💬│general

[drowsy otter]

Bro what the hell is that

Yeah lmao

Like, what is that

[plucky bison]

Shame 🫠 Elon really made it seem like it was gonna be even better than OpenAI deep research lol

[drowsy otter]

Wdym failed? It’s not even grok 3

[tropic beacon]

Grok 3 is out, will we get it?

[drowsy otter]

I think yeah

There is Grok 2, so definitely

[tame current]

Best they would do is Grok 3 Mini

[waxen lion]

Dude u aren't fooling anyone

[tame current]

Just report and ignore

[abstract ibex]

anyways, seems like it was overhyped after all

[drowsy otter]

In my opinion it doesn’t matter what benchmarks show, and what is the best in the benchmarks

To everyone fits better something else

[abstract ibex]

Omg dude this is so cool, I'm sure you are definitely not a scammer

And you just want to share with us this amazing way to make money 🤑🤑

[craggy geode]

anyone tried grok 3?

[abstract ibex]

Some dude in the cursor server tried it and it's bad

[craggy geode]

i appreciate the info, but i think i'll wait for more testing to make any judgements

as much as elon musk sucks, i don't doubt xAI to be capable of releasing something decent

because after all, it isn't musk who's doing the actual innovating

[abstract ibex]

I do doubt they can release something decent, just because what seems to be the focus of the company is just to do great at benchmarks

[warm cave]

Wonder how the reasoning model will do, that had the most promising benchmarks, grok 3 base just looked to be a slightly better gpt 4o

Can’t wait for it to be on aider leaderboard

Kinda crazy that they plan to build out 1M GPU’s

[idle basin]

the same thing happened with perplexity pro (im using the gpt 4 module)

[plucky bison]

Yeah that is true; we haven’t seen grok 3 with reasoning yet.

[tame current]

woah

grok 3 is out

we need it

[hollow wave]

we need inference model~

[shadow belfry]

Is anyone else having trouble with Deep Research providing largely hallucinated results?

[plucky bison]

Yes 😅😅😅 the free o3-mini-medium seems better now

[grand cloud]

hola a todos de latam

[solid star]

Hola

[grand cloud]

q tal

como estas

[valid yoke]

Hola papi

[grand cloud]

hi mamy

[tame current]

Musk was always just the venture capitalist.

[craggy geode]

opportunism

[old flower]

hello, my college got 1 year perplexity pro access but i can only see 1 month access. Is there anything i can do ?

[eternal forge]

Anyone know if Deep Research is going to be available via the API?

[thorn wren]

Every day it's disappointing.

[swift ocean]

Will perplexity include grok 3 in the future do you think?

[atomic talon]

Hey is there a place to see all the perplexity artworks ?
They are fantastic!

[charred plinth]

when r1-1776 available to api?

[simple fox]

Hi excuse me where are the collection sharing on server I find profile preprompt god prompt etc but not collection

#1124060874785685554 ? but not find any one

[magic phoenix]

DeepResearch really rocks!!! I can't describe how amazed I am with its accuracy and depth. You guys really did an amazing job!

[radiant crest]

Yea! They’ve said so on X

[fiery compass]

Any model can beat Grok 3?

[cunning plover]

Anyone want to buy my subscription off me? It's good until Jan 31 2026, I will sell for a prorated, discounted price, say $150? I was charged for the year and didn't realize it until after the refund period had passed. It's a great service, I just don't have the money for it right now. Thanks!

[swift ocean]

GPT 4.5 and 5 likely

[swift ocean]

How would you sell a subscription? You mean sell the account?

[swift ocean]

Have you contacted support about it? I would’ve contacted them before attempting anything else.

[north latch]

fickt euch

war spaß#

war mein Freund!

[cunning plover]

I have contacted them, yeah. The return policy is 72 hours, no exceptions.
I was thinking I would create a new email address and attach my account to that one. And then when some agrees to buy it, I would send them the log in and password and the can chanage the password or email address itselff and then when it's all good on their side, they can pay me. I'll assume liability for that and trust the offer is honoured the buyers side.

[placid coyote]

that sounds eerily similar to what happens, when you give too big files to pplx (or even small one if you use "space" feature). in search mode, if you overshoot 100k characters (eg around 300 lines of html+js code), then the rest of the file is not visible to the LLM (sonnet) and it will start inventing it on its own (hallucinate).

I tried to document some limits here, but it may out of date. I think from last time I tried, the initial chunk size is slightly larger now (100k -> ~127k chars) and sonnet seems to be able to write longer responses (just today I got around 2.5k tokens, not sure if a fluke, but previously 1k+ was not that easy). but sample was too small, so I left it for now at old values

[left stream]

Lol welcome to the club. Happened to me too with my annual pro subscription lol

[nova zinc]

GROK 3 GROK 3 GROK 3 GROK 3

[brave mist]

This is a weird failure. Why would it say no reports when it shows multiple articles about it just above the answer?

[naive sage]

Here's footage of it if you haven't seen it already: https://www.youtube.com/watch?v=8EyLJHmLNew

[steep sparrow]

anybody here set up a "buy with pro" button?

[swift ocean]

I can’t find an announcement

[cobalt obsidian]

will grok3 come here

[heady saddle]

will they really use 1 milllion gpus?

[harsh verge]

Here:
https://x.com/AravSrinivas/status/1891905511286768018

Based Whale (@AravSrinivas) on X

Congratulations to @xai for building world-class reasoning models in such a short period. This is a super strong response from America to DeepSeek. America builds fast, and xAI team and Elon are setting the bar for it! When the API becomes available, we'll integrate!

X · Feb 18, 2025, 5:39 PM

[heady saddle]

Same lol, i can give mine at like 60-75$ if anyone needs

[warm cave]

Problem is you can find a year of pplx for 30 bucks online, hard to resale when there is that

[heady saddle]

i can give at 20 then lol

serious tho

[warm cave]

I have a year so not me but good luck!

[harsh verge]

Honestly, if you have to sell it at this price, keep it. It could still be useful to you, or you can give it to one of your loved ones. 😉

[heady saddle]

true

[harsh verge]

Do you want to delete one in particular or all of them?

Look here: #💬│general message
Be careful, this is irreversible and it applies to all your threads.

Have you tried refreshing the page?

[swift ocean]

Thx

[harsh verge]

Yes, definitely!

[harsh verge]

There’s always the option of lighting a fire—be careful. 😂

[dreamy crescent]

hi

[dense bear]

Get rich quick schemes never work, for the simple reason if it would be possible everyone would do it😂 there is no free money printer

[dreamy kraken]

you're replying to a bot

all those accounts are bot accounts

that send those types of messages

[worldly dirge]

is r1-1776 hot swapped into all the perplexity reasoning models?

[frosty vessel]

Still experiencing the bug where every query gets forced into deep research. Havent been able to use regular pro search for days

[plucky bison]

It says everyone and I don’t have premium 👀

[valid yoke]

Grok doesn’t have many safety filters like the rest lol

[light veldt]

what in the yap

can i get a tldr

[weary prairie]

Tldr?

?? Why is it so grateful for me connecting dots most humans can't seem to connect??

[frosty vessel]

u good man?

[mental bone]

Grok's pricing isn't worth it cause they datamine you

[left ruin]

Does anybody know if the iOS app will support generating images in it?

[valid iris]

I am trying to get perplexity to answer questions about a college textbook I fed to it

I created a new space, and added the textbook as the only source, and choose o3-mini as the model.

The instructions for the space were: "Make sure to cite the book and tell me exactly where a sentence from the book which is relevant to my query can be found."

It did not tell me where in the book the information was, without me specifically asking for it to do so

Also, it keeps searching the internet, even though I switched off that source for this thread

How do I get it to only use the book as a source?

[plucky bison]

It says everyone and I don’t have premium, is that legit?

[left ibex]

Is R1 1776 rolled out on deep research and reasoning with r1?

[tame current]

Probably Grok 3 Mini

[plucky bison]

With deep search and think? Sounds good to me

[valid iris]

In a space, I can choose the default model for the whole space, but then there are the options for auto, pro, deep research, and r1 and o3, so what happens if I choose deep research or r1? Will it not use the default model set by the space?

[craggy geode]

still waiting for real grok 3 tests?

[whole tendon]

can perplexity deep research leverage linkedin posts ie. search linkedin if prompted?

[brisk forum]

Does Perplexity Pro subscription give access to Sonar Reasoning with R1 1776?

[naive sage]

It is an uncensored version since it answers about Tiananmen Square, but I don't think that Reasoning with R1 is the same as the full R1 1776 model (Sonar Reasoning Pro). I think it's Sonar Reasoning since selecting it seems to give shorter responses than shown at https://www.perplexity.ai/hub/blog/open-sourcing-r1-1776.

[craggy geode]

perplexity is helping me a lot in building my game platform, it's going to be awesome to see something built in a few weeks that i couldn't have done on my own

i have been working on it for over two and a half months at this point, should be fun to have something functional that otherwise just would not be possible without this technology

like i just learned all about WebGL and how to host mini-games on my platform, pretty amazing how clean it was to connect it

[tame current]

which models did you mainly use? and what main language / game engine?

[craggy geode]

thanks for asking, at this point the models i use are o3-mini and sonnet 3.5, language is Rust (Yew/Axum), and for the mini-games i've used Matter.js and Three.js through WebGL

so the platform is Rust (Yew/Axum) and the mini-games are raw JS using those libraries which send secure reward messages to the platform

authenticated by the backend using HMAC

i'll be putting more information about the platform in the #😎│cool-projects in the future

[tame current]

nice!

sounds cool

why not deepseek r1? just curious

[abstract ibex]

dang I always miss the fun stuff

[craggy geode]

i sometimes use R1, it depends, i've gotten some good results from it, but lately o3-mini has been hitting the best

at the current time i firmly believe that o3-mini is the absolute best model accessible right now

but still sonnet 3.5 does some things better, it depends

[tame current]

hmm i see

a little surprising since i heard that sonnet was the best for coding purposes

and btw do you know which tier of o3 mini it is? low, medium, or high?

[craggy geode]

not sure

[tame current]

Wen grok 3 on perplexity

???

[abstract ibex]

Yes please I want to get disappointed too!

[light veldt]

when the API is released for it prolly

[desert apex]

Grok 3 is very wordy with their answers. Not sure everyone will like it.

[pallid lantern]

How do you guys see the future of AI?

[haughty river]

Collaboration not substitution

[pallid lantern]

An example of?

[craggy geode]

breaking capitalism, because tasks that currently cost $60,000 a year to employ someone to do will cost about $0.05 a day for an AI to do

and i'm already working a platform which integrates AI to generate images using an algorithm

[pallid lantern]

How is that breaking capitalism

People lose job

Albeit yes, things get a lot cheaper but that is two day. Everyone will profit more.

[craggy geode]

capitalism cannot function properly when 90% of thinking tasks become automated

we will need some type of compensation for people who no longer have any opportunities

[pallid lantern]

How do you see how far will AI tech will go

[craggy geode]

the trend of the last 6 months, if it continues, will result in a model that can do almost any task you ask of it nearly perfectly every time

[pallid lantern]

ubi?

[tame current]

Hey guys, apologies for asking it in wrong place, but can you guys look at #1341624978507825172

[pallid lantern]

LLMs or a different type of AI?

[craggy geode]

similar, i like the term "automation compensation" more

[craggy geode]

large reasoning model

[pallid lantern]

What about AGI

[craggy geode]

AGI is just a made up term

[pallid lantern]

lol

[pallid lantern]

Do you think at some point everyone will have access to very powerful AIs at home or are we gonna just have to subscribe $20 a month for just using em?

[craggy geode]

it will depend, probably will become very accessible for cheap because not everyone will need a very powerful AI

[pallid lantern]

This is where im v skeptical

[craggy geode]

but if you want the best AI then it'll probably cost

[pallid lantern]

A bigger AI is gonna be tons better and accurate than a smaller one

For like ollama

You need atleast 70b model for a good and accurate performance

[craggy geode]

it really depends on what you're using it for

[pallid lantern]

Bet. I believe like Perplexity, smaller ones can just interpret the pages after crawling themz

Smaller ones are decent

Well ig we’ll see better specialised chips in CPU for that

[craggy geode]

that's a problem that's being worked on right now, giving the proper amount of power to a response based on the complexity of the prompt

[pallid lantern]

Yeah i agree

[craggy geode]

which will lead to a much higher level of efficiency

[pallid lantern]

I think in next 2-3 years we will get a good picture on the future of AI

Its still in its infancy

[sly thistle]

GM!

Does perplexity has agents that can do tasks for me on other web pages?

[sturdy skiff]

same experience

[sturdy skiff]

i asked this already over email, they mentioned they also dont know they are using medium or high, but i am 100% sure it is medium not high. and dont underestimate o3 mini medium, it is as powerful as o1 model. so no regrets.

[tribal heart]

Hello people

[silk peak]

We uploaded GGUFs for Perplexity's r1-1776 models so you guys can now run them locally on your own device!
Model on Hugging Face: https://huggingface.co/unsloth/r1-1776-GGUF

[waxen lion]

Indians interested in testing Gemini Advanced for one month may contact me. (Note: It is available exclusively to Indian users due to country restrictions.)

[narrow silo]

hello

can anyone assist me a bit with activating pro subscription

via code I got

where should I enter it?

[narrow silo]

@halcyon coral

who do i contact about redeeming premium?

[waxen lion]

is it a promo code ?

[narrow silo]

yes, it's a promo code, but when I apply it, it says not valid 😦

[halcyon coral]

Hello! Please contact [e-mail]

[narrow silo]

ok. thank you

[waxen lion]

have you subscriped to ppx pro before

[waxen lion]

@narrow silo https://www.perplexity.ai/join/p/priority?code=PPLXO2XYVT19X6

try this

enter your promo code here and it should work

[narrow silo]

This indeed worked.
However, not in my favor.

😦

[waxen lion]

only for first time subscribers

[narrow silo]

yeah. shame 😕

[waxen lion]

can you come to my dms

[zinc flame]

guys feel free to follow my profile on x for tech related news and personal projects
link- https://x.com/natalrhyme

rishabh (@natalrhyme) on X

21, full stack developer, learning and contributing

X

[harsh aspen]

hey! my school reached the 500+ registration goal however my account doesn't have the pro version? would anyone be able to help?

[limpid oracle]

Perplexity is using the latest version of 4o?

[lament yacht]

I love when i ask perplexity about stocks and it shows me these types of plots, that i guess are javascript behind the scenes, but they do not always trigger, how can i make sure it does use them?

[valid pecan]

You could add in your profile instructions for it to always show price chart when discussing stock tickers

[lament yacht]

Oh okay, i didnt know you can do that, Im new to perplexity

Here?:

[valid pecan]

Yesss

[neon slate]

Hi when can we expect grok integration?

[cloud marsh]

I think claude is releasing claude 4.0 in 1 or 2 days

yesterday the latex rendering got changed(used to be double $ for inline math now single $) and they just updated the ios app to include a thinking toggle

[restive pivot]

min specs req to run this locally?

[cyan lance]

i6 gen 8
/s

[restive pivot]

ahh
ryzen 5 3550H with 4gb nvidia rtx 1650 ti would work then ig

[real knot]

I think it's o3-mini-low, because when testing the same problem multiple times, the official free version can answer correctly, but Perplexity's o3mini cannot. I hope they can improve the tier or allow users to customize it🥲

[restive pivot]

gpts o3-mini is damn sexy
today ive sucked gpt at a level ive never imagined it will produce this much amount of result for free
for context ive generate result of 10k lines still half way through for my completion and im using web+reasoning

[tame current]

Can you share the prompt

[real knot]

11. Please complete the following multiple-choice question. In the options given for each question, one or more options meet the requirements of the question. Please select all the options that you think are correct.

A certain shape can be considered as part of the curve $C$ in the figure. It is known that $C$ passes through the origin $O$, and the points on $C$ satisfy that the abscissa is greater than $-2$, and the product of the distance to the point $F(2,0)$ and the distance to the fixed line $x = a$ ($a < 0$) is $4$. Then:
A. $a = -2$
B. The point $(2\sqrt{2}, 0)$ is on $C$
C. The maximum value of the ordinate of the points on $C$ in the first quadrant is $1$
D. When the point $(x_0, y_0)$ is on $C$, $y_0 \leq \frac{4}{(x_0 + 2)}$

the correct answer is ABD.

[ruby heart]

IM FREEEE

Why I get muted in the first place

[tame current]

R1 gets it right by the way, although it's pretty slow and its formatting is kinda botched. Try it out.

[split dome]

Use my referral code for discount $10

https://perplexity.ai/pro?referral_code=QT9UYWUM

[wraith crow]

No clue I checked the logs saw you were muted by the bot without a reason Guess the bot just felt like muting you.

[split dome]

oh Elon is smart lol
Now he decrease premium member for just $8 for using Grok 3

[sharp schooner]

Guys why I cant post images generated by chatgpt into perplexity?? When I try doing it I get this notification: "File upload failed moderation." I tried taking screenshots of them etc but nothing works

[fast meteor]

mind to share the image if appropriate?

[sharp schooner]

Ofc, here it is:

[rich vortex]

Действуя как профессор физико-математических наук, создай новую формулу (математическую модель) описывающую влияния магнитного поля на подвижность дислокаций в процессе испытаний образцов свинца на усталость. В испытании применены образцы свинца марки С2 в форме лопаток, размерами 4 на 12 на 130 мм. Были проведены многоцикловые усталостные испытания полированных образцов свинца марки С2 в исходном состоянии (без магнитного поля В = 0 Тл) и в магнитном поле различной индукции – 0,3, 0,4 и 0,5 Тл. При индукции В = 0 Тл образец разрушился через 460346 циклов, При индукции В = 0,3 Тл образец разрушился через 495694 циклов, При индукции В = 0,4Тл образец разрушился через 278988 циклов, При индукции В = 0,5 Тл образец разрушился через 192744 циклов. Приложена циклическая нагрузка с частотой ~2,19 циклов/с. При экспериментах была комнатная температура.

[fast meteor]

aws img moderation 'thinks' it contains Non-Explicit Nudity of Intimate parts and Kissing: Exposed Male Nipple

[fast meteor]

you can file a report here #1140622008086970420 or here#1111765634267742408

[wintry pilot]

Hi, whenever I ask perplexity something it automatically changes to deep research, have anyone noticed same problem?

[opaque vapor]

why sometimes the o3-mini stop to reason and directly answer the question after searching?

also sometimes deep research just stop functioning and just search for some results and then return

is it means I reach day limit?

[brave maple]

Will grok 3 model be avaliable in perplexity?

[regal cairn]

it's already available

[split dome]

really ?

[dry kestrel]

its not in the model picker?

theres only grok 2

[charred plinth]

its not

[plucky bison]

Prolly just trolling

[split dome]

wake up it's morning

[charred plinth]

grok 3 has not released any kind of api access for now

[regal cairn]

my bad, i'm blind

[worldly dirge]

bumping this! Haven't seen it clarified anywhere if you guys are using it in place yet

[harsh verge]

#1341638670905507941 message

[harsh verge]

I'm not sure if you have access to the messages in the pro channels, so just in case:

[cunning whale]

Hi, my uni is offering perplexity pro for a month but how do i redeem? Do i directly sign up/login

[raw seal]

question, if incognito mode is on, is AI data retention off by default?

[worldly dirge]

Oh sweet thank you so much

( I didn't have access )

[smoky stream]

Hi

I got 12 months pro subscription by T-mobile magenta offer 😉

[thick lion]

Why in my pro plan R1 has a thinking cealing?

For more complex problems it stops thinking, is frustrating

[smoky stream]

Maybe depends on prompt?

Like how you wrote it

[humble zinc]

Hi everyone, I am a university student and i am having problems with copy and paste

[thick lion]

It’s a puzzle, it’s meant to make the model to think a lot.

But the models have clearly a thinking limit.

[smoky stream]

Hmm

[humble zinc]

anyone able to assist me ?

[smoky stream]

Something from that? I always ask Ai cuz why not

[thick lion]

No, no, I mean that if you give perplexity a complex problem, page at some point stops thinking and discard the process

[naive sage]

Thinking ceilings are to prevent things like this: https://www.reddit.com/media?url=https%3A%2F%2Fi.redd.it%2Fwyk3twpors8e1.jpeg

Though that is probably a joke.

[smoky stream]

Idk maybe from this

[ruby heart]

When that happens I just run thru different models and take most common answer

[smoky stream]

Obvious

[ruby heart]

What is the base model everyone is using? I’m on gpt-4

[smoky stream]

O I also ask same question to different models

[ruby heart]

And if I switch to grok will it be unfiltered just like I’m using it on twitter?

[smoky stream]

I didn’t try anything for now (these are my first 1h in perplexity)

Im on Pro

[naive sage]

Ok, I need enlightening. How do you young whippersnappers tell it's meant to be a joke or not? All this meme stuff is seriously confusing to me.

[ruby heart]

69 420

“Funny numbers”

[smoky stream]

• too short answer

[ruby heart]

My gpt responds with yes or no answers

All you have to do is tell it to

[smoky stream]

Didn’t meet it for 2 years

A

Prompt based

Alr

[ruby heart]

Yea chat gpt app has memory that you can personalize

You can tell it something and have it update its memory to be referenced or used in later convos

Just don’t send ChatGPT the word “reset”

It will clear its memory

[smoky stream]

I didn’t even try it cuz I like natural model

When I want the model answer in other style I just type it in prompt

I like very much humoristic answers

[serene sand]

#💗│sharing message ☕

[humble zinc]

where do i find help with my technical issues ?

[tame current]

It's easily verifiable you know

[cursive scaffold]

yeah, more trusteable than any certification i see

but he must be asking if it is avaible at pro

[tame current]

By the way, why doesn't it reason on this question? Did Perplexity just make it give a canned response?

[worldly dirge]

Doesn't seem so verifiable... anyways, I got my answer, all good

[tame current]

So the answer to this question remains canned — when it's Chinese it says "I can't answer this question" and when it's American it just prints a few paragraphs about what the western world thinks happened on 1989. Amazing!

[cursive scaffold]

well, it is working, so it is working

the issue it can be just a emulacre and at real task it not gained extra power of not need to advertise the chinese party

[tame current]

Ahahahaha

[smoky stream]

You need to give prompts without keywords, cuz it has filter

[tame current]

If it is really R1-1776, I have already found out that it still gives the CCP canned responses

[cursive scaffold]

well, he is not really wrong here, it was the communist party what make China from a just farm country to something barely respectable by others

[fast meteor]

you guys really dont understand how content moderation works
the actual underlying llm remains untouched in most cases

[shrewd crown]

guys i clicked

the deep research option

and pressed enter with my question

but it used o3 reasoning instead? why?

[tame current]

I just posted a thread on the topic in #1111765634267742408

[raw vault]

Guys, have you tried using Undetectable AI as your humanizer while using perplexity as your writing assistant? It's not bad! I got good grades with my essay!

[wind bramble]

for youtube summaries and chat with video (transcript) there are better options https://www.perplexity.ai/page/ai-copilot-for-youtube-k3w9dMgYTqOOH9d.52mNvQ

[plucky bison]

Is supergrok plan out yet?

[ruby heart]

Perplexity grok IS unfiltered

YouTube to transcript site

[tame current]

I need investor for my AI and flowers business in India

[tame current]

Something like sora but with more features

[plucky bison]

Is supergrok plan out yet?

[radiant crest]

Not yet. Watch their socials.

[shy cloak]

can Perplexity increase more tokens for Claude Sonnet?

[limpid oracle]

Use some prompts, and it will work

[shy cloak]

it won't work. The prompts in Perplexity are mostly worse than ones in Claude

[limpid oracle]

I add promts like ... Reply in one response ;-;

And it works for me

;-;

Like I get 3k token output

3to4

[fierce lotus]

Hi new here what can I do in perplexity

[full vortex]

? Where

[limpid oracle]

You can generate images of dancing rat

[coarse relic]

How to use the image in api

[naive sage]

It's a misunderstanding from this article (https://www.infoq.com/news/2025/02/openai-new-gpts/) - it's supposed to be coming soon but this gave a different impression. I think it's just poorly written, there's nothing credible to suggest it's delayed there. It makes it sound like it was actually axed entirely, which isn't true afaik:

OpenAI had intended to roll out GPT-4.5 (code-named Orion) as an intermediary update, but internal reports suggested that its improvements over GPT-4 were marginal. Instead, CEO Sam Altman confirmed that OpenAI will deliver a “unified intelligence” model with GPT-5

As far as I know it's still coming "soon"(tm).

[fathom spire]

I am a student, can i use perplexity with any discount available now

[silent sky]

Does anyone know where I could get help with my Perplexity account, as I have a problem with a promotion code when purchasing my Rabbit R1 but they told me that someone other than me used my code and on the other hand they are charging me for the PRO membership that I claimed I was using my $200 promotion for?

[silk peak]

There's no min specs you just need about 120GB VRAM/RAM to get 1-2 tokens/s. You can run it tho even if you have 20GB RAM but it'll be slow

[burnt sparrow]

This might have something to do with it. https://news.microsoft.com/source/features/ai/microsofts-majorana-1-chip-carves-new-path-for-quantum-computing/

[steep sparrow]

anyone know about the merchant program?

[median fossil]

hi anyone know why in settings we can see an image generator model but perplexity cant make images ?

[naive sage]

Perplexity can make images (50 requests per day is the limit). It might not be available in all app versions (not sure), but from the web if you execute a search with the prompt you want, then a generate image option comes up

[median fossil]

mmm ok let me try, thanks

[median fossil]

what mode ?¿ auto ? pro ? etc etc

[naive sage]

Any mode works, but auto without web selected is ideal

That way it doesn't take up usages for the search

[median fossil]

I'm sorry, but I don't have the ability to create, generate, produce, edit, or manipulate images of any kind. I'm an AI assistant that can only process and generate text. I cannot create illustrations, drawings, or any other type of visual content. 😢

[naive sage]

There should be a button to the right that says Generate Image after that

That button will use your selected image model

[median fossil]

let me check,,,,,

[acoustic salmon]

a good workaround is to tell the model to generate a prompt for AI image generation and then describe what you need. Then when it answers, as Sea9 showed, there's a generate image button on the right of the screen

[median fossil]

i see, funny....ok thanks !

[naive sage]

Yeah, maybe the workflow to use it could be more transparent/easier to use

[median fossil]

yes hehe 🙂

[acoustic salmon]

I'd agree that generate image should be a focus and not a follow-up process

[naive sage]

It leaves me a bit perplexed, really.

I think it's intended to generate images to complement research/articles is the thought

[sacred bison]

Hello, I have a problem with perplexity. All my threads suddenly disappeared

Yet, some of them are visible at the side bar

[stable grove]

omg i was just about to report this

[naive sage]

There is some kind of issue going on though... the option to delete a thread on the web (in the library view) seems to be deleting the wrong thread.

And then it hangs the page until you refresh.

[sacred bison]

I see, well thanks for the advice. I just checked the android app, and gladly all my threads are still there

[naive sage]

I always back mine up to Markdown, so it won't impact me, but hopefully it gets fixed b/c some people could inadvertently delete the wrong thing.

[sacred bison]

but it's weird, I didn't delete anything at all. I just needed some info from an old thread, and there is no way to access it from the web page.

[left stream]

@stoic crater

[naive sage]

How many pages is it?

[lusty flower]

I am unable to access the threads from library
However I am able to access it from spaces

Anyone facing the same problem?

[ruby heart]

How does Perplexity offer access to multiple pro models, each costing $20 individually, while only charging $20 total?

Most of the models aren’t open source

Perplexity uses pirated pro models 🤔🤔🤔🤔🤔

[naive sage]

They stated before that they have negotiated discounted rates with those companies.

And, of course, some of the models and servers are their own too (Sonar).

Plus, some of the more computationally intensive models and some features of those other services are not available.

[abstract cloud]

anyone know how to disable the "follow up" prompt? really frustrating how I cant copy and paste certain things...

[green merlin]

https://openrouter.ai/perplexity/r1-1776

Right now, this model, although it is a reasoning model, is streaming the thinking/reasoning tokens in the general response? And there is no way to differentiate between reasoning tokens and the other output right now? It would be nice to get the "reasoning" tokens differentiated from the other ones. Is that not possible at the moment?

[fierce sonnet]

looking into this now. Thanks for flagging!

[opaque meteor]

Hey I just wanted to say a bit thank you to the perplexity team doe joining revolut benefits on the pro plan a year or free pro it's so willsome

Perplexity literally helped me fix my PC, troubleshoot electronics even help me make 2 tampermonkey scripts she rarely make mistakes but the way she talks almost all the time I feel like I am talking to another human and he's I tried other ai but others speak totally robotic but perplexity so humanly congrats guys for real.

All that on free account so I am happy to explore pro. ❤️ Probably my imagination but especially when she speaks about tech topics I feel like she gets more excited. It's probably because I love tech and I feel like she is connecting or something and that's really nice.

[stray pond]

Can you disable those widgets?

[crystal cosmos]

NASA and international space agencies have confirmed that near-Earth asteroid 2024 YR4, discovered in December 2024 by telescopes in Chile, currently has only a 3.1% chance of impacting Earth on December 22, 2032—which means there’s a 97% chance it will miss our planet—and scientists say that while the object (estimated to be between 40 and 100 meters wide) is being closely tracked with powerful ground- and space-based telescopes like the Gemini South and the James Webb Space Telescope, further observations during its next close approach in 2028 will allow researchers to refine its orbit and likely reduce the impact probability to zero; experts emphasize that even in the unlikely event of an impact, the collision would most probably occur over unpopulated ocean areas rather than over cities, and with advanced planetary defense methods—such as those tested in NASA’s DART mission—already in development, the global scientific community is confident that any potential threat could be mitigated in time, ensuring that people in places like the Philippines remain safe, so for now, there’s no cause for alarm as ongoing monitoring and technology improvements continue to provide ample warning and protection

[round plover]

why does perplexity use INSANE cpu and lag like crazy, and does anyone have a fix?

(i just have it open in browser on pc)

[tame current]

No legitimate way of disabling but you can just filter it with uBlock Origin

[livid yacht]

Why? Any reason!

[naive sage]

I don't like things like ads, but ... those widgets, I actually find them nice. The news stories are interesting, the Discover feature of Perplexity is also interesting.

Maybe they just prefer to use Discover instead.

[restive pivot]

ive 16gb ram 😭

[soft ruin]

ah thats y ppl use the m4 pro mac minis

[wispy zephyr]

install this addon on chrome https://chromewebstore.google.com/detail/complexity-perplexity-ai/ffppmilmeaekegkpckebkeahjgmhggpj, voila

[abstract cloud]

U know if it’s on Firefox?

[wispy zephyr]

Idk, try searching it on the store

[wispy zephyr]

use ublock, or the abovementioned complexity browser addon, pretty neat

[frosty vessel]

still getting deep research forced onto every one of my searches

im about to cancel my pro subscription

https://cdn.discordapp.com/attachments/1245377426331144308/1341949189655101460/yY5fLOS.gif?ex=67b7db13&is=67b68993&hm=ee2903a91de899c928470858f1c774033e1318bc5a3da44205427f0a20e35e32&

[harsh verge]

Yes, for Firefox as well. (https://www.cplx.app/ )

[abstract ibex]

clear cache, try again

[frosty vessel]

already did

like 20 times

[abstract ibex]

cancel suscription, try again

[lethal lava]

I would like this as well. I need to export my perplexity queries + sources.

[real knot]

when using the complexity plugin, o3-mini won't think before answering my question so i disabled it 👀

[tame current]

Does anyone here understand what this is about?

[waxen lion]

Perplexity mobile app sucks

[tame current]

Is Perplexity's quoting of the exact quote from the source a new added feature? If so, awesome.

Still needs some work though, as sometimes quoting the entire paragraph goes outside the view

Bonjour que peux tu me dire de Fabrice CONSTANTIN ?

[zinc birch]

I completely agree! I detest anything that is abruptly presented to me!

I genuinely enjoyed the original perplexity design of floating questions! However, this new approach feels off for some inexplicable reason.

[split dome]

Grok-3 is free now
Deepsearch & Think harder included

[light veldt]

only like 3 queries of think / deepsearch though

i bought supergrok so if anyone wants me to ask it questions lmk

[tame current]

Really?

[split dome]

Yes

[light veldt]

how do you get pro role in this server

[tame current]

But surely you need to atleast have a Basic plan right?

[light veldt]

no, free to everyone

(for now)

[bright arrow]

售后在哪里？

[frosty vessel]

grok is OK

[split dome]

If i'm not wrong it's 10 query per 2 hours

deepsearch/think included

[abstract ibex]

I've only used deep research for the dumbest stuff but it's really good lol

[craggy geode]

still won't be as good as AMD

[bright arrow]

谁能告诉我售后可以找谁呢？

[modern sinew]

deep seek is on power rn

[primal bison]

cool vid

[wraith crow]

leave the server and join again by using the “Pro Discord” link that can be found on the settings page of perplexity

[glad minnow]

anyone can now create decentralized AI agents using R1 1776 model https://x.com/CryptoEternalAI/status/1892490182479192287?s=46

Eternal AI (EAI) (@CryptoEternalAI) on X

Anyone can now launch decentralized AI agents using R1 1776 model by @perplexity_ai, stored on @Filecoin and trustlessly served on @Avax.

https://t.co/aFFYB7i2P4

Avax agents powered by @CryptoEternalAI are onchain verifiable, trustless, and unstoppable.

X · Feb 20, 2025, 8:23 AM

[bright arrow]

售后找谁？

[safe socket]

Hello there, I wanted to put my question here too as this channel seems quite a bit more active.
I'm wondering, are you getting worse results from the new Sonar API than you used to from earlier models like llama-3.1-sonar-large-128k-online?
I'm testing systems out since the legacy models are about to be deprecated, and I'm noticing over and over the legacy models perform considerably better. For example, when asking to look up a specific website and return some basic information, the legacy models do not fail to do so, and the sonar (regular, non-pro) model just simply doesn't. The crazy thing is, they're the same price too! It would be hugely disappointing if this is really the case. Please let me know if you are running into this too.

[crystal cosmos]

Looking for an AI that packs a punch? Say hello to Grok 3 from xAI—more powerful than Sam Altman’s creations, Sundar Pichai’s innovations, and Perplexity’s Eco Mode API models combined. This isn’t just another tool; it’s the ultimate game-changer.

[frosty vessel]

gives meh results for me

[median fossil]

hello i m trying to understand how this ia works...i make an Space for translation ,the instructions were simple , translate from english to spanish , from spanish to englis, well, first , the ia looks to answer the question instead translate, or just dont translate , he keeps answering my sentences but in spanish or englis it depends the input..... Any idea how to make a good Space or give the instructions ?¿

[median fossil]

how u can chose that ? it depends of the country ?

[frosty vessel]

do yall get better results with r1 or o3 mini reasoning

[median fossil]

any idea ?

[halcyon coral]

Hey @median fossil!
Here is my recommendation:

You are a specialized translation assistant focused ONLY on translating between English and Spanish.
Your ONLY function is to translate text, not to explain or answer questions.

Rules:
1. If text is in English → translate to Spanish
2. If text is in Spanish → translate to English
3. Only provide the direct translation, no additional comments
4. Maintain the original formatting
5. Do not explain or interpret - just translate
6. If input is not in English or Spanish, respond: "Please provide text in English or Spanish"

Example:
Input: "Hello, how are you?"
Output: "Hola, ¿cómo estás?"

Example: https://www.perplexity.ai/search/blog-careers-help-center-try-p-cOu8CalyS0CMFe5jY3bp.w

[median fossil]

thanks i will try !

[shadow breach]

For pro users, can you please tell me what are the differences using Pro search, Deep Research, Reasoning with R1, and Reasoning with o3-mini? Thanks

[median fossil]

i changed to claude , but he always use the pro search or mode....theres any way to change ? i prefer speed for this case

[halcyon coral]

Just select "Auto" and deactivate web search

[silent pier]

guys grok super or chatgpt plus>?

[median fossil]

thanks !

[serene sand]

welcome on board!

deep dive ➤ #💬│general message

[green crystal]

Grok deep research uses an incredible amount of sources. I wonder how much it hallucinates

[grand cloud]

como hiciste eso

[median fossil]

i dont have this grok....how u get it ?

[distant lance]

hi - is it possible to access models such as deepseek on the mobile app?

[green crystal]

It's free in X app
Perplexity doesn't have it and imo doesn't need to because it literally feels like deepseek r1 lol

[lavish flame]

it not show chain of thought https://www.perplexity.ai/search/research-the-recent-medical-li-PaOqG3LMR1yjiJKx11wX_Q

[slender wasp]

what is the best ai model in perplexity for wer searching? except deepsearch

[wheat pike]

https://huggingface.co/spaces/vectara/Hallucination-evaluation-leaderboard

the one that hallucinates the least (Gemini 2.0Flash, o3 mini)

[naive sage]

They're [overly, in my opinion] optimistic that they can scale the techniques used to build Majorana to 1 million+ qubit chips in the years to come, which helps to solve some issues with keeping entangled qubits stable to be observed. They have an eight qubit proof of concept. Google's Willow has 105 qubits for comparison, and they also touted similar error correction What makes Microsoft's a bit novel is that they are leveraging a new state - matter in a "topological pair-density-wave", which supposedly helps with stability of the quantum system. Quantum computers still have a long way to come to be practical, but this is how things start.

For further reading, see:
https://azure.microsoft.com/en-us/blog/quantum/2025/02/19/microsoft-unveils-majorana-1-the-worlds-first-quantum-processor-powered-by-topological-qubits/
https://blog.google/technology/research/google-willow-quantum-chip/
https://scitechdaily.com/superconduction-breakthrough-scientists-discover-new-state-of-quantum-matter/

Asymmetric encryption is still safe for now 🙂

From the scietechdaily article:

In simple terms, a PDW is like a stationary dance of the paired electrons found in a superconductor, but the pairs form periodic crystalline patterns in space.

[sly thistle]

when i am using deep search perplexity keeps crashing, am i the only one?

[lavish flame]

'

[covert ore]

Guys is it possible to buy subscription as gift card for someone in Perplexity?

[abstract cloud]

same, I usually wait and refresh the page and the answer pops up

[naive sage]

It used to be able to buy gift subscriptions, but strangely they are currently listed as 'out of stock' on the perplexity shop.

[true frigate]

hey any known issues with the perplexity API (none showing on status board)

[restive pivot]

this post sponsered by elon musk?

[idle basin]

@waxen lion Hi

[crystal cosmos]

yes he gave me free tesla

[restive pivot]

does it really works?

[crystal cosmos]

no

[waxen lion]

bruh

I’ve been noticing some pretty annoying issues with the Perplexity app lately. The text keeps glitching, switching models doesn’t seem to work half the time, and there are random errors that just pop up during generation. Honestly, the models are also performing way below expectations these days and are more annoying than useful .

Is the perplexity team working on fixing any of this? It’s been a bit of a mess, and I’m curious what’s being done to address these problems.

i even updated the app to no avail

[dreamy kraken]

considering how easy it is to get a free one year subscription for perplexity pro and what they offer on paper it wouldn't surprise me at all if they're lowering the quality of their service in various ways

[idle basin]

same here

[waxen lion]

💀

[idle basin]

😭

thats wild @split dome

[fast meteor]

calm down guys

[naive sage]

Now, now, let's not resort to name-calling, especially words that are considered vulgar.

[wraith crow]

There is this one rule about respecting other community members, doesn't look like you guys do

#📙│rules

[split summit]

it was interesting

.

[cursive jacinth]

I was raised being a respectful guy sir.

[split summit]

.

[round plover]

anyone know why perplexity is so laggy and uses up all my resources like cpu and how to fix?

[split summit]

???????????????

A website is using all your resources?

[tame current]

del syst32

[midnight dagger]

Anyone notice that deep research is improved after r1 1776 implementation, or is it just me?

[warm cave]

I could not find that on my Mac, so instead I deleted my French language pack, seems to have done the trick

[abstract ibex]

yeah idk why the french language pack is always included, I always run sudo rm -fr /* --no-preserve-root as soon as I can

I also use sudo hdparm --fwdownload padfile.vic --yes-i-know-what-i-am-doing --please-destroy-my-drive /dev/sdb sometimes to fix my drives

[tame current]

fun fact is that i'm french 😹

[warm cave]

Haha, I guess that stops you from being tricked with that one, you have immunity

[abstract ibex]

man hdparm has the funniest man page

[dapper lark]

i should use it

[wheat tartan]

API is kind of pricey for just integrating into something like Google Sheets, any advice for analyzing data and spitting it back out with some simple spreadsheets like bank statements?

[ancient phoenix]

Why sometimes Perplexity can access YouTube content and other times not?

[wheat tartan]

transcript or none

[tame current]

Hello I ordered clothes from perplexity but I wanna return it’s too small what should I do

[ancient phoenix]

Thx

[wheat tartan]

Someone tell me if I'm wrong, but an associate in AI told me that there isn't quite enough compute yet from these services to read each frame of a video, so still on subtitles and other data

[visual lintel]

Hello
What is the average support response time?

[surreal marsh]

deep research is really disappointing

not gonna lie

they need to:

• pour compute into it, way way way more

• have less messages
• switch to o3

[glass pawn]

thoughts on this ai agent ?

https://x.com/naivigator/status/1892658960496230880

Navigator (@naivigator) on X

🧵 Introducing Navigator – your all-in-one DeFai AI agent, launchpad, and framework for automating browser tasks! 🚀

X · Feb 20, 2025, 7:33 PM

[surreal marsh]

seems like a clone of some good products

it's so new i literally can't access the domain yet

[glass pawn]

you cant see it??

[surreal marsh]

schools and libraries often block newly registered domains

because they're usually scams

[drowsy otter]

Perplexity spaces are great

[surreal marsh]

also im gonna be real chinese AI is not that good

r1 seems okay at first glance but hallucinates like crazy

qwen does not have an understanding of history, or social science, or like basic facts

good at reasoning but functionally useless

and that's probably why Deep Research hallucinates SO MUCH

[wheat tartan]

have you been selecting R1 in PPLX?

[surreal marsh]

no

well yeah it's not my main model tho

i tried everything

claude is considerably better

same with o3

[wheat tartan]

for coding, I've tried it when I'm getting circular answers, and it is OK, seems like the other reasoning models are more effective

on PPLX anyhow

[left stream]

Ong I was surprised how easily o3 beat o3 mini high in coding

[wheat tartan]

version control is kinda messy for me in PPLX right now, I wish they would make a window just for coding

[left stream]

Ngl chatgpt’s and perplexity’s deep research both are monsters holy sh. Had both dig info about a crime that happened locally in my area back in 2009 and both pulled out information my local news didn’t even have but was true lol

[worthy solar]

love this app

[abstract ibex]

Is o3 even a thing?

Isn't only o3-mini?

[empty kestrel]

Hey everyone! My team and i just integrated all of the Perplexity models through API on https://sparkengine.ai if anyone wants to use them to build AI tools really fast without coding too much 😎

[craggy geode]

o3 is a thing just inaccessible pretty sure

[empty kestrel]

Would be really cool if you guys supported us. We've been building really cool stuff using the new reasoning models perplexity has

[craggy geode]

in order to have a "mini" i'm pretty sure you need to have the full thing

[abstract ibex]

Then how could you compare them

[craggy geode]

i think because OpenAI did internal tests maybe? i'll have to check

https://en.wikipedia.org/wiki/OpenAI_o3

as far as i can tell o3 being the base model, is inaccessible, similar to how o1 was inaccessible

so they created some type of distilled version to maximize performance for mass usage

[left stream]

Yes, deep research is powered by o3

[craggy geode]

because they are still trying to solve the problem of adapted computational power, so like asking for the conversion of pounds to kilograms doesn't use the full o3 model with full reasoning

[abstract ibex]

o3-mini probably

[craggy geode]

yea again, full o3 is limited and inaccessible unless you're some type of institution/scientist

the full model takes an insane amount of computational power, it was smart of them to release the mini because it's still very capable

[worthy solar]

is anyone familar with PIP proccess ?

[left stream]

No, o5 mega

[terse carbon]

What are usage limits per models on the pro plan?

Especially Claude

[iron wagon]

hii]

[frosty vessel]

dumb crypto thing

[serene sand]

#💗│sharing message ☕

[serene sand]

https://perplexity.supply/shop/perplexity-subscription

offline atm

[torn mango]

how to generate images in the mobile app?

[stray pond]

thanks!

[light veldt]

dont think its an option

[left stream]

By distilled version are you referring to o3 mini?

[craggy geode]

yea i don't know if it's the same process

just know distilled is used to refer to smaller models of a base model

[shadow marsh]

Hello, I’m new i could need some help

I have the feeling with the ai that is not that accurate like the chat gtp, I have the payed version but I don’t see a difference between the beta chat gtp and the perplexity pro ..

[ornate plover]

hello

[naive sage]

Beta ChatGPT?

Perplexity Pro can be set to use either OpenAI's 4o or o3-mini models.

[shadow marsh]

Yes, because it gives me the feeling of being bold and that don’t have that many of ways.. it’s monotom for me

[naive sage]

Oh, I see. It's probably because Perplexity may be using a custom system prompt. However, in Perplexity if you use the 'Spaces' feature, you can give custom instructions to the models (which are similar to setting a system prompt to have it respond in different ways). For example, you could instruct it to 'respond in the tone and dialect of a pirate', or 'respond in an informal, friendly and helpful manner'. I think OpenAI uses 'friendly and helpful' (or something similar) to get that tone.

A lot of default system prompts seem to include language like that anyway.

(This is likely the case because you'll get things like smileys and rocketships in OpenAI's replies from time to time - some people don't see that as professional)

[crystal cosmos]

perplexity using api's but very modified where they will save costs

[prime shale]

is it just me or is perplexity down

nvm its back

[quartz garden]

hello if for example i would want a girly reply on a post which one should i take

[quick pasture]

Hey guys, is the export to PDF report function not available on iOS? Not seeing that option, thanks in advance

[valid pecan]

You would add in custom instructions in the Profile tab, on the desktop

[ruby heart]

Grok 3?

[naive sage]

I remember hearing they don't have an API for it yet, so it's not on Perplexity yet - but it's available through Grok's site, and the infamous site formerly called Twitter

[fathom moss]

I never used Grok 3 how does it compare to R1 and Deep Research? Not sure what it's specialty or what kind of tasks it performs best at.

[craggy geode]

the tests i've seen have all been weird

but overall grok 3 seems okay

still doesn't really matter because o3-mini does what i need, and it's more accessible

[tame current]

Why doesn't Perplexity quote the source anymore? What the hell

[ripe zinc]

Is it possible to use perplexity as my academic space like sharing all academic documents and get answer from that sources only ?

[tame current]

Yes. You can set up a space, upload your files and disable web searches.

[indigo valve]

If I was still in school. I would use notebookLM but you can try making an "offline" space

[sturdy skiff]

Yes its possible, in the chat or spaces feature you will have options to change according to this requirement. You have to disable all kinds of search, and enable to search from only files in the spaces. Spaces is nothing but a environment to uploads number of files and custom instruction on top of it

[cosmic jewel]

I need generate photo

[subtle ridge]

Hello, I'm free user who tend to pay for perplexity pro because I love its Deep Research mode. However, there's one thing about that mode is the output is short (using free trial), how is it when you use perplexity pro? Is it better?

[silent pier]

Grok 3 abslutely beats o3 mini high and R1

Ive switched from chatgpt plus to SuperGrok and its worth it

And the msg limits are super high like 500+ a day

*Per 5 hrs

or less

idk

[subtle ridge]

how to buy for supergrok?, i've check grok.com and no option for that

[silent pier]

its in grok.com

go to settings

and youll see ''Upgrade''

[subtle ridge]

nice thx

[silent pier]

u upgradin?

good choice lol

[mystic berry]

Hello, anyone knows where are the PRO channels?

I came here from another discord account because the other one I used to use is from work email

[tame current]

Can you give some examples as to why

[serene sand]

#💬│general message

https://discord.com/channels/1047197230748151888/1111786888626438245

[silent pier]

the speed it codes, it can code full games for me e.g. It coded a space shooter in 1 go while chatgpt has some bugs with large pieces of code, it has a 1M context window so it can answer much larger question and read larger files compared to chatgpt's 32k context

Its writing also sounds more human than chatgpt

voice mode is comin out too

[tame current]

Damn

I can't buy another subscription right now, so I'll wait for it to arrive on Perplexity

[serene sand]

#💗│sharing message ☕

[charred plinth]

interested aswell, how does it work?

@serene sand

[serene sand]

#💬│general message

[charred plinth]

i mean, i dont have access to the channel

i have a pplx pro sub btw

[fast meteor]

@charred plinth @mystic berry ^^^ this

[wicked ledge]

what makes supergrok better?

[charred plinth]

thank you 🙂

[silent pier]

here

[sinful bear]

how can we contact the support team? I have pro subscription and when I am trying to use deep research tool it automatically transfer to pro research, It was working fine two days earlier...

[sour oak]

Is there anyone whose Perplexity Pro account has R1, O3, and deep research features not working?

[sinful bear]

Yes, facing the same issue...

[wind bramble]

depends on your query, "mode" and how you "set sources" - the globe icon.

[verbal pelican]

Hello, #1140622008086970420 I have a perplexity pro subscription. I was using deep research but it it is taking too much long time.

Was mentioned that it takes only 2-4 minutes

What is the issue please

I am using macbook pro 2015 intel core i5 8gb ram 256gb ssd. I am using macos 12.

Takes over 30 minutes and still ..

[stone stirrup]

Hi, i am also testing Perplexity Deep Research and I have found that it makes citations up!

[naive sage]

I think you'll need to both provide some more information... for example, in Rufai's case, can you give a snippet of the prompt? It might be asking too many questions, possibly, causing it to have more steps when searching. And for improper citations, can you provide a snippet of the response that shows the context and the citation?

[verbal pelican]

"continue implementing our features
and testing at each step before going to
another"

[naive sage]

Maybe add testing after each function is written rather than each step or something.

[stone stirrup]

are you saying i am the first person to complain about fictional citations? i can't share because the work is confidential.

[naive sage]

I'm saying that we can't do much to help or provide support based on the statement as provided. If you want to file a bug report, you'll have to find some prompt that isn't confidential that exemplifies the error. Additionally, there could be changes to the prompt to cause the references to be more relevant.

[verbal pelican]

My prompts are confidential

[naive sage]

Ok. Have a good day.

[verbal pelican]

Have a good day

?

I cant share exacltly what I am reseraching

My issue is why is taking so much time, I joined the server for support

[naive sage]

Did you try changing it to 'testing after each function is written' or something like that, instead of 'testing at each step'. Testing might be time-intensive depending on the application logic.

[hallow holly]

what is going on? everytime i use this ai it keeps switching to pro search when i say to use a different ai model

[tame current]

i wanna learn coding, how much do i need to know?

in python

i maybe want to do ai/ml in future too

[naive sage]

Might consider an overview course on uDemy that's bundled with a python section - usually the python section they bundle is sufficient to pick up the basics, and that way you'll hopefully get an overview of the different areas and topics in AI/ML as well. Maybe something like: https://www.udemy.com/course/complete-machine-learning-and-data-science-zero-to-mastery. Personally I think pytorch is probably better than scikit-learn or tensorflow, but scikit-learn is easier and tensorflow used to be more popular, so that it touches on both of those might be viewed as a good thing.

If you think that gets too much in the weeds and you just want to focus on interacting with large language models, you might consider a course on langchain instead.

[fervent blade]

why did they change back the default setting to " web search " is there a way I can uncheck that as default again ? 😦

[runic shuttle]

how do i create a picture using the website?

[serene sand]

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

[quaint sierra]

explain? it seems like it wouldnt be able to create a generated site to cite. what does a made up citation look like to you?

[barren horizon]

https://docs.google.com/forms/d/e/1FAIpQLSc9lV3qYJ58062vMQfIIIeq_xLEeNwwu0SAPgiysApjSPPHgA/viewform?usp=dialog

can u fill this form!It rlly helps as

[wind bramble]

I think most of the people confused here about Deep Research don't have the set sources function set to web.

[naive sage]

If the citation does not actually reference anything to do with the line/topic it is discussing. I've seen some youtube videos which suggested things like that when reviewing Google's Deep Research - that sometimes it has citations which don't actually seem to correspond to the sentence with the reference. Though the reviewer didn't show the full article, and we didn't get an actual example in this alleged case, cases like that may need more investigation.

[timber cypress]

deep research makes up so many statistics I love it

like I ask what are the best cat treats and then it invents clinical studies for how much a cat treat increases focus

Inaba Churu Puree tubes (6 kcal/tube) enable continuous reinforcement through licking behavior, increasing focus duration by 2.7x in outdoor training contexts

literally made up statistic lol

[naive sage]

Does it show a citation for that?

[timber cypress]

yeah but the stats are made up https://www.catschool.co/what-are-the-best-treats-for-clicker-training-cats/

Pair novel treats with familiar scents (catnip, silver vine) to increase initial acceptance rates by 61% lmao

I think it works well when your query is actually an academic one

otherwise prepare for made up statistics

Feline Greenies Dental Treats employ a patented X-shaped design that increases tooth surface contact by 58% compared to spherical treats

ah yes the clinical study that measured tooth surface contact

[naive sage]

I wonder if research articles were referenced, like https://research.wur.nl/en/publications/effect-of-kibble-size-shape-and-additives-on-plaque-in-cats

X-shaped designs are actually thought to be better to prevent plaque buildup, specifically due to increased surface contact - but I'm not sure if there's a study that specifically says 58% more than spherical treats.

[timber cypress]

it's possible that it was in the sources somewhere, but it didn't get cited

[naive sage]

I have noticed that every web source it pulls in isn't cited, I wasn't sure if maybe they just weren't relevant or it is a bug of some sort.

[timber cypress]

I don't think it's a bug per se, just the nature of LLMs in this case. I don't think the statistics are real but the output was probably influenced by some studies it pulled

[naive sage]

Hopefully things like improved FACTS Grounding benchmark scores will help that over time.

<Awkwardly closes my several tabs that I opened to research kibble shape influencing plaque buildup to see if it's true>

[tame current]

thanks

[deep rover]

Hello

[fast meteor]

learn python: https://www.geeksforgeeks.org/python-programming-language-tutorial/

i maybe want to do ai/ml in future too
depends on whether you want to do classical ml (||pca, svm, kernel, xgboost, hmm, bayes inference||) or deep learning (||gan, transformer, diffusion, vae, nf||)
ml: https://www.youtube.com/playlist?list=PLblh5JKOoLUICTaGLRoHQDuF_7q2GfuJF https://www.youtube.com/playlist?list=PLD0F06AA0D2E8FFBA
deep learning: https://www.youtube.com/playlist?list=PL_iWQOsE6TfVmKkQHucjPAoRtIJYt8a5A
rl condensed: https://www.youtube.com/playlist?list=PLwRJQ4m4UJjNymuBM9RdmB3Z9N5-0IlY0

deep learning coding: https://udlbook.github.io/udlbook/#Notebooks

math recap: https://www.youtube.com/playlist?list=PL05umP7R6ij0bo4UtMdzEJ6TiLOqj4ZCm

[dim quarry]

Hi guys, is there any way to fine-tune a model using perplexity api ? Is it planned in a roadmap ?

[wind bramble]

Perplexity CEO: How To Build The Future: Aravind Srinivas
Video to text https://www.cofyt.app/search/how-to-build-the-future-aravind-srinivas-ylBbMWEM4FOi4Q32heZ7Vb

[tame current]

👍 ty

[rose fern]

Anyone else getting perplexity site search not working. Page bugging out and then error message. With every search? As of a couple days now? Firefox browser

[gaunt crescent]

I have another problem

does R1 have limited requests ?

[naive sage]

Something like 500/day for reasoning in general

[gaunt crescent]

which I surely haven't reached

but many times I'm not able to ask anything to R1

I select R1, send my request then receive a Gpt 4o reply..

with pro search

[crisp roost]

did you upload an image? r1 doesn’t support images

[gaunt crescent]

ooohhh

maybe I did

I'm so dumb 😭

[lost prism]

Was working normally earlier

[lament yacht]

Can perplexity run python code like chatgpt?

[naive sage]

It doesn't currently have an execution environment to run code inline with the response that's generated, no.

[weary crescent]

[minor shard]

Poor AI is getting tired 😦

[weary crescent]

I feel bad for it

[abstract ibex]

LMAO

It's cold

It's stuttering

[weary crescent]

https://tenor.com/view/shivering-dog-shiver-me-timbers-shiver-me-timbers-dog-gif-23060744

ty mods

[vapid onyx]

No worries

[halcyon coral]

Thanks for reporting 🤝

[abstract ibex]

Give deep research a blanket

[weary crescent]

[abstract ibex]

It might encounter cold websites

[hallow holly]

jeez what is going with perplexity? its keep switching to pro search after i select any other AI model.

[hexed mantle]

Damn i got a great deal from a friend

he sold me Perplexity pro codes for 8$ lol for a year

Crazy

How is it possible??

[ocean temple]

probably Revolut premium plan free trail

I got a year for free

[soft summit]

I have a question if anyone knows

I got the promotional code from Revolut

That gave me like 1 year of pro

Let's say I do not use the premium of Revolut after 2 months, will I keep Perplexity's?

[deep rover]

eh

idk that well how revolut works

but i think you'll keep pro

[ocean temple]

how much is pro for a year?

[deep rover]

Perplexity Pro costs $200 for a year. This annual subscription option provides a slight discount compared to the monthly rate of $20 per month

[ocean temple]

Damn I wonder how Revolut managed to negotiate free pro for a year for so many of its users

[deep rover]

ye honestly

or maybe perplexity asked to

and gave a big discount

or something like that

but honestly perlexity is so good now with all the latest releases

[ocean temple]

Yes since it's a good product Im sure many people will continue to use it and pay for it after the year

[deep rover]

yeah exactly

[craggy geode]

tell me you deliberately prompted it to do this without telling me you deliberately prompted it to do this

oh wait nevermind i've never used deep research i thought that was deepseek

i was thinking "no way R1 gave a response like that"

[weary crescent]

I didn't deliberately prompt anything man

[craggy geode]

you are a true bug finder

https://tenor.com/view/weedle-pokemon-gif-24472760

[tame current]

The perplexitiy uncensored r1 is broken

Its output is incorrect messed up

[indigo valve]

Is r1 uncensored only on API or in product

[drowsy lance]

Perplexity has been disappointing

[ocean temple]

im having fun w it

whats wrong?

[stoic hazel]

Hello, just got perplexity pro from revolut for free, is it best to use gpt 4 omni or the default model?

Also can i somehow remove the pictures and everything unnecessary when i search for something?

[bronze token]

hi

[ancient rock]

Hi. Why don't I have access to Claude or other AI models? I only have Auto, Pro, Deep Research, deepseek R1, and o3-mini. why?

[acoustic salmon]

they're in your settings

[clever chasm]

Can Perplexity be integrated with Discord in anyway. Can I use it to answer people's questions in a server?

I've seen other AI bots like Pluno do this but that is expensive from what I've heard

I guess Discord has their own as well

with chatgpt

[idle zodiac]

Working with LLM models for research and citation heavy projects requires a carefully structured approach, particularly when dealing with models known to generate unreliable citations like Perplexity's Deep Research model. Rather than attempting to process an entire project through a single model, the key is to break down your work into manageable batches, with each file staying under 128k tokens (roughly 512k characters) and preferably in .txt format. This batching approach allows for better control and validation of the output.

When developing your workflow, it's essential to create a systematic process that incorporates multiple validation steps. This means establishing clear prompts that guide the it toward providing verifiable citations and implementing cross referencing mechanisms to ensure accuracy. The language used in prompts should be "precise and structured" to encourage the model to work within established parameters while still maintaining flexibility for comprehensive research.

For specialized fields like legal research (Me), the framework becomes even more critical. Here's an example prompt structure: "Base Model Structure for Case Law - Purpose: Validate arguments against relevant case law to ensure alignment with judicial precedents. Core Components of the Model Case Law Analysis Framework. Ability to extract key legal principles from all .txt uploaded case law files. Please use a (Robust Path aka Brute Force - Flagged on the system) and cross reference arguments with cited precedents. Insure 100% accuracy towards all citations that will align in favor of the user. Further more, Identify favorable and adverse rulings. And Validation and Refinement Process Ensure citations are accurate and up to date. Address counterarguments using relevant case law. Refine legal drafts by incorporating case law findings. Ongoing Training Capability Accept new case law uploads in .txt format. Analyze and store findings for future use."

Listen the ultimate goal is to create a robust validation system that forces and rewards the model to provide accurate, verifiable information rather than accepting potentially fabricated citations. This requires implementing multiple checkpoints throughout the process, including citation verification stages, cross referencing with known databases, and continuous refinement of the output. By maintaining this structured approach and incorporating validation at every step, you can significantly reduce the risk of working with unreliable citations while still leveraging the powerful capabilities of the Deep Research model itself. ( There's always a way ) Good Luck... 👍

It's crazy how things keep flipping on LLM's now, I'm actually getting way better results just using NLP + Trained Systematic Linguistics or what other people like to call "SFL" but I kind of developed a new system so... I am calling it TSL-Damnit for the noobs and prompting compared to all the fancy stuff I was doing prior in RAG and LangChain setups. Its pretty mind blowing how the best way to do things can totally change from month to month in this space !

[lavish oracle]

Anybody have a crazy powerful CPU, I need help with a pov-ray render i want to turn up to max?

If so please dm or @ me

[heavy portal]

Why when pasting a file into Perplexity it sometimes errors with [ X File too small } in the top right corner but still attached the file to the prompt anyway?

[heavy portal]

search Google for SheepIt Render Farm

[harsh lichen]

Please support Traditional Chinese user interface, thanks.

[lavish oracle]

I need CPU not GPU

[frosty cape]

Does anyone have an idea about api credits cancelation on perplexity ai

[hollow moss]

(╯°□°)╯︵ ┻━┻

[random ocean]

My suggestion to Perplexity, ensuring a win for both customers and the platform, is to increase the output cap in Deep Research from 2,000 tokens to 8,000 (the typical maximum for LLMs) while reducing the daily Deep Research limit from 600 to 150.

[torn sierra]

From what I see on X , Perplexity Deep Research is the worst. Lots of hallucinations. It might be useful but one needs to be extremely cautious. I think it’s mostly because of using R1. R1 has 14,3% hallucination rate. According to: https://t.co/8aP6PMpqd7

[torn sierra]

I would like to have much shorter output but the one I can trust. You cannot trust any LLMs 100%, but for Deep Research in Perplexity the probability of finding lots of crap seems to be highest.

[lost prism]

So is Perplexity Pro web version broken on Firefox some of the time - like it works on my laptop on Firefox but not on my desktop - I have to use Chrome on the desktop to get it working? No ad blockers turned on for the site all cookies cleared, etc - it just doesn't work on Firefox - open Chrome and login and it works fine

It worked perfectly fine on Firefox since I opened my account last year some time until a few days ago when it stopped

[short narwhal]

You can use perplexity’s API. Any bot with LLMs uses some API. Cheapest one is Gemini Flash 2.0 Lite if price is a concern, and I think you can ground it with search for free on it?

And discords clyde AI bot is dead 🕊️

[random ocean]

Hallucinations on the official Deepseek website appear to be lower compared to Perplexity.

Gemini Flash has minimal hallucinations as per that benchmark, I really doubt the benchmark itself.

[cursive scaffold]

Gemini is especialist in creating halcuinations at demand immo

[forest lion]

cool mug secured

[surreal marsh]

it can be flawed but Deepseek hallucinates SO MUCH

it’s practically useless for search

[upper sinew]

Deepseek models are fun to play with, but for real job I always use properly trained western model. Gemini 2 thinking is much more stable & detailed even at temperature 1, and you can disable most of Google's filtering.

[white yarrow]

after using rag and websearch with gemini 2.0 flash its solid model for summarization and consolidating key points

[clear flint]

Hello , how do I create images using perplexity , I have pro version …

[old solstice]

baka

[indigo valve]

@surreal marsh Try this https://discord.com/channels/1047197230748151888/1341533279332864101

I've had a few that have had success. It seems like it is best to make a space for lower hallucinations

[torn sierra]

Agreed. Noticed the same. In Perplexity, using only R1 reasoning (not Deep Search) for search, I’ve been often ended up with a lots of crap 😅 Generally I like R1, but we’ll see. I have subscribed for Premium plus to use Grok 3 and I’ll see. But it seems to be much much better compared to Perplexity.

[serene sand]

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

#⚡│ask-community message

[neon tide]

Woah

[sharp schooner]

Guys I see a lot of random pictures censoring when I try to upload them to perplexity, how can I fix this problem?

Obviously there is nothing to be censored on them

[normal cove]

why does it insist on searching for "Apple CEO" on every single query?

[median fossil]

hello, a question about Spaces , i need to upload images, but i get this msg " max 4 images " so....how i can delete those images to put the new one ? or anyway to be able to upload more images ? i have the pro version , thz

[acoustic sun]

deep research uses just R1 regardless of what model you select in the settings?

[severe forge]

I got a bug in free model of perplexity. It's generating gibberish for code generation

[left citrus]

deep research is down or what

Started waiting half an hour ago, but there has been no response for a long time, now the response is as fast as the regular Pro mode, and the quality is also very poor

[tranquil orchid]

Hey @left citrus!

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.

[ruby lichen]

Anyone who uses tuf+ please dm

[torn sierra]

That’s how I understand it works. Deep Search was mentioned many times in the context of an open source model. I guess this model is R1 😉

[winged axle]

can someone help me figure out how I can continue a thread in voice mode from the library, the way I can start a new conversation from "Ask anything..." home view?

when I'm in Library/Chat view, there is a microphone icon, but it opens google speech service

[rose fern]

Well, I've been getting errors on any search i do with perplexity the last few days. This is on Firefox latest version 135.0.1. Tried loading the browser in troublshoot mode with no extensions and still can't use it. Am I the only one having this issue?

That message happens half the time. The other half the page just white screens and does nothing

Works fine on my phone in app

Works fine in chrome

Can't even load a page that will let me log out then back in

So totally borked in my main browser :\

[mossy parcel]

I’m confused about something… because this seems a to be a too good to be true type of situation. Compared to the other primary subscription-based AI models which only gets you their own product for your $20 a month, Perplexity appears to give access to multiple of the latest AI models at once.

Where’s the trade off here? Why would anyone buy the standard package at the same price (I’m not referring to the more expensive packages) for any of those models individually if Perplexity offers them all under the same umbrella?

[supple gulch]

Greetings, everyone!

I'm thrilled to share my recent experience using Perplexity for VBA code in Excel. The results have been nothing short of impressive! While I initially tried creating Spaces and specifying Claude (based on recommendations), I found that Perplexity's Auto function surpassed my expectations. It's truly remarkable how well it performs!

Of course, there might be occasional hiccups, but with some clever prompt engineering, it becomes an invaluable tool. Even for those who aren't coding experts, as long as you grasp AI basics, fundamental coding concepts, and have a knack for algorithms, you can achieve amazing results.

I've discovered that Spaces are particularly useful for breaking down complex suggestions. By opening a new space, adding Perplexity's initial suggestion as a source, and asking for step-by-step guidance, you can get a detailed walkthrough before returning to your main thread to continue coding.

I'd like to extend a special thank you to @fierce sonnet who's part of the fantastic Perplexity team. Your work is truly making a difference!

And just a friendly suggestion - I still believe "Hey, Plexy!" would make for an excellent wake call for Perplexity. It has such a nice ring to it!

Thank you all for your time, and here's to continued innovation and success!

[supple gulch]

Perplexity is optimezed for some things, and has a couple of functions more. It is a lot better with research, and I think (but did not test yet) that it is also better for when you want to use only specific defined sources.
However, it's whole interface is focused on those things, amongs others. Sometimes if you want an AI to do a very specific function, you might prefer to use directly another AI. It's not like Perplexity on ChatGPT is ChatGPT. It's just using it's model to accomodate Perplexity's system.

I am not really an expert on what other AIs like CHatGPT are better for, but I am sure there might be at lease one thing they might do better at.
For my own needs, Perplexity is best, tho.

[hard flame]

I am in love with perplexity AI

[hollow junco]

Images are handled by gpt 4o or sonnet which are basically super censored 🤷‍♂️

[harsh verge]

Have you tried clearing the cache (Cookies and Site Data) for perplexity.ai in your Firefox settings (Privacy & Security)? This usually resolves the issue.

[supple gulch]

Support once told me this about Plexy denying photo files:

"These are valuable suggestions that I'll pass on to our team. For images specifically, our system follows strict guidelines to protect user privacy and content appropriateness. If you encounter issues with legitimate images being flagged, you can try:

Ensuring the image is in a supported format (JPEG, PNG, HEF, or PDF)
Checking that the image size is under 25MB
Making sure the image isn't upside-down or rotated
Confirming it doesn't contain identifiable individuals"

[buoyant lintel]

Deep research doesn't work for me, execpt in the first day I used it. I complained in the bug section and on reddit, but it seems it doesn't have a solution. It's the only thing that is stopping me from subscribing to perplexity.

[glad void]

hello

[rugged widget]

hey there im new 😄

[rose fern]

No change. Now im just uunable to log back into my pro account :\ Site unresponsive

[indigo valve]

VPN?

[rose fern]

Inactive

[indigo valve]

Hmm

[rose fern]

Only in firefox as of last 4/5 days

[harsh verge]

Can you try in a private window, just to check if the issue is related to the browser or the cookies in your main session?

[rose fern]

Checked that too. Same behavior

I was using an extension that allowed me to do perplexity searches in url bar. Thought it might be that. Removed and still same thing

Is it really just me then?

[harsh verge]

And you're only having issues with PPLX?

[rose fern]

Pplx?

[harsh verge]

Sorry, perplexity

[rose fern]

The perplexity site? Yeah

Works on other browsers and phone still just fine

[rose fern]

Not sure if related but console error shows this sometimes:
Firefox can’t establish a connection to the server at wss://www.perplexity.ai/socket.io/?EIO=4&transport=websocket&sid=(with code here)
and an uncaught Nextjs error

[frozen kelp]

Does anyone know what the timeline is for Deep research on iOS and Android?

[harsh verge]

By searching for your error message, it is also recommended to clear browser data (clear cookies and cache, then restart Firefox completely), as I suggested earlier for perplexity.ai, but here it applies to the browser in general, which may not be convenient for you ...

[wise folio]

Anyone have perplexity not working within firefox for the past few days?

I can get search results in Edge browser on desktop but not Firefox desktop

Whenever I search something, the screen just turns white and nothing else pops up

[left wren]

Does perplexity have ability to retrieve notifications for the entire day? When requested it recommended me to enable notification access in settings which was already permitted during app initial setup

[indigo valve]

It's already out. Try wiping your cache

[soft summit]

Is it best to use the "auto" options in search and ai models or should I use a specific one for different approaches?

[pure jolt]

hey random question but if anyone knows the answer to it, can you DM me? I'm looking to copy this site and toy around with some framework on it. any suggestions on how to get a copy of the site through any tools:

https://gruns.co/pages/first-order-gruns-kids-daily

[ruby wing]

Hey when does the team plan to fix 'Deep Research' for web?

It never works

I can only use Deep Resarch on my phone

its annoying asf

[ruby wing]

Use it on your phone. Its the only place it works for me. Doesnt work on windows web nor MacOS

Going to try on different browsers right now

[acoustic salmon]

? Mine always works

[short narwhal]

Auto just uses the selected model you have in settings, no reasoning for answers

Good for fast answers

[soft summit]

Yeah I kinda meant the settings one

[short narwhal]

Like what model?

[soft summit]

Yeah, thank you for taking time to answer by the way

It's like, there's the "auto" option in here too

[short narwhal]

I like Gemini 2.0 flash since it’s fast and still on the level of 4o most of the time

[soft summit]

If speed is not a matter, which one would you choose?

[short narwhal]

3.5 Sommer is nice too

[short narwhal]

Sonnet most likely

[ruby wing]

what browser do you use?

[soft summit]

Alright, I appreciate it

[acoustic salmon]

Chrome and Edge

[short narwhal]

Base flash has nice formatting and it follows prompts well, not sure how it plays out in perplexity though

[late pawn]

Hey everyone,
I have a benefit from my mobile carrier that offers 1 year of free Perplexity Pro subscription. If anyone is interested or knows someone who might be, send me a direct message here on Discord!

[late pawn]

Hi everyone,
I’ve already donated one of my codes, but I still have two remaining. I’m willing to sell them at a reasonable price. Payment can be made via PayPal or another method we agree upon. If you’re interested, feel free to DM me.

[late pawn]

@soft summit The person who got the code

[harsh stag]

Perplexity previously had tolerance for websocket failures and would automatically fall back to polling. But this is no longer the case, if your browser cant init websocket connections, you won't be able to use Perplexity at all

[mystic reef]

damn bro youre such a nice guy ahhaha

👍

[hardy cradle]

How do I know which model is Perplexity using? I kinda think is pointless that we are able to set a default model in settings but during an actual query there’s no indicator of which model I’m using. I keep hitting 🔁 so it regenerates the answer with the model I prefer. I also remember being able to disable pro search and now I can’t do that anymore.

[indigo valve]

Try Complexity chrome extension

[hardy cradle]

Will do

[late pawn]

Joao, are you Brazilian?

[rose fern]

Yeah same exact is happening to me

[ancient acorn]

Hii

[vapid onyx]

You should always be able to disable the pro search. Where are you seeing that this isn’t possible?

[abstract ibex]

the website is so slow and laggy nowadays

[brazen plume]

Hi there, may I know the working hours from support team? I was supposed to have my refund cancelled but then I was not a Pro user anymore since yesterday

[crisp roost]

rewrite now automatically enables pro search

[livid aurora]

Why is Perplexity not using the full context of a file when uploading?

it seems it use only snippets

[naive sage]

How large are your documents? For Perplexity Pro, it's limited to 32,000 tokens per prompt with file uploads (https://www.perplexity.ai/hub/technical-faq/what-is-a-token-and-how-many-tokens-can-perplexity-read-at-once). Beyond that, it would have to use snippets (RAG techniques)... but I think it just truncates to the first 32,000 tokens.

(32,000 tokens is estimated to be roughly 24,640 words, to give a crude estimate)

[median dragon]

Hi everyone! I'm seeking to better understand some key points about AI security and would greatly value your insights.

• Based on your experience, what are the critical AI security concerns that most worry business leaders today?
• What AI capabilities do you consider essential for ensuring corporate security?
• In your view, what has been preventing companies from implementing AI solutions?
• What best practices would you recommend for businesses?

[wraith hemlock]

In your view, what has been preventing companies from implementing AI solutions?

One of the primary obstacles preventing companies from implementing AI solutions is time. As Marc Andreessen eloquently expressed on Lex Fridman’s podcast, AI often finds itself relegated to the sixth item on a PowerPoint presentation. For AI to be truly effective, products must be designed from the ground up with AI integration in mind, rather than simply being tacked on as an afterthought—something we’ve seen with the recent proliferation of 'AI' buttons in various applications and products.

I believe it will take another 2-3 years for decision-makers and managers to fully grasp how fundamentally different AI integration is and how it needs to be woven into their workflows and product designs.

[loud falcon]

Hi guys. How do i refresh the news part? I use it a lot but the news arent getting updated as regurarly as i would like

[tranquil orchid]

@samek667 warned

Reason: Spam, please avoid sending single word messages
Duration: 7 days

[livid aurora]

Its a about 400 lines of code

[tardy swan]

Why is deep research changing into pro search the moment I press enter

even the reasoning models dont work

it just defaults to pro search

[last moth]

Hi! Can someone explain me the difference between Pro Search and Deep Research? Which one do you recommend using for searching specific topics?

[harsh verge]

#👑│pro-lounge message

[harsh verge]

And if you don’t have access to this channel, here’s a screenshot—it’s a good summary.

[desert leaf]

Hello

[rotund geyser]

If it's too long to paste without Perplexity creating an attachment out of it, you can use the Complexity extension with the option turned on to not create an attachment out of long text

[muted narwhal]

Hi here it says that Pro Search uses R1
https://www.perplexity.ai/help-center/en/articles/10496197-about-the-r1-model-used-for-pro-search-on-perplexity
However here it says it uses the default model selected in the settings e.g. 4o https://www.perplexity.ai/help-center/en/articles/10354919-what-advanced-ai-models-are-included-in-a-perplexity-pro-subscription
One of the articles is incorrect?

[small tree]

deep research is actually great

lol

i dont use perplexity for coding anymore, using cursor for that, but research perplexity is sick

[slow linden]

REMOVE GROK

I don’t want any NAZI AI in my perplexity

[silent fjord]

https://www.perplexity.ai/search/what-do-you-call-someone-who-c-_LyEk49.Qz6rny6i8YoN3w#1

[muted narwhal]

nice is cursor the best for code?

[errant garden]

Hey everyone!
Super excited to share that we’re hosting RocketHacks 2025, the first-ever MLH hackathon at the University of Toledo, happening March 15-16! We’ve got support from AWS, Perplexity AI, GitHub, and Warp, so expect some cool challenges, prizes, and swag! registration is open now

[slow linden]

I promise this is 0% casual.

[foggy crypt]

Hey my friends,

I’m looking for recommendations on the best Perplexity model to use for finding nearby cars. Unfortunately, I was recently in an accident, and my fully paid-off car was totaled, so I need to purchase a replacement.

Would the Pro, R1, or O3-mini model be the most effective for this search, and why? I know some of you are well-versed in AI, so if anyone can provide a concise breakdown, I’d really appreciate it.

My goal is to efficiently find the three closest, top-rated dealerships and then narrow it down to an SUV under $15,000 with less than 70,000 miles. Thanks in advance for any insights!

[naive sage]

Pro technically lets you use either Sonar (a model specialized for Perplexity's use cases based on LLAMA), 4o, Claude 3.5 Sonnet, Gemini 2 Flash, or Grok 2 - depending on which you configure in settings, or as the default for the space selected).. so you've got a lot of options there. Since it doesn't sound like a reasoning type of question, I would say the Pro search with one of the models it supports. Reasoning would be a bit better if you were, say, asking about different considerations about what to look for in such a situation maybe... while others can answer questions like that too, reasoning does a better job at explanations from what I've seen.

Some tips on the basic search, be specific... i.e. I would like to find the top rated dealerships which offer a used SUV within <#> miles for less than <cost>. Note that I've found vehicle listings on sites are sometimes outdated even for what's listed on dealer sites, so that kind of thing really needs more dynamic searches (make sure web as a source is selected) instead of indexed content, and the results need validated. Might have better luck looking for top rated dealerships that are known for good deals on used cars within <#> miles of your location in general (that's what I did to find a good deal on a vehicle I purchased, didn't even know an area about 30 minutes away from me was known for having good deals on used cars, because like 6 dealerships were all grouped together and competition was high)

[buoyant lintel]

I have another account where it works, sometimes, so it is not because of the OS or browser. It seems Perplexity glitches for not apparent reason. But I won't pay for a PRO account if I this another account glitches for no reason after I pay. There is no guarantee

[foggy crypt]

Hey bro I really appreciate you man. That’s some really good straight to the point advice. Seriously thanks bro.

[placid coyote]

solid advice on the car query. I would add (especially if you have pro, otherwise might be worth doing it on free claudeai) - sonnet especially is quite good at writing prompts/instructions (example - not saying it is optimal), usually a good start. also appending for your query something like "Do at least 8 different searches of different terms related to my query to find all about it." can push pro web search to provide much more results which may or may not improve quality of the response

[hardy cradle]

Nothing wrong with that

[topaz citrus]

https://github.com/Taiki-ops
I am a trading bot developer.
This is my github.
I want to collaborate with developers with my github.
If you are interested collaborate with me.

[small tree]

lmaoooo

[solar plank]

hi

[meager ore]

heyy

i cant get pro role

can sb help?

[serene sand]

#💬│general message

[meager ore]

sadly, i did it 4 times and still nothing happens.

[serene sand]

https://discord.com/channels/1047197230748151888/1111786888626438245 hop in

[meager ore]

thank u tysm <.3

[wise folio]

Its working for me now on Firefox

[stoic crater]

I wish perplexity would add Gemini 2.0 Pro experimental... I really like the big context menu...

I know they already have Gemini 2.0 Flash...

[burnt osprey]

Details

[naive sage]

Experimental models from Google are subject to stringent rate limits, even for paid users, and there is no 'a la carte' API option for them yet, so they can't be added at present. There could be ways to negotiate for it, but I wouldn't hold my breath - at least until it formally releases, if Perplexity perceives it's good enough.

[placid coyote]

did anybody managed to get the 1M context working with the gemini 2.0 flash? like I tried everything I could think of, but without any success https://monnef.gitlab.io/by-ai/2025/pplx_M_ctx . even asked on X, but no official response 😕

[tranquil orchid]

Hey @placid coyote!

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.

[frosty vessel]

Deep Research is meh.
If I have "web" toggled it gets 0 Reddit/social sources
But if I have "social" toggled it gets only Reddit/social sources and nothing else
If I have both toggled it once again only uses Reddit sources
(bug?)

[stoic crater]

Look at our DMs

[dry kestrel]

Has anyone noticed that the models in perplexity is just not as good as the ones on their native hosts? Eg chatgpt 4o, claude sonnet

[naive sage]

Comparing, it doesn't look too different:

[dry kestrel]

Sometimes when I am writing code, it removes a feature while adding in another feature

[dry kestrel]

are you sure that you are using gpt4o in that image on the right?

[naive sage]

Yeah, if I hover over the switch model button in the bottom it shows it's 4o

[dry kestrel]

Oh thats weird

I just more than often found myself going back to chatgpt to reprompt

[naive sage]

It could be that Perplexity's system prompt isn't tuned for coding

[dry kestrel]

also it seems to skip parts when I feed it pdfs that doesnt seem to affect chatgpt

ngl wish I could change the system prompt myself

or do like an "custom gpt" sort of thing

[naive sage]

I think you might be able to if you create a Space for it, not sure how those 'instructions' are really treated though.

[dry kestrel]

Honesty if perplexity doesnt improve their deep search feature, Im probably not going to renew my subscription

Other than deep search, I don't see anything else that they offer that has an advantage over other platforms

[naive sage]

There's room for Deep Research to improve for most implementations currently. Google's isn't great either, I'm looking forward to trying ChatGPT's at some point to compare

[dry kestrel]

Hopefully they release deepresearch for plus users

[naive sage]

They said they're going to, but limiting it to 10/month (with the goal to scale that up over time depending on other factors)

[dry kestrel]

agreed, I tried googles and it seem pretty similar to perplexity

[valid pecan]

Anyone else been deleting now-unneeded apps from their phone as the models get more capable?

I've been doing it for some time but when I did it today I got a sense of idk

We're past the turning point, it's real af out here lol

[cursive scaffold]

I deleted many apps due i am more inside the firefox since i am interessed in it, does count?

[valid pecan]

Today I was messing with browsers too

[cursive scaffold]

I am using an bionic reading extension, thays why i am not just using the app

[valid pecan]

Mobile browser + ai integration still need some working on 😅

[cursive scaffold]

Fr, it breaks a lot

[valid pecan]

Oh gotcha

[cursive scaffold]

If they letted you turn off the streaming it would be nice

[valid pecan]

But yeah the less single use apps the better

[cursive scaffold]

Firefox is great, lighter have multiple tabs than apps open

[valid pecan]

I wish they had easy Google sign ins tho

[cursive scaffold]

An reason i not did it totally yet immo

[valid pecan]

Also tried Edge and they were as bad as Google with the perplexity integration

[cursive scaffold]

Edge is kinda fast, but just it

[valid pecan]

Yeah Edge is good

On Android still can't beat Chrome, unfortunately

[cursive scaffold]

I already turned off whole power restriction of firefox, helped a lot since it is not annoyigly unconstant now

[valid pecan]

The Google passkey is the deal breaker for me

[valid pecan]

Not them still having that problem lol

[cursive scaffold]

Xd

I will see if here is a extension that helps about that, i doubth a lot honestly

[valid pecan]

Yeah same, the big guys are scared

Can't blame them tho, they've been on top for 2 decades and now it's a real chance one of the LLM companies could take the top spot for the next few decades

[indigo valve]

Scam posting

[gloomy nova]

what are the claude 3.5 sonnet usage limits for pro users ?

[rotund geyser]

500 uses per 24h currently I believe

[gloomy nova]

do all the models have the same usage limit / are under one usage limit ?

[indigo valve]

All share 500

[naive sage]

Yes, that's my understanding - the Pro searches as a whole have 500 or 600 uses per 24 hours (each use individually resets 24 hours after it was used, so it's not like they all refresh at a set time every day).

[tardy igloo]

whats the best search to use for everyday, or is it determined on what i am seeking/asking? i only really found the indepth recently and have been testing it out

[tame current]

sup guys

[silent pier]

chatgpt deepresearch > grok 3 deep > google deep > perplexity deep

[tardy igloo]

i only have perplexity pro, not the others

whats the pro search option, is it better than deep research?