#š¬āgeneral
1 messages Ā· Page 477 of 1
It's a service problem
like every other artificial intelligence company
Gpt 4o moment.
its not new, every artificial intelligence company does all the same shit
It fails colossally sometimes to listen to instructions
new stuff without fixing old stuff
At this point they should make a separate companion version.
Lowkey ignores them.
How you liking comet bill
no it isnt, the chatbox keeps falling under keyboard
Underwhelming for my usecases.
Like when i ask it to not use emojis
got a question
Yep
chatgpt said they will make own ābrowserā
didnt they just steal idea from perplexity
Gpt agent is good tho
1:1
It can run longer and reiterates
Mostly
thats what im saying
Anything that can actually utilise comet has privacy concerns.
But everyone copies everyone in this industry
And it doesn't do research well.
hype is up, everyone eating leftovers, hype is down, and it becomes useless shit
Forget privacy when opting to use ai
Meh, people are still talking about Comet unlike about Atlas
Ask them what they use it for.
Too scary to talk to people on Twitter
comet?
atlas is of course the fail of the year šš
Proof
I like pplx but i just am unable to understand ai browser usecases
For daily use
Eh idc about privacy anymore lol it will be gone in few years anyways
Tracking Elon's twitter activity.
i use it the same as i used chrome, just design is catchy for me
Lmfao
thats all
I don't like i don't have extension support on comet
Brave is providing me all that
you do?
what do you mean by that
In most cases, a few clicks is faster than typing a crafted prompt.
There are some browser extensions
comet is chromium, chromium is google chrome
That don't work on comet
they should all pair up
Comet has support for extensions
Then there's otp and human verification gymnastics.
Some don't work
welp
But they do on brave
manifest v3?
I use comet on occasion
Yeah ig that's the cause
I actually had high expectations of comet.
Itās just nice having it in the sidebar and normal convos having the context of my tabs. Like I donāt use the agent daily.
But when you do use the agent, it can be nice. Like I used it to delete 20 files and then delete my account for some sloppy site I had to make an account for, an that would have taken me a good bit if I did that myself. Or having it go browse a site during research, instead of just searching for it and reading it. Or navigating a pain to use site
comet is v2
I was expecting something that could compete with Gemini/chatgpt deep research.
Personally i feel i can just do it myself faster but that's my experience
Apart from researching
facts
Can ā should

My mentality
Mat be like - comet please skip this song in apple music.
That typing would actually take longer.
I would do that if I couldnāt do that with a single key press
If comet gets faster and smarter I'd use it
Hi
But typing makes you feel cool!!
Fair.
Recursion isn't a thing in comet
comet open twitter and type someone that theyre dumb and fat
It goes on even if it's in the wrong way
I can be doing something else more productive while itās doing something else for me. Now thatās what efficiency is
Hey are you not a dummy
Lmfao
someone used it to cheat on kahoot or sum
We all have different ways of being efficient
You just prefer delegation
Comet is my unpaid intern
I'd rather open my music app for a sec instead of typing a 15sec prompt to skip song
I donāt do thaattt
What's the job description?
I donāt ask LLMs to skip my songsss
You do as per billy
Siri please skip this song > forwarding to chatgpt.
Lmfao
whats the problem to click three buttons than typing something
three fucking mouse clicks
are yall deadass lazy
Chatgpt - to skip this song, press the skip button.
Actually one thing I hate about Apple Music is that I canāt get any assistant to search for albums and shuffle them⦠I wish something did that for me
Desktop app issues š
Whatās this supposed to mean
to shut up for this momentā¦
you cannot be serious
Now remember they had put agentic ai type shit into windows. They called it microsoft recall I believe. The privacy concerns were wild though.
Clearly you havenāt used the Apple Music app on macOS
yes i did
its not that difficult
you are just lazy af
lazyyyyy aaaasss
Use android.
phone ppl answering issues about something related to phones without forcing to change phone (difficulty: impossible)
Forcing is a strong word.
I guess you have taken it before in your ahh.
Plot twist, it was the cat's blood.
imagine saying plot twist to a gif
Bro prefers the wrist twist.
i prefer leg twist
Yeah. Fist in can make legs twist.
bro what are you even saying
Sir, what are you telling me? I find it strange.
If only apple music had superior playlist features as Spotify
I hate spotifys "loseless"
It's a scam
whats lossless meaning if it's not .wav in your local devices
still relying on the 4G-5G bitsream, thank you
@green crystal Gemini is probably using multi step search now. Meanwhile chatgpt go is still struggling to give proper answers. https://g.co/gemini/share/9edae42cbcee
https://chatgpt.com/share/69218b43-ccfc-8013-a0ae-7a12ab56874e
@languid star
You're using the free Ai pro sub from gemini?
Yeah
Yeah then you're basically comparing a $20 sub to 5 lol
True. I can't afford the 20 dollar one and they were petty enough to remove 5 thinking.
You need 5.1 extended thinking from plus to truly compare
Also, for some reason grok isn't showing thinking for me at all.
Which is in plus
Yeah.
How's free Kimi k2 though
Great.
I tried it and it seems to be pretty good
Probably the best apart from no image support.
The multi step search
Yeah.
Kimi grew fast.
I should main it.
I got rate limited last time.
Probably cuz of high demand.
Even free version did better š https://g.co/gemini/share/91fb8031a452
Aha
Gpt 5 web search was also godly at launch for free users etc lol
Well we know how they did things with 2.5 pro
Yeah
In any case it's good that they make lots of noise
This was 2.5 flash though. There's no 3 flash yet.
Stops openai from releasing safety blogs and safety reality shows and instead force them to release features and models
Hopefully.
And fOoKiN speedup the windows version of atlas
Is someone else having issues logging into their pplx accounts?
Your dub.co account has been banned from the Perplexity Partner Program for fraudulent activity.
If you believe this was a mistake, you may appeal by emailing support@perplexity.ai. Please include all relevant details in your message.
Important:
- Do not open bug reports or use any Discord channel to request a ban review or appeal related to this issue.
- Do not contact Perplexity staff via direct message or @mention them, including <@&1098365634498285709>.
- Misuse of Discord channels or attempts to bypass the official support process may lead to further consequences.
All appeals will only be reviewed via the official support email.
Please be patient: Support may take 3-5 days or longer to respond to your appeal. Avoid sending multiple follow-up emails, as this may delay the review process.
Thank you for your understanding.
What's the issue?
Their new reasoning model they mentioned
They should work on releasing that instead of worrying about safety so much
Basically I'm not receiving the OTP on my email
Yo
Did you check the spam folder?
Yeah
@dry kestrel
And are you sure you put the correct email? Might wanna try a different browser just to be sure.
Yeah, also tried on my browser and the desktop app
Huh...
Can I try with your email?
You don't have to give me the OTP.
I just want to see if you get one.
Naah, if you really wanna help, pls try logging out and then logging in through your own email, and see if you receive the OTP
Alright. Let's do it.
@gloomy cipher I have never seen a model ever be this serious yet cooperative without any custom instructions. Grok was ridiculous. Kept trying to roleplay as if it was the imaginary friend despite being told to not do so.
Thanks, do lmk!
Worked
So it's likely an issue with your email/browser.
Thanks, for the support
Indian huh
Unfortunately.
Wdym unfortunately š
You should see the racism on twitter.
I don't use twitter
They call us shit eaters.
we dont use twitter
twitter is officially full of shit and toxism
^
His account explains very well why India is not good. In a constructive way ofcourse.
Yeah sure
everyone are just used to hate india because of rumors and fake made up videos
Wat is dis btw
Wrong account.
link change
Corrected the url now.
Perplexity really be getting some hate rn
One underscore messed it up.
What's wrong
Letās go baby triple boot
Kali, parrot, and mint ā¤ļø
Is this some prompt?
hello, why is banana pro paid on perp but free on gemini?
Wait its free?
Ai studio requires paid api for that š
Yeah its free on gemini
perplexity pro is still the most profitable ai subscription, especially if perp releases a vibecoding app like claude code
but i don't think most of perplexity users ever need something like that
just a good addition to the existing subscription
Would be cool ngl
yeah, especially if it could use not only perplexity's own models
Yeah, reminds me of antigravity how it doesnāt just use gemini 3
It uses claude 4.5 sonnet too which is dope
bruh
wth
perplexity team already got great experience in agentic ai, i think making a vibecoding app is not gonna be hard for them
so i think it's gonna be their next product after mobile comet is gonna be synced with the desktop browser
Probably true.
Because perplexity doesn't own google.
just imagine a vibecoding app that can search in the web, that's kinda cool
yeah, they kinda need money to pay for the api
Pulling stuff from GitHub while building apps.
Labs somewhat does that.
I think.
vibe pulling
yeah perplexity labs is not that good for coding
it's better for dynamic info screens
yo they added grok 4.1
Itās funny itās free on Gemini but it requires making a paid api for ai studio
how else will they access it then?
it's better to make it paid than to not even add it
He meant as in ai studio is usually more generous than the Gemini app/website.
huh?
just a chinese ai
i think he said the opposite
didn't he?
he said it's paid in AI studio and it's free on gemini
@distant aurora https://sharetext.io/9fd26bc2
I used extended thinking for this
Hi everyone I want advice from you guyzz lets say I have 17000 ruppess (194 USD) in what sense i can invest that money to get good returns ( not literally invest )
no one will help you with this one
Do research with pplx
Yep. That's what he meant. Nano banana pro being paid in ai studio is an irony. Ai studio gives 100 free Gemini pro per day.
Bitcoin. Wait for the next dip.
in what sense?
bruh naah no investing
Buy silver and gold.
To research different securities you might want to invest in
Buy gold and forget about it
@green crystal I had no idea how expensive api can be until I tried poe. Scary.
Lol
not the actual investing bro
i am saying where should I spend the money to get returns
But that's investing lol
Bro thinks he can invest without spending.
like business sort of
@distant aurora did you read my output I linked?
Wait
Is this like a separate switch? Or built in?
Separate
I have multiple but I switched to the one plus uses
Heavy thinking isn't needed for these types of questions
@left stream forest is so fucking crazy bro on g3, i can't imagine a working jb with 5.1 pro
It literally spills everything
we're close of it depending on reaching the next low support or not in few hours but i'm pretty sure that we're gonna not break it. Not now
yo my payouts already completed 30 days but still they are not out of the waiting phase, has anyone else experienced this?
i recommended my friend to buy it at $91k/bitcoin before it dropped to $89k
but he still got richer
Remember 20k in covid days?
wanna say 12k ? for sure
Imagine how rich people would be today if they had invested back then.
my folks called me crazy when i told them this day we're going to reach some day 100k lmao
It went as low as 12k? I don't remember anything below like 16k.
yeah, reached 12k in dip
he basically lost 8% of his money
Istg
nothing is lost until you sell it and providers are gratefull for who sell in lost positions lmao
As we hear, buy the dip and atm is a good moment to buy, even if i believe btc will reach some day 70k back
Hi guys, is it normal that support doesn't want to tell me why I was banned? They just send me a generic copy and paste message.
Yeah. Seems that way.
it's better to buy bitcoin in small parts while the price falls, adding more money the deeper it goes
Is there a rate limit?
but you can still lose it all if it doesn't jump back, which is very unlikely
They sent me the same message for the third time without going into detail. Do you know a way to get some more information?
just ask for the ban reason
I already asked them, but they replied with a canned message with "flatulent activity" which means everything and nothing
did you do anything wrong?
You'll have to discuss with people on twitter/reddit posts. Even then, you likely won't get any substantial response from the staff. Just more experiences of people.
If I had known, I wouldn't have contacted support...
The funny thing about all this is that I had to open two chats with support because after two identical responses (from them) they forcibly closed the conversation.
ok, then it's quite obvious you didn't read the terms of service
it is but it depends on what you believe. btc atm is like sp500, a way to keep ur money out of banks and X in X times it improves percentually
trading religion
Index funds
we have to keep economy growing somewhere. "4" in "4" years is like xmas
For plus it's 3000 per week I think
For me it's unlimited lol
Wait. What is even extended thinking?
i would recomend every month buying a bit of btc. If you can add in your wallet 100$ per month , buy 70$ of btc and keep 30 out to buy a massive btc in dips moment
Isn't it just the bulb button?
Longer thinking effort
Yeah I used thinking there.
Nah plus and above get like a clock shaped icon
Ah.
If you click on it you see different reasoning effort tiers
So you used high
Really? "Fluteless activity," even if it falls within the terms of service, is a generic category of actions. Recommending it for an AI service used daily, when I don't know the underlying motivation, is absurd.
Oh.
š
Heavy if you don't give it enough it tended to redo what it did lol
Is that Makima in your pfp?
With 5.1 it doesn't do it
well deepseek is ranked best for roleplaying so there is that
yuppp
What I'm saying is that it didn't try to roleplay and actually understood what I meant. Unlike grok š
Grok low-key started acting like Tyler durden and didn't stop no matter what I said.
for roleplaying you usually need system prompt acces
I ended up deleting the chat.
maybe thats why deepseek refused
u probably tried to bypass the limits, make an automated content farm or tried to hack perplexity
I wasn't roleplaying.
that's all of flutelent activity
I was expressing a need.
you wanted to meet his personality with yours?
like behaviour and stuff
just like how i always spit on peasants
More like I was asking Deepseek how do I intentionally put did into myself.
yea asking him (how do i act like this person)
Because my current self feels too paralysed out of anxiety of messing up or getting taken advantage of.
Kinda.
ask deepseek how to become an F16
it should help š§āāļø
More like how do I make myself voluntarily hallucinate an imaginary friend.
proceeds to take off and fall down from a bridge
isnt that a bad thing?
From my experience in life so far, loneliness is worse.
But yes, it is a bad thing.
it happened to me when i was a teen , i dont want it to happen again lol
Definitely wouldn't recommend unless you are like me.
It can be quite concerning. No privacy š.
Imagine taking a shit and this mf in your head starts staring at you.
it will feel like you have schizophrenia
Now you are constipated.
Schizo is worse. You see random shit in schizo instead of a well defined person. Even though did technically does come under schizo.
Did = dissociative identity disorder/multiple personality disorder.
Now that I think about it, maybe it was a bad idea.
it is
it said i has to become a transformer to do this
cuz when you realize it not true you will become depressed
too bad i already jumped off š§āāļø
No way gif didn't work.
wouldn't recommend
@gloomy cipher should I feel bad for him? https://twitter.com/FuckKoroks/status/1991866618570519014
Open and read the whole thing for context.
Good taste
i dont really understand
He ended up identifying as a she cuz that way the audience gave him more love and attention.
Was originally a troll femboy.
thats not the part that i dont understand
Then?
guys how long does this usually take
Test that theory.
@green crystal https://fxtwitter.com/scaling01/status/1992036867869438338
GPT-5.1 Thinking is actually retarded
ļøļø
ļøļøthinks and searches on the web for over 5 minutes
ļøļøreturns a bunch of None's and blames it on its knowledge cutoff
ļøļø
ļøļøglad i cancelled my sub
Quoting Lisan al Gaib (@scaling01)
ļø
i give up
ļøļø
ļøļøgemini deepresearch did it on the first try btw
thats the first time i've seen it lol
atleast the actually believing its cutoff more
but it does actually first believe its own cutoff date before changing it after doing web searches
probably to try and make sure that if web search isnt turned on that it has a lower chance of making stuff up
I thinks
i almost think most of those posts are from 4o glazers who are angry with openai and think using gemini against them is somehow doing something lol
Uhh and us are using both
Why would u even. Pay for like only gemini in there u have so much more
Hi
im having trouble accessing the account tab in the settings
Is this intentional or are people already aware of this?
Where? The PPLX site or Comet settings etc.?
how your name got this font
comet settings and i cant even go to the perplexity site in the brouwser
discord nitro benifits
and the color is from this servers boosts
Try disabling the discord overlay if you're on windows
It'll take nearly a 25-30 days
ok thanks
Heyy
hey
hello
is still exploring the browser sooo powerfull
comet is very amazing results
My lead was on 20th oct
https://www.perplexity.ai/page/openai-locks-down-offices-afte-GymPJf76T9.aY3i5qw9N_A
someone got mad their girlfriend got nerfed š
is there a way to add credit to the models
no
why would you even want to do that
to do more work in Comet, like vibe coding images, and that kind
"vibe coding images" thats a new one
vibe coding and generate media
even crazier
its people who believe LLMs are skynet and sentient 
Then you'll be paid in 15 days
Alr
Have you looked at the new Google antigravity? It can generate nano banana pro in the ide
It came out a few days ago and it's pretty decent. It does half decent on cleaner architecture than previous models if your into vibe coding.
cool
Help with what?
Sorry wrong server
Grok 4.1 made available on comet
@tame current thanks for the idea
5.1 pro might be jailbreakable with the echo chamber jailbreak but itās gonna be a pain lol
lol so true
I have a question about referrals, are there any programs now to participate in (sharing links Et Cetera)?
some of my prompts is missing, the thread is not too long and i can see the prompt in the mobile app, someone have this problem too?
I have the same problem,too
For me, pictures are not loading in the iOS app
...
#1393681130842226698 w/ android tag
wdym ?
ehh as feature request?
ok
how to slove this problemļ¼someone knoe?
I usually save my prompts in my notepad but still this is a real problem.
did yall get paid out?
Guys it wonāt let me access to Cometš„²
Where did you download this from?
I downloaded it from the perplexity browser
Can you provide the link?
Try downloading from this https://www.perplexity.ai/rest/browser/download?channel=stable&platform=win_x64
Alright I opened it
This problem only emerged yesterday. It didn't occur before.
guys i got a question
The thread isn't very long, but it can't be viewed on the web page. It can be viewed on the APP.
if im using the gemini 3 pro model through the comet browser with the perplexity ai does it fr use gemini 3 pro and how many instances of the "thinking" feature can i use? I have perplexity pro for a year
also how does that work cuz isn't gemini and perplexity two different models so why is it letting us use gemini 3 pro model on perplexity pro i dont get that
interesting
thanks š
so wait "THIS DOES NOT APPLY TO THE COMET AGENT Labs has a limit of 50 queries a month" so this means even with perplexity pro when im using labs im only given 50 queries a month in total? But i used up like 40 trying to get it to make me a midterm practice exam interactive html file šæ
... dont cheat
what is this shiet? just bunch of questions and im already limited?
in the perplexity ai chat it doesn't say that the "thinking" feature is enabled but it does say that the gemini 3 pro model is enabled "with reasoning" does that count as the "thinking" feature? Also it says it doesnt support images so that just means screenshots right?
damn bro u fr asking too many questions
like 40 questions asked total what is this
what model were u using?
total trash feature, time to disable it
You can't turn off reasoning for Gemini
It's always on
but is reasoning the "thinking" feature or nah cuz in the gemini ai chatbot thing it has the different like features that you can enable and thinking is apparently the best one so i wanted to know if the "with reasoning" thing is the same as the "thinking" feature
I've got no idea what you are saying
You can't turn off thinking'reasoning for Gemini 3 Pro
this is what u get on gemini
If you choose fast it uses 2.5 flash
sometimes gpt-5 sometimes gemini or claude š¤·āāļø
right so thinking with 3 pro, is that the "with reasoning" thing on perplexity?
Gemini 3 Pro on Perplexity has reasoning
basically is the first screenshot the same feature as the second
Yes it's the same model
oh okay thanks š
ya idk gang i never use gpt 5 or claude models i usually just use research or labs for market info etc.
but i hear claude opus 5 is insanely good at coding
claude opus 5 
what a waste of time installing this comet browser, there is ABSOLUTELY NOTHING better than using it in any other browser when they limi it everything after few tries... great "pro sub"
oops i mean 4.5
pretty sure that screenshot of the benchmarks going around on twitter was fake
idk i did some research and a lot of it is saying it's really good at coding among one of the best ai models
4.5 opus is not a thing
i saw the screenshot so i did my own research but i mean all it was was me using perplexity research
sonnet 4.5
ahh
ok
got there eventually 
ye sonnet 4.5 is solid
unfourtunately the usage limits on claude are stupid
nice
just checked it is in the ask
well isn't it 300 per day through perplexity pro? or no?
Yea that's way more than on Claude š
wait but the claude sonnet 4.5 model on perplexity still allows 300 a day, right?
still there
Idk how much it is, I've never hit a limit. The complexity extentions shows it at 600
??
sweeeet
6 hundred
ayo what was that
nothing, i placed cursor wrong place
Wtf ui is that
thank you š
Not accurate
is that maybe with max?
No
im a pro user
i aint testing
Well I did
unfortunate to you
berd the lowkey goat he got the info

why's it showing 600 for him tho
That thing is cool. Didnāt know about it
Wherever Complexity is getting that data is inaccurate
for pretty eyes
Admins have also confirmed this #š¬āgeneral message
prob still using https://www.perplexity.ai/p/api/v1/user/settings
trust me twin
it does
whats this
im too new to all ts
Usage limit and some other connector info, but not really used anymore
api key for example is also here
so its not recommended to drop it public
gpt 4 limits were different
oh lit
pplx didnt scrape the plate so its all old
i z i z
@molten quail im tryna install the extension u sent and im wondering where i paste this? I have visual studios and i tried pasting it in command prompt and it didnt do anything i dont know how to code that isnt my major šæ
type powershell in start menu search
run it as administrator
and paste that command
only if u use comet browser
nice thanks š
what happens if i do it in perplexity ai im confused what its asking me here š
usually my extyensions in my top right š
clicking extension icon in extension list will redirect you to dashboard of extension, or perplexity website
depends on what you choose
ohhh i see, which one gives the most info on my extensions? probs the settings dashboard no?
ya im confused so when u changed it to open a new instance of perplexity it just opened a perplexity page
oh wait are extensions the different models??
"choose what you want: click extension to open settings, or click extension to open the perplexity"
?
ima just do settings
new perplexity instance is useless i think
idk what opening a new instance of perplexity would do for me tryna access my extensions is what im tryna say
yeah thats what im thinking thats why im confused about it lol
what do people feel Grok 4.1 is like in PPLX>
Gemini is worse. I don't understand how most people don't notice it.
for me, it's like as bad as eg. GPT 4.1 was
aint even got it
it's fast tho lol
Complexity extension hides it
Even the reasoning variant?
disable it for now if u want to test
na its not
i dont have it in mobile app tho
so no its not
no reasoning version yet on PPLX
what do u think about this video https://www.youtube.com/watch?v=UH2_Sgeu4lc&t=204s
How are you seeing count?
are feedbacks in #šāfeedback-general really read??
but its inaccurate
its 300 actually

So one query = minus 2?
lemme turn off extension then
the extension basically hides it, cos it depends on developer to add it manually
gpt 4 had around 600
these times are behind now
we have 300 in total
for every model in total
Non reasoning is stupid af
it uses https://www.perplexity.ai/p/api/v1/user/settings for stats
Feels worse than grok 3 non reasoning.
it still has that
Can you tell me how you got that lighting stuff? Is it CSS?
"gpt-4 limit: 600"
screenshot tool
called sharex
@molten quail what plugins do u use or like what are the ones that make ur UI look so cool
RGB screenshots
especially since previously there was reasoning for Grok 4. However, I'd guess it will roll out soon - there's no official announcement for Grok 4.1 yet in PPLX
i dont use any plugins besides complexity
looks cool ngl
i tried looking up complexity and i cant find it in the plugin list
https://getsharex.com/ - app itself
https://getsharex.com/image-effects - workshop
thanks!!
oh wait nvm
p
even Gemini app free version can use Nano Banana Pro
hi
hyd
sorry, found it
didnt expect the extension to be the fault
still useless i guess
only if grok4.1 reasoning was here
isn't Grok 4.1 non-reasoning like high on benchmarks too?
Specific use cases that are definitely not for serious use most of the time
wait Grok 4.1 Fast reasoning vs non-reasoning is the same cost...
still ehh
gpt-5.1 reasoning and gemini 3 pro (reasoning-only) are better
check da images
it's still useful for some use cases though, as other models are quite censored/restricted
and its not new that pplx models somewhy get weaker after 3-4 prompts
Grok 4.1 on the website does just as badly tbh
like on Grok's site
was hoping with PPLX's RAG, search would be better lol
did that pic in the morning
four prompts were fine
then it started cutting the tokens
around a 2 grand
@quaint ravine
wow u managed to get a single prompt with >11,000 token response?
I thought I was special getting that lol
ya ig
normal token response is more like 500-1500 for most default prompts
I get around 5000 tokens for one of my Space due to custom instructions and complex prompt
and only GPT-5.1 Thinking comes close consistently to that 5000
other AI models don't follow that
used claude sonnet 4.5 thinking in that situation
or only very occasionally
Claude doesn't follow targeted token/word count as well as GPT-5.1 for me in PPLX
thats why the numbers are swimmin
2159, 2705, 1876, 1551
lol
well is it because the prompts got less complex?
like if u asked it 1+1..it's not gonna give u 11,000 token response
no basically because i was giving it strict instructions to explain every word and code
and write whole game series in unity as dumb example
right, but still, it can be affected by complexity of prompt
i did, seriously
and then demonstrated it consistently went lower and lower
ah that's interesting then
so with the identical prompt, u got 11,000 tokens vs 1900 tokens etc
identical is best tbh
even with almost identical its too massive
and with custom instructions?
yea, of couse
but no space instructions
this is Claude right?
just "Introduce yourself"
sonnet 4.5 thinking
yea
yea, it is
introduce yourself I've had mixed outcomes
sometimes it compeltely ignores it
but might just have been buggy
thats a weird one
at the time of testing
hence why I stopped using and use Space instructions only....altho that has bugged a bit in past too
use memories
they are like instructions
not at all
guys what are u talking about with tokens n stuff
Space instructions is better
tokens = amount of symbols
like ai memory
oh so characters?
ya
all pplx models have 128k limit i think
oooo i see
im extremely interested in complex prompts but i cant find any that i could use personally or edit personally
but as shows the practice, more limit doesnt mean more performance
anyway, DM me if u exceed 12,000 token single response in PPLX (my record is like you, around 11,000), and especially if it's non-coding related
i heard about how u can give the model a super complex prompt so it gives a really good response
do i get edible chocolate medal
you got me sweating
Not really. Too many directions & long instructions can mess them up. Just using natural language for what you want is more than enough
future is behind, models need breaks now
strange, a friend of mine was sitting in on a seminar that a guy was giving cuz he makes models using ai algorithms for market predictions and stuff and she told me that he has this really complex and long prompt that allows him to get better responses
hes a corporate guy working with big firms
yet using ai to solve stuff
I adapted a few custom instructions to Perplexity, give it a look later if you want
https://discord.com/channels/1047197230748151888/1441240513050054757
well thats his whole thing thats why hes wroking with big firms cuz its working ā ļø
thanks :D
do we like, drop it in space with deep research threads
as space instruction of course
thats how I use it
sounds smart
and if they have reasoning they follow it
i hear it's racist
to who? to models?
i saw a video of it calling someone a chimpanzee and then the hard r š
and yet you ask me why did i call grok as bullshit
nah i agree lol
we gon hide the fact that video was staged
i think its a relatively weak ai model but i havent done any research its just from what i've seen
damn
is that propoganda
unfortunately
my referrals disappeared⦠why?
they disappeared from pending but dont appear in processing
i agree - it's more complex than one thinks and probably related to search/RAG limitations too, especially for complex queries that do better with searching
no matter what I put in my custom instructions for this prompt "what are the latest AI models now?"
I can't get any AI currently to 100% always get this right, regardless of platform
sometimes eg. GPT-5.1 Thinking will get things 100% right in its response (even if it misses mentioning eg. Grok)
but many times, they will say Grok 3 is the latest
or GPT-5 is the latest (when GPT-5.1 has been out for weeks now) etc
uh
sometimes I've got Grok 5 as the latest lol...due to a website that reports Grok 5 in the table (mistakenly)
anyway, we are very far from AGI lol
thanks
Womp womp lol
On grok lol
oh alright
my b for no context
Hi everyone, I have a question. Today I tried using Perplexity normally, but I noticed that my conversations aren't loading all the messages completely. They're only loading the first few, so to speak, and the current history isn't loading. This is happening both in the app and in the browser, and even in incognito mode. It works fine on my phone, loading the entire message history. Does anyone know what might be causing this? I tried from a different network, and it's the same. I haven't tried from another computer.
grok always sucked, i dont know what are they even doing with the model
its always ass
People want it to say slurs
ofc it will go underneath
Hi!
Finally
Hey, guys, I have a question, what if my connected Google account was deleted (hacked in my case), how can I access my pplx account, if it requires confirmation code from non-existent email?
Try to write an email to the support
+1
idk what happened

@distant aurora https://youtu.be/JOx2wZm5DFg?si=gbCYFQMwg6rqCkSC
In today's video we'll be testing Gemini 3, Google's new LLM. We'll be comparing its answers to a couple of research maths problems that should be tractable for an LLM. There's been a lot of hype for this model, will it stand up to the test?
The most based channel
Someone who didn't Yolo his life savings into GOOGL
yea thought so
Wow that's good
that was my general feeling too for my use case(s)
which aren't maths lol
but I thought that might be the case
yep PPLX summaries of youtube videos are excellent usually
Yeah gemini 3 is definitely not bad but it's not some godly model that'll make other Ai labs shit themselves
GPT-5.1 Thinking is superior ime
The true "OH shit!" event for Ai was when deepseek released r1
it's all just incremental improvements as I thought
around June/July 2025, I predicted that if by end of 2025, there wasn't any big Grok or Gemini breakthrough, that the plateau is confirmed
i think Grok is the most telling, as Elon wouldn't hold back
i still like to hope/immerse in the hype of Mo Gawdat etc
For now it's up to openai to show if their new reasoning methods and rl methods are that much of an improvement
yes hopefuly Pro get access to Nano Pro eventually too with limits
as i said before, even Gemini free has access
Google can eat the cost since it's their own product while Perplexity uses the API which is quite pricy for an image model
how does the cost compare with GPT Image 1?
isn't that cheaper than GPT image 1?
hey is the kimi thinking model useful or not
You saw the news?
Nursing declared not professional and other fields too
^
Wow Perplexxity has already added Gemini 3 Pro to the models list. Addtionally they added Kimi V2 and it added "hosted in the us" for some reason maybe they're running the model themselves if it's open source i've never used kimi
ye +1 idk why
Some one tell me what actually kimi is another ai model like chatgpt or geimini? or like grok and claude
What can you say about the new Grok 4.1 in pplx?
Same
not tried yet
did anyone get payout of comet referrals? @everyone
i think payout is bugged?
I don't see either Grok 4.1 and Kimi K2 better than GPT-5.1 Thinking in PPLX
and often they are worse
at least for my use case
would be interested to hear others
no my 125$ is processing for 6 days
no clue
is that your first payout?
I have perplexity pro and it states that I've hit my monthly limit which is strange, did the amount seriously decrease lately?
yea on Paypal
i already said and again saying it takes around 10 days
after penfing is completed
oh- there is said that it takes 5 business days and thats why i am curious
yes 4-5, for processed and then 5-6 for stripe
10 we can say
but im paying out on Paypal
idk abt paypal sorry mate
it is not my first payout ( from others ) so it could get fast
my one referral disappeared js rn
bruh
ic
COMET IS OUT FOR ANDROID. idk since when tbh. it just popped up home page as i signed for auto install as soon as its out
yeahhh the assistant experience is not as expected
will give it time
mail them even i did
damn still no google ai pro for my uni
that's sad
yes i need more notebook lm
Damm
Only people who can help are support. Reach out to them.
Youāre not eligible?
nope. its still not in my country
Is Grok 4.1 not thinking?
And Kimi is self-hosted š
yes no thinking version for Grok 4.1 yet on PPLX
otherwise there'd be a reasoning toggle like with GPT-5.1, Sonnet etc
or a toggle that is permanently on like with gemini and Kimi
Hey, are y'all getting rerouted to Best after trying to use the thinking model of Claude Sonnet 4.5?
It's happening for me
yes a lot of people have been reporting that since about Nov 1st
Huh? It only just now happened to me..?
i've noticed basically 1-10 reports of that per week across just reddit and PPLX discord here
How to check? The indicator shows right Sonnet Thinking
They are scamming us out of our money š¤¬
what model does the lab and research model use
It barely took me time to realize
@dry kestrel you smart dog
I just realized something funny
Dms
kimi is self hosted because
- its open source
- its asian
so whats the problem??
theres no damn difference?
but yea, grok 4.1 is without reasoning
but
BUT
it doesnt matter
grok always sucked unfortunately
Hey, actually I used the Study and Learn mode on the web interface but I cannot access that mode on the Android app. Are you guys facing this issue? If somebody knows the answers please reply!
thats all models
see: #š¬āgeneral message for grok comparison to others
screenshot?
What screenshot?
Maybe the model will be outdated soon, maybe it works better on Chinese servers 
See
same for ios?
isnt study is just regular search with more source searching
I used Google AI to craete an image of how my blindness looks (eventually it'll be fully black but for now it's like this)
Why this feature isn't available in android app?
it isnt available for ios either
so yea
damn?
does the image like
