perplexity-pro

[tame current]

Thank you

[sleek vortex]

540 or 600 (in ab testing?) pro searches per 24 hours (leaky bucket type ratelimit)
most likely 540

and then 50 opus per 24 hours

[sour valley]

@sleek vortex thanks for the clarification 😊

is this info pinned by any chance? i'm sure ppl will run into this same question

[tame current]

Skibidi

[fickle sorrel]

costs only 200 euro per 1 million token

[fickle sorrel]

https://media.discordapp.net/attachments/1160358610480660597/1267645283508945029/High_Performance_this_clever_evil_girl_just_burne.mp4?ex=66a98a2b&is=66a838ab&hm=d0381b78fc20a9543bab2ab0b9fa4b7838c53ba3264a425317df4eb7cfd0e4e6&

[tame current]

Why can't perplexity 'read/'view' youtube videos when linked anymore?

I was able to do it all day before now it says it can't

[tame current]

Just heard chat gpt launching it's own search gpt like perplexity.

Would this affect perplexity and it's fundings?

[fading moth]

No, because OpenAI will be bankrupt this time next year

:3

[tame current]

Lol

[blissful thicket]

i need help @grand mauve @toxic moss
my gmail account that was used to subscribed to perplexcity was banned by google for whatsoever reason

i can't change my email address on perplexcity and now i cannot login to perplexcity on my mobile app because it requires OTP and i cannot get the OTP

I paid 200 USD for perplexcity pro but now i cannot use it on other PCs/ Mobiles because Google screwed me royally.
I have no idea why. I did not create my google account with a bot. I tried to appeal to Google but to no avail

[dense bear]

If you had a gcp account you could try to contact their support, sometimes they can help you 🙂

[blissful thicket]

What is gcp?
I tried appealing already but to no avail I hope perplexity can help me by simply changing my email to another one

[dense bear]

Google cloud platform, their AWS alternative 😁

[dense bear]

I would contact their support directly.

[torn sierra]

All these tools are still imperfect. This also applies to Perplexity. It's interesting to see companies pointing out each other's mistakes on X. For example, felo.ai claims to be more accurate than Perplexity, while the CEO of Perplexity draws attention to a post where ChatGPT made mistakes (its standard search version, not the new one that's supposed to be better). I've also seen You.com comparing their results with Perplexity's and praising their own... https://www.theverge.com/2024/7/25/24206488/openais-searchgpt-demo-results-arent-actually-that-helpful

[torn sierra]

It won't go bankrupt at all. I've seen similar stories before. It's just media hype and nothing more. A lot of money is being poured into OpenAI. The news about bankruptcy is just media sensation 😃 but I guess you are just having a laugh wrtiting this 😎

[proper frigate]

Hi guys!

Quick q: The search results are from... well, which country? Or how does it work?

[tame current]

Or Google One alternatively. https://one.google.com

That's the only way that you can get direct support from Google, I think so or by GCP as you mentioned before.

I wouldn't rely on Google or any service, they can shut your account down whenever they feel like. https://youtu.be/dll-lsLdm28

But sorry for you, hope you can get it sorted. 💙

[haughty mist]

Hi

[slow light]

https://techcrunch.com/2024/07/30/perplexitys-plan-to-share-ad-revenue-with-outlets-cited-by-its-ai-chatbot/?guccounter=1

[austere kestrel]

i don't see why both can't be true:

• OpenAI (like all companies developing frontier LLMs afaik) is haemorrhaging money, both building and deployng the models
• OpenAI won't go bankrupt (investors will keep pouring money if the AI thesis/dream remains as promising as it currently still does) i.e. it can keep haemorrhaging colossal amounts of money without going bankrupt

[wise edge]

why Academic mode hidden the reference and paper urls ? i need the papers' url

[harsh stag]

hello beautiful people, im back. Thanks to @halcyon coral for his quick reply 😁

[hallow linden]

And just like that, perplexity is now the most ethical ai search engine in the world

[alpine flicker]

In good or bad way?

[serene sand]

https://x.com/aravsrinivas/status/1818279260517499062?s=61

Aravind Srinivas (@AravSrinivas) on X

Today, we're announcing the Perplexity Publishers Program. Our success relies on ensuring answers are factually grounded in high-quality sources of information. The most scalable and sustainable way to work with creators and publishers is to share our revenue and make our

Twitter · Jul 30, 2024, 1:35 PM

[fleet sandal]

question I saw that gpt4 has a kind of chat trained for a library that I am using, my question is can I through perplexity somehow access to that kind of chat?

[subtle ridge]

Now, we have perplexity api llama 3.1 which supports 128k context window, will we have that size on website?

[vital fossil]

As soon as I’m seeing ads as a Pro user, I won’t be a Pro user anymore.

[harsh verge]

I'm not sure this is entirely good news for the users.

[tame current]

you are asking if you can access perplexity from chatgpt ?

I don't think you can, but there are some other GPTs that have access to the web

[fast citrus]

I know claude 3.5 is good with web design with css and its artifacts

Does wrapping perplexity with claude allow all the benefits of artifacts?

[fiery dust]

I don't get why there's a publishers programm now? Scraping off these websites is totally still possible. Is this just a PR move after the backlash?

[woven cosmos]

On Perplexity, there is no artifact, but you can use the "Complexity" extension.

[junior blaze]

brands will pay perplexity to push related follow up questions at the bottom from what i understood

it's me or perplexity is kinda down right now ? i can access my previous chats but can't create any

[lost crater]

any issues happening on perplexity? But slow and unresponsive

[junior blaze]

yeah me too

[lost crater]

Yes. Started getting the issue right now

[junior blaze]

they're probably doing some things on the server side after the announcement

[hollow hearth]

Same here

[junior blaze]

personnaly i started getting it like 20 seconds before the announcement

[muted pine]

Pls make Perplexity the Best IA CODING !!

[tame current]

and voice assistant on pc not just mobile 😛

[violet inlet]

Yo is perplexity down right now?

[junior blaze]

yeah it is

[tame current]

was thinking that lol

[hollow hearth]

yes, some experiencing issues, likely linked to announcement

[woven cosmos]

Yes, it's super slow

[violet inlet]

What are y’all’s alternatives to perplexity?

[hollow hearth]

GPT

[tame current]

before it was down, if you turned pro mode off in writing mode would of it worked or needs web for all?

[tame current]

nah cap limit is a joke

[mental harness]

Oh good, I was scared that my chat bot storytelling had gotten to wild and my account deleted

[fast citrus]

is perplexity down rn

[hollow hearth]

Read ^^

[junior blaze]

chatgpt when you need web search, otherwise claude, and when you run out of claude credits, le chat mistral with mistral large

[rapid pebble]

phind

[tame current]

correct!

[quartz nova]

getting the same

[violet inlet]

Bet. I appreciate you

[tame current]

yes its down lol

[violet inlet]

I’ll check that out

[fast citrus]

lol everyone was discussing this in general

[tame current]

pablo escobar meme when perplexity is down xD

[warped marsh]

dafuq is this

[sage skiff]

hello everyone!

[gritty pasture]

yo, shouldn't we be prioritized for paying $20?

[violet inlet]

Lmao chill its just a bio

[still owl]

That's probably a joke

[tame current]

bro just coz we pay doesnt mean its gonna go down lol

[woven cosmos]

It's a general breakdown

[fast citrus]

how is wrapping a model around an LLM make it better than an LLM? As I think Perplexity is better than GPT-4o

[junior blaze]

https://knowyourmeme.com/memes/discord-user-is-a-suspected-terrorist-copypasta

[tame current]

i feel like gpt4-o is better then claude and others

[fast citrus]

how is that

[tame current]

more accurate

[woven cosmos]

For me claude 3.5 sonnet is better

[gritty pasture]

"We're getting a lot of questions right now and have reached our capacity."

This is an overload problem, not a general downtime

[tame current]

compare ur answers with the rewrite button

[woven cosmos]

Then it depends on the application

[violet inlet]

Best ai for coding questions?

[fast citrus]

I like that it has GPT bots but aside from that I don't know whether it hallucinates

[tame current]

large mistral shits on most but seems like gpt4-o still the top dog

[woven cosmos]

3.5 sonnet

[fast citrus]

last time I used GPT-4 it couldn't do Geometric math with tricky angle for grade 10s

[tame current]

3.5 not bad but doesnt dig deep enough

[fast citrus]

I was wondering whether not paying $20 until your subscription is auto cancelled (2 weeks i think) and then paying $20 helps you hack the monthly subscription thing

[trim mantle]

Surprised about the outage message. Makes me wonder about the sustainability of their new Pro search.

[woven cosmos]

It's just a breakdown, it happens to everyone

[tame current]

The value of perplexity has gone up massively in terms of features and quality compared to when it launched, and it remains the same price, so I don’t really see a point in complaining about what’s a temporary inconvenience

[trim mantle]

Could be a server outage. It does read like a budgeting issue; maybe they should change the message to better inform Pro users.

[fast meteor]

might be one dev pushed some bad codes...

[trim mantle]

Yeah, no issues here. Just curious about the underlying issue.

[still owl]

For real. We are using what people in the early 2000's would call science-fiction tier tech

[fast citrus]

Do you guys spend a lot of time prompt engineering?

[mental harness]

Like our concept of what "talking like a robot" means has completely transformed over the last 2 years.

[violet inlet]

This may be a stupid question. But do any other ai services compare to perplexity in the way it has complete access to the internet and sources everything it finds?

[still owl]

Copilot does at the moment

[mental harness]

Not that I've seen

[oblique trout]

I feel like gpt4o’s math is much better than the previous versions

[trim mantle]

I didn't use to, and thought it was a joke; but now I do after reading some basic AI prompting tips that actually work.

[oblique trout]

Could use it for my real analysis things

It sucks at something like computer graphics tho

[mental harness]

Yeah I got my perplexity bot working the way I like and it's much more personable to me

[fast citrus]

will have to try it out

[still owl]

Looks like servers are back?

[fast meteor]

is back

[oblique trout]

I kinda use sonnet and gpt4 interchangeably

[placid coyote]

hmm, is that big llama on pplx behaving okay? it seems to output much shorter responses (and ignores ai profile) compared to sonnet

[woven cosmos]

Claude 3.5 Sonnet boosted in perplexity 😍

[fast citrus]

how how did you do that

[woven cosmos]

It's an extension named Complexity

[oblique trout]

Omg look good

Ill try

[oblique trout]

So Claude wrote the entire thing? 💀

[woven cosmos]

In 2 tries because he had just written in my language so for the example I told him to rewrite.

[oblique trout]

It looks pretty good. One step closer to lose my job

Well tbf actually a lot of softeng things are not about coding and currently AI’s code quality is still not good enough

But its def better than what i saw before

[fast citrus]

How are the new Models Sonar Large and Llama 3.1 like

[oblique trout]

I used Claude 3 for similar things but it wasn’t this good

Tried llama 3.1

I dont think it is as good as gpt4o and Claude

Based on my tests

Perhaps need more tests tho

Sonar large isn’t as smart but is uncensored

[fast citrus]

did you build this extension can't find it online

[oblique trout]

🙏 what's this site

[woven cosmos]

https://www.perplexity.ai/page/Complexity-Perplexitys-New-yl0q3mHYQz6RhRyuvjvN4w

[fast meteor]

https://chat.lmsys.org/
leaderboard

[fast citrus]

Its only for beta testers

cool stuff

[woven cosmos]

Ask the admin in dm to be alpha tester

[fast citrus]

ok

[woven cosmos]

The extension is super cool, take the time to read the documentation and test it

[tame current]

should have windows program not third party since ios and android do :/

[placid coyote]

sonnet 3.5 got quite good for these kinds of tiny projects and even flashy one-prompt pages

[oblique trout]

yeah

[placid coyote]

saw some controversy on twitterx that it can reproduce in quite high accuracy (very similar looking result) some nice looking not entirely trivial shader from a tutorial 😅

[oblique trout]

i kinda wanna test sonnet 3.5 on complex industrial projects tbh

the issue with ai generated code was

the code was usually not really maintainable

i prettu much just used ai as an advanced google

[placid coyote]

yeah, you have to steer it a lot. and have good tooling around (some code graph rag, user giving it specific files etc)

[oblique trout]

did u try sending AI something like

UML

[woven cosmos]

Good job

[oblique trout]

and ask it to implement the code

[placid coyote]

I wouldn't dare to 😅

[oblique trout]

i've only tried asking gpt4 to write SQL queries based on my database schema graph

last time i tried it it was not really capable of writing code for complex things

[placid coyote]

technically most of ailin code developed sonnet (small project, around 5k LoC TypeScript). but in small parts, gradually and I was making bigger decisions. and, well, most of it is front-end, so a lot of css and with styles I had to do a lot of manual changes (for "starting" styles, eg of a new component, it is good though). started using cursor ide and it's pretty nice. in there opus is 10 daily uses, but a lot of times I end up wasting 3 prompts and 5-10 minutes with opus not getting what I asked for, while sonnec usually handles such cases on a first prompt

not sure why, opus on pplx seemed to handle code better in its free period, but maybe I was giving it so much less surrounding code, context?

[tame current]

does beta perplexity extension not work now?

[woven cosmos]

What is the beta extension of perplexity ?

[oblique trout]

looks good then

maybe i can test it on front end as well, nowadays i'm doing things like backend also some low level stuff

for front end iirc ai was pretty good at writing boilerplates

i use react a lot it was good for doing something like writing redux slices etc.

for backend or just OOP programs in general i feel like

it's good for writing code snippets

but u can't really rely on it

yeah as u said u need to steer it

for low level things ai still sucks, but my things are perhaps too low level cuz sometimes i need to write asm directly

[placid coyote]

for front end iirc ai was pretty good at writing boilerplates
for boring repeating code spread in several files, it is amazing. like adding an option to config (so types and default value) and in config modal adding a toggle for it, and in simpler cases (existing functionality on/off), it fairly reliable does it in one short prompt when "linked" relevant files

[toxic ruin]

Will paid users be advertised to? Wrt #📢│announcements message

[still owl]

https://tenor.com/view/second-futurama-scruffy-gif-20187509

[white raven]

For what it's worth, I really hoped there would be a little more push back against Forbes and exposed the fact that they sell guest posts, backlinks, profiles, AND author pages. If you ever see someone say "I got featured in Forbes", chances are they paid around $500 for that. Making 90% of the content on Forbes not worth paying a single penny for. It's also an extremely slippery slope. 85%+ of the internet never sees a single human eye. Where do we draw the line? Should the sports blog that only has 40 users/month deserve to get paid as well? Just my two cents that no one asked for and something to think about!

[urban ermine]

hi

[toxic ruin]

Everybody that wrote content that is returned by our queries deserves to be paid for their time effort and content. I actually really like that because it incentivises the creation of quality content and then it's perplexities problem to pick the right content.

However, as a paying customer I absolutely expect to not be advertised to. That's a fairly standard expectation of any online service. If that will not be true for perplexity then that is troubling. I'm the customer and not the product.

Am I making these comments in the right place? Is there a support channel maybe... looking

[viral current]

So wordpress.com is part of the publishers deal. Does that mean they will just hand over my blogs to you?

[torn sierra]

Still not resolved. No reaction…. https://discord.com/channels/1047197230748151888/1266708591763521626

[azure mountain]

Any update on if paid uses will have ads? At this point the silence is starting to feel like a yes.

[tame current]

Skibidi

[shell jasper]

Hi please your help cause perplexity app is not working

[tranquil orchid]

Hey @shell jasper!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[south kindle]

https://arxiv.org/abs/2407.04620

[trail ivy]

I'm completely unable to use perplexity labs — I keep getting ERR_NAME_NOT_RESOLVED, anyone else have this issue?

[tame current]

Vpn or dns

[trail ivy]

I flushed by DNS, restarted laptop, not using VPN

the page loads but it won't respond to anything and the error is in console

Normal perplexity works though!

do you think it has something to do with me being in Hong Kong right now

[toxic ruin]

Asking here https://discord.com/channels/1047197230748151888/1267965574944788633

[trail ivy]

actually Labs isn't working in Singapore, either. Are you guys able to use Labs?

[tame current]

Tbh perplexities literal upside is just their search

[charred flicker]

Yes, but you have the same opt out option for your sites content that was already in place from WordPress.com

[remote sinew]

Hey all we’re having some issues connecting to the api from zapier. Looks to be formatted correctly etc. anyone have this setup and working?

[trail ivy]

It was good to be able to use the llama LLM in Hong Kong without being restricted like other AI services are

[brittle garnet]

Sonet large 👀

[sage lily]

Wait what where?

Do you mean Sonar large?

[terse knot]

Does Perplexity restrict the number of searches I can perform?

[livid kite]

I was thinking about subscribing, but not if Perplexity is going to push ads and brands even on the paid tiers.

[vital fossil]

I hope they give us some clearity on that soon, silence is never a good sign. Good thing I didn't get a year subscription anyway.

[gusty sand]

Just updated my app and noticed this.. Still waiting for confirmation if these will be gone if I upgrade to pro?

[torn sierra]

Ads are in a free version. I am pro and there’s no ads. You may also try felo.ai. It’s free, no limits and no ads 😉

[torn sierra]

From what I read: „Starting later this quarter, Perplexity will let brands buy “related follow-up questions” that will appear below the initial answer to a user query. The questions will be clearly labeled as sponsored, Perplexity says. The brand will pay Perplexity a certain (undisclosed) fee for every thousand user views of the sponsored question.” (https://www.fastcompany.com/91163900/perplexity-selling-ads-ai-search)

[austere kestrel]

i could be mistaken, but i'm not sure that they're paid ads - i think they're more like fancy citations for products that can bought online that are mentioned in a response

they seem to be generated with Pro enabled too
(i'm not really sure though - i don't use perplexity for shopping.. just playing around)

i don't it's any different to something like this - meant to be helpful ig

[serene sand]

if we don’t see any scammy links, helpful it is

[livid kite]

i think also perplexity would be the perfect platform to "sneak in" persuasive sponsored content into the generated text output.

[heady perch]

why perplexity does not support zip files like GPT does ? im not able to upload my src code file

i just bought the subscription

[tired quartz]

That would absolutely be a horrible idea. They could insert ads everywhere, but if they include ads in the generative responses, I would abandon it immediately. That would transforms pplx from a "helpful" platform to a "harmful" platform.

[austere kestrel]

i think the idea is to somehow insert the ads into the 'suggested queries' which appear at the end of responses

which tbh doesn't particularly bother me - so long as advertisements don't have any influence on the actual response (or the sources from which it is derived), I can deal with (ignore) monetisation efforts on the periphery

[heady perch]

how to use this

[austere kestrel]

https://www.perplexity.ai/page/Complexity-Perplexitys-New-yl0q3mHYQz6RhRyuvjvN4w#1c2552be-080e-440f-8bfd-e1d31fbd2aa8

[dense kraken]

Is it down, or am I restricted?

[vital fossil]

Works fine for me

[grim trellis]

Agree! Also not sure if I misread this but if you are letting advertisers pay you to alter the recommended questions (and therefore the “thought path” curiosity leads you down), you are essentially letting advertisers pay to manipulate users thought process.

Now one could argue advertising does this inherently, but I think doing it via recommended questions is a dangerous / scary way to do that bc it is so hidden and psychologically manipulative

[timber slate]

Are ads coming to pro tier as well?

If so game changer and will probably stop using the platform - its entire use case is it’s direct and ad free so you can get information concisely and quickly imo - ads are the antithesis of that

[tired quartz]

I rarely check and use the related questions so it's fine.

[prime wind]

hi i just joined perplexity discord. i want to ask how to create a chart like in the picture. even if i write the prompt like in the video example it still doesn't create a line chart

[trail ivy]

How can I share code with Labs without it breaking bc it thinks I'm trying to inject code or something

[sullen mural]

u need to use pro search i think

[sullen mural]

see in #1224467082079371363 or ask in #⚡│ask-community

[fast meteor]

i think @tepid portal has this exactly issue, might be geo based?

[tepid portal]

Idk, it still doesn't work in my region

[opaque stump]

Getting triggered by my router

[fast meteor]

malicious site 🤣

[vapid onyx]

Now that’s funny

[toxic ruin]

https://discord.com/channels/1047197230748151888/1267965574944788633 vote/ watch here

[fast meteor]

a months old reply said it certainly not affecting pro
but things may/could change

[toxic ruin]

Can you link to that?

[compact bough]

Wait. Does wordpress partnership mean individual bloggers and site owners the use wordpress or what exactly?

[trim mantle]

Do I need to enable this? I don't see this view with the pro version?

Th artifacts view that is

[icy cloak]

is there a way to export the entirety of a created 'Page'? Including sources

[tame current]

Share the link, right?

[main bay]

Is this URL to view limit no longer working for everyone, or it's only me that can't view?

https://www.perplexity.ai/p/api/v1/user/settings

[wanton torrent]

Why did the Perplexity app access my contacts on my iPhone?

[harsh stag]

it's broken for everyone

[icy cloak]

not to share the link but to export the text and sources in a word doc for example. Or other markdown format

[tame current]

Truly IMPOSSIBLE. Impossible. Never going to happen. Never. Impossible.

[fossil sentinel]

Could anyone let me know if drive uploads work for them in aistudio? I've been having issues trying to upload any pdf over 100/200k tokens today

[fossil sentinel]

'Failed to count tokens for this attachment' 😔

[placid coyote]

that's a custom project on top of perplexity, https://gitlab.com/monnef/ailin (mostly a userscript). though if you are not a dev and don't want other features, you probably should rather try the Complexity extension https://www.perplexity.ai/page/Complexity-Perplexitys-New-yl0q3mHYQz6RhRyuvjvN4w (much easier to set up, should support the html embed as well)

[main trout]

Hey guys, I just got here. I am in love with perplexity. Changed the game for me.

I am a plaintiff in an extremely important lawsuit and I need highly precise yet complex search. Most LLMs just hallucinate fake cases. Perplexity helped me find super relevant cases from the internet and then crunches thay in proper LLM fashion.

Does anyone know how I can apply RAG to it? Say I have 100k documents in discovery and I want the AI to be able to search through it all and crunch the data for me.

Thanks in advance. I should get some rest. I've been so giddy and excited just playing with it

Oh and any other advice regarding optimized use cases for legal/lawsuits.

[crystal elbow]

Dude can someone tell me how I can paste text as TEXT and not as a attachment ? It's really so annoying, why is that even a thing? I'll press the "attachment" thing if I want to attach a file.

[crystal warren]

I believe it is due to the number of characters of your pasted text. I've experienced it as well and I've just dealt with it. Just remember to remove it from subsequent messages if you go off topic.

[chilly nova]

did also your perplexity freeze?

it's not working now

it hangs forever

[sleek vortex]

try reload window once

[chilly nova]

tried already 3 times

it doesn't generate the answer, completely frozen

[sleek vortex]

hmm… try clear application storage in dev tools (F12) > application tab > storage > tick them all > clear

then reload again

[chilly nova]

nah, it's just that particular thread

[sleek vortex]

oh what

no real clue then

perplexity just being bugged

[chilly nova]

nice... now it froze in another thread

this is so dumb

wtf it started now after 1 minute of hanging

this is crazy

[wraith crow]

Because you gave perplexity access to your contacts. Your contacts are used when you click on the person icon in the top right, to share the app with your contacts.

[livid aurora]

I found multiple OpenSource Alternative to Perplexity. If you want the list Ping me

[glacial anvil]

Perplexica🤔 ？

[livid aurora]

there are more

[tame current]

I do

I know of some, like morphic.sh and devv.ai (oh wait, that one's not open source)

[livid aurora]

Yeah and a few more.

Not sure what the difference is between them and Perplexity

[serene sand]

#💗│sharing message ☕

[final ruin]

hey @signal hamlet and @frank rivet I am Aron.

We are creating a new tool and we use Open AI and Perplexity APIs.
We are having a hard time getting PP to provide us real URLs and information.

Example:
We might ask the AI to provide a list of 5 courses at a local university
that might fit the interests of a student.
It sends me 5 made up courses and the URLs to each course are faked.
We can't get PP to provide real URLs.
We have it crawling the web ok -- just not producing real URLs
HELP please. this is urgent for us to roll out our update.
Thanks!
Aron

[sleek vortex]

uncovr :)

[tired quartz]

This page not works for me anymore.
https://www.perplexity.ai/p/api/v1/user/settings

Why?

[placid coyote]

hard to tell why (new api version or they didnt want users to know limits?). see https://discord.com/channels/1047197230748151888/1268412632306810900

[tired quartz]

Just changed?

[tired quartz]

I cannot get access to the link.

[placid coyote]

oh, it is in a pro-feedback, sorry. but rest assured others have reported it, now at 5 pluses