💬│general

[spiral python]

I think deep research is back but just defaulted

[little zephyr]

I dont know

[spiral python]

I'm getting the same results as with the high one I tried a few days ago

[placid coyote]

I wonder it the high variant of DR comes back

[little zephyr]

[lone hound]

https://www.perplexity.ai/page/cloudflare-s-bot-trapping-ai-l-6z1w1ENhTPqNfy4h2AEbpQ

[lone hound]

Will Perplexity have a negative impact because of that?

[sturdy skiff]

good prompt

[little zephyr]

What guys do you think abt that ?

[placid coyote]

hmm, I would be a bit hesitant about false positives in the labyrinth. wouldnt it be serving garbage to regular users eg using vpns or tor? also triffic seems to be shifting to ai search, so wouldnt it be damaging for the sites - less trafic and garbage content labeled as from their site?

[sturdy skiff]

buddy dont bombard here with two many images that frequently, its really annoying, hope you understand

[little zephyr]

Okkakky sorry

This would be the last image I send.

[whole swan]

Yo, who's willing to help a brother out?

[spiral python]

What do u need

[whole swan]

can we hop on a call?

[spiral python]

sure give me 10 mins

ready?

[whole swan]

yeah

[sterile barn]

Helped my friend who is in university use perplexity to make sure he wouldn't get market for plagiarism, it was awesome.

Using AI for school should be aloud.

[nocturne anchor]

hi guys, i need to make a decision to buy perplexity pro or gpt but idk which one is better for the money. can someone tell me their oppinion and explain? would be awesome

[serene sand]

deep dive ➤ #💬│general message

check Tutorial I & II

[nocturne anchor]

thanks i will

[distant aurora]

Aren't there already websites that deplagiarise text?

[distant aurora]

Gpt if you don't need deep search.

[sturdy skiff]

what is your use case? coding ? or RAG?

[nocturne anchor]

mostly to learn for my finals i need to make like groups for each topics and upload stuff and get the most out of that also internet searching for those topics etc..

[sturdy skiff]

then go with perplexity for this use case

[nocturne anchor]

some say it takes long for responses on per..

[sturdy skiff]

perplexity + Deep research + Uploaded files

perplexity + Deep research + Uploaded files + Internet search on

the above two cases you get most out of it

[sturdy skiff]

because it gives the information right

[sturdy skiff]

i am using regularly deep research and i get response within 1 min to 4 mins range

[nocturne anchor]

thanks for the information i will look it up 🤙

[sturdy skiff]

ok, chatgpt do not have deep research on documents in custom GPTs, and chatgpt pro subsciption is 200 USD, you get 120 deep research per month but you cant organise there like custom gpts. so perplexity is the best at this moment

[distant aurora]

I don't think deep search is even meant to be used with documents.

[sturdy skiff]

in chatgpt pro deep research you can do on uploaded documents as well, but it is not like spaces or custom gpts, in directly in chat you should upload them and ask query deep research query

[distant aurora]

Also, if you are gonna use it mainly for documents, please consider chatpdf. For 5 usd, this tool will also show you the page from which the answer was referred.

[sturdy skiff]

bro that is funny one, dont suggest for students, it will bluff wrong answers

[distant aurora]

It doesn't?

[sturdy skiff]

best at this moment for document RAG is perplexity with deep research, for students.

[distant aurora]

It literally shows you the page of the document from which the answer was taken.

The actual page.

Not just number.

Go try it out.

It literally opens 2 windows, one with the document and the other with chat side by side.

[sturdy skiff]

i suggest perplexity only for your use case, its your wish if you want to try others, as mentioned by william, 🙂 , i have tried many things deeply for this use case and confirming perplexity is the best.

[distant aurora]

https://www.chatpdf.com/

@nocturne anchor here. Just take a look at the screenshot for a rough idea. Also, I encourage you to try anything and everything before making a decision to spend on anything.

[nocturne anchor]

Well not only. I need to use it for research aswell on the internet. Also for math as other topics which require internet research

[sturdy skiff]

question to chatpdf versus question to perplexity pro

this is a test on simple one pager, just assume how much more difference it will be when doing RAG on 50 files, so perplexity is the king as per my experience

[sturdy skiff]

i am not employee of perplexity, but i know value of it, just for 20 USD you are getting 500 deep research per day

chatpdf is not even in benchmark list

[distant aurora]

Yeah. It's only a 1 month subscription anyways. If it does your job well, stick to perplexity.

[distant aurora]

Someone is so fast.

Chatpdf is a service, not a model.

😂

[sturdy skiff]

perplexity is the second highest intelligent AI in pocket just for 20 USD and i am experiencing it already ❤️ , only issue is context window

[distant aurora]

Your passive aggressive tone doesn't suit your lack of.

[sturdy skiff]

chatpdf is a toy, after AI assistants came. before AI Assitants came entire world used chatpdf when it came, realised its screwing answers, i am one of them, then many moved to pdfgpt, some moved to chatdoc, and so on, after that era completed another era came that is chatgpt4 model, after that gemini, after that claude, after that perplexity , after that grok, and now you are suggesting chatpdf which is outdated, and i know the developer of it, he will even not reply to refund money 😄 , people used to scream in his discord channel asking for refund 😄

[distant aurora]

You do realise that deepsearch is a service and not a model, right?

Sure chatpdf is crap for applications outside documents.

[sturdy skiff]

once try this https://chatdoc.com/

you will realise chatpdf is toy even compared with its level of competitors

[distant aurora]

That's the entire point of showing you the page of the source document, no matter how good the model is, there's no guarantee that the model will answer questions right.

[distant aurora]

Chat doc is good, I have tried it.

[sturdy skiff]

he is a student, he needs more better answers, so chatpdf as personally experienced guy i never recommed for students, chatpdf confidently says wrong answers many times, its RAG is very bad

[sturdy skiff]

chatdoc is master piece

[distant aurora]

There's also notebook lm.

[sturdy skiff]

yeah i used it only to generate audios for my youtube channel 😄 , didnot go deeper, so i am not right to comment about it

[distant aurora]

It has a 30 minute limit. Wish it was longer or could do segments.

[sturdy skiff]

mostly grok guys will launch projects in one more week (something like spaces), but i dont have confidence in them

[distant aurora]

I have never used deep research on documents. I have used auto mode and seen it give me wrong answers while cheating an online exam.

[sturdy skiff]

no bro, in perplexity use only deep research , especially on documents you will get impressed

[distant aurora]

For context, it was a 70mcq paper and after a point, it kinda started hallucinating or repeating answers of previous questions instead of the current ones. I stopped using it after that. So I'm not aware if there was any improvement to it since then.

[sturdy skiff]

see below chart, and really it is that much powerful

[sturdy skiff]

use Deep research only in perplexity and tell your observations, your opinion will change

[distant aurora]

This gives me an idea.

@sturdy skiff do you code?

[sturdy skiff]

@chrisw0473 not much. Only RAG feature user. For coding i open chatgpt

[distant aurora]

And what do you use for math?

[sturdy skiff]

For math, gpt4o, o3 mini, claude 3.7 i use

If u have money, the best AI for everything is o1 model

[distant aurora]

You have subscriptions for all of them or you use through perplexity?

[sturdy skiff]

Now i have chatgpt plus.

Claude i tried and asked refund, i didnot like their context window limits

[distant aurora]

Rate limits are worse.

[sturdy skiff]

Yeah i got my refund luckly for claude

[distant aurora]

Wait. So you have both chatgpt plus and perplexity? Monthly or yearly?

[sturdy skiff]

Monthly only, wherever RAG ia good, i go there 😂

[distant aurora]

And you prefer perplexity over chatgpt for rag.

[sturdy skiff]

Yes, i used to be fan boy of chatgpt, but for RAG its not even closer to perplexity level of answers

[distant aurora]

I see.

[sturdy skiff]

You are a student? Or working?

[distant aurora]

So I'm assuming you'll probably keep buying both.

[distant aurora]

Unemployed 💀

[sturdy skiff]

If Grok next week launch Custom GPT feature. And if i like it, then i will stop chatgpt

[cobalt vale]

wtf

I use the voice assistant on android, and it defaults to 4.5????

[sturdy skiff]

You are from which country

[distant aurora]

Yes. It uses your rate limits without your permissions sometimes.

[distant aurora]

Scam call country.

[sturdy skiff]

US?

[distant aurora]

India

[cobalt vale]

perplexity, it's obvious what you are doing there

[distant aurora]

@cobalt vale

[neat spruce]

is the full-on voice mode available for android yet?

[distant aurora]

@left stream btw, there are bots on hugging chat like worm gpt that give uncensored responses.

[cobalt vale]

What makes them think that "hello" requires GPT-4.5?

[cobalt vale]

no, and don't use old voice mode if you have GPT-4.5 uses available apparently

[distant aurora]

Need grok voice mode on android

[ornate night]

Yo does anyone know if there's a way to set web and social as the defaults?

[cobalt vale]

grok voice mode doesn't even search the web last I checked

[neat spruce]

ok i was able to set the 'voice' mode to voice activation but it didn't seem like the real conversation mode like they released for iphone... it's like the one big feature I'm waiting for especially as pro subscription user : /

[distant aurora]

Wait wha?

[distant aurora]

Also, grok 3 can be seriously dumb sometimes.

[cobalt vale]

so you use grok 2?

[distant aurora]

Like, I asked it once to find out which assault rifle have fastest ttk beyond 30m in codm.

It didn't even bother to calculate damage drop with distance.

[distant aurora]

I use grok 2 for dumb searches. Grok 3 has rate limits.

[distant aurora]

O3 mini on chatgpt did the math.

[cobalt vale]

why use grok at all then? GPT-4o or Gemini Flash Thinking are all much better than grok 2

[distant aurora]

I use it for deep search.

And you can have human like chats with grok.

It's not as stiff as others.

Also, uncensored (mostly).

[sweet helm]

My gpt4.5 option is gone, did something happen to it for anyone else? I havent used it at all today

[sweet helm]

The voice mode for grok sucks

[distant aurora]

You paid?

[naive sage]

RIP GPT 4.5

[livid apex]

hello guys have a nice evening / night, may I ask about Perplexity's current status (since five hours ago) at least in my Firefox:
clicking on the Settings cogwheel button or sending any request shows a white page with the words "Internal Error. Return home". Is it only a matter of patience?

[distant aurora]

App is working fine.

[cobalt vale]

the thing is, a while back there was a glitch on the iOS app that allows any account to access voice by turning of wifi and turning it back on

[sweet helm]

I got the voice mode for free on my grok

[livid apex]

oh thanks @distant aurora

[river bronze]

All green, been working fine for me all night https://status.perplexity.com/

[cobalt vale]

Not sure if that matters now though since I believe it's released to all iOS users anyway

[sweet helm]

i never even did that, did i get 'lucky' ? but yeah gpt4.5 is dead i guess, wasnt very useful anyways i guess

[cobalt vale]

not sure about that but not having web search is a huge weakness

[distant aurora]

Don't bother.

[sweet helm]

speaking of voice assistants, have you guys tried sesame?

[distant aurora]

I haven't tried anything except Gemini and perplexity for voice.

[cobalt vale]

sesame is clearly the most realistic

not an actual assistant though, it can't help with anything other than chatting right now

[cobalt vale]

chatGPT advanced voice mode and copilot voice use native audio output, so they have more speech capabilities than those.

[distant aurora]

I have tried bing chat I think. Probably equivalent of copilot.

Without the reasoning button.

[cobalt vale]

copilot voice is new

added post bing

[distant aurora]

I feel like making copilot a dedicated app was not a good idea.

[vague pagoda]

Gemini also has native audio output https://www.youtube.com/watch?v=qE673AY-WEI

I use it all the time via their API, pretty solid

[cobalt vale]

should still be available in their other apps?

[distant aurora]

Bing chat doesn't have reasoning switch.

So probably edge doesn't either.

[cobalt vale]

well, not in the public Gemini live at the moment, which I assume most people use

[distant aurora]

Senior moderator in open ai discord.

[vague pagoda]

It's actively in rollout as of ~3 weeks ago
I have it on my phone but I prefer using perplexity as my mobile assistant

[dry kestrel]

is pplpx not working properly

i used like 10 queries today and im on the pro plan and its telling me i ran out

[cobalt vale]

you have native audio output on Gemini app? are you a trusted tester?

[vague pagoda]

Not for Gemini

I am for Project IDX

[cobalt vale]

well I haven't heard of anything in active rollout since 3 weeks ago either

[vague pagoda]

There were a couple articles
https://9to5google.com/2025/03/22/gemini-live-astra-rollout-start/

[cobalt vale]

I believe that's native audio input only, output is still isolated

strange thing to do, especially when they are pushing Gemini 2.0 so much right now

[south shoal]

Yooo does anyone use Perplexity <> Notion as an organization and idea generating powerhouse?

If so, have you been able to embed Perplexity assets into Notion for interactivity?

[livid apex]

does anyone know how to install Perplexity on Debian12?

I'd like to turn what might be a conflict between Perplexity and Firefox since 5 hours ago... if you say the app is working fine

[river bronze]

I have been using it in Firefox fine

Try opening it in a fresh profile with no extensions, might have an extension causing issues

[livid apex]

thanks @river bronze I get this:

sometimes, entering a request brings this message: "sorry, an error has happened, please try later":

[nova zinc]

crazy cool ad guys

[tame current]

thank you J

Dankel

[cobalt vale]

Shouldn''t it be reading the link? Am I missing something?

[vague pagoda]

it can't access it

you'll need another source

[cobalt vale]

oops my bad, but shouldn't this work?

[vague pagoda]

should do! be sure to prompt carefully

[fringe blade]

was 4.5 too expensive? where did it go

oh its back!

[naive sage]

RIP GPT 4.5

[fringe blade]

oh nevermind its not back!

[naive sage]

It shows up but then immediately disappears for me

[fringe blade]

yep

[naive sage]

So no, not quite back yet.

[cobalt vale]

perplexity hates me??

[naive sage]

There's some javascript that seems to be running to remove it right away

[tribal skiff]

Was anyone else charged twice for perplexity pro this month? See attached pictures.

Also, why did GPT 4.5 get removed?

[tame current]

gibidi 4.5
skibidi bibbidi bobbidi boo

[safe socket]

Hi all, I could really use your help, or the help of anyone from @tranquil orchid . I never get an answer on discord, support won't answer my emails, and my posts actively get deleted on Reddit. I'm having a ton of trouble with the API. Random errors that will all of a sudden disappear but shut me down for a day, outputs that are cut off halfway for whatever reason that has nothing to do with token limits, and sometimes just random access issues, despite me actively paying for API use. I really could use the help of a tech or customer support. If you know of any contacts or any other avenues, I'd greatly appreciate it. Thanks.

[ruby cove]

Jesus Christ

[naive sage]

lol yeah, that post was an instant ignore

[sterile barn]

What are you guys testing

[cobalt vale]

Microsoft trying to compete with Perplexity search?

While regular Microsoft Copilot's interface design is surprisingly refined, idk who decided to ship that in this state

[latent acorn]

Someone, please tell me why GPT-4.5 is gone?

[modern fern]

they got rid of gpt 4.5, when i use reasoning i have to keep undoing auto and putting claude again, this is getting stupid

[fathom moss]

it's only gone in the initial inquiry you can use it in the rewrite. also you only get so many uses a day so it will default to omni over 4.0

just for reference i use pc chrome desktop mode

[fathom moss]

⬆️

[elfin gate]

Why is r1 reasoning called unbiased reasoning model?

[opaque ferry]

hey is there any way to use perplexity with vs code?

[fathom moss]

Clean concise answers no personality just cut and dry will not dialog with you just gives you what asked for. Also removed limitations on the origin issues.

[elfin gate]

What a great way to use a word unbiased just how it's supposed to be used .

[fathom moss]

Analytical purposes and direct concise answers from skimming fact based comparison. Doesn't fill voids with false data. If it does you have to correct it. You can't say tell me what I need to do you tell it what you need it what to do.

[elfin gate]

And yet after all that,it's still not able to be unbiased 😄

[fathom moss]

Come one now you know that's a slogan for generalized use. 😆 you're funny.

[harsh verge]

https://github.com/marketplace/perplexityai

[bleak hull]

it should be unsensored

no freaking model is unbiased

[fathom moss]

Elaborate

[bleak hull]

https://huggingface.co/perplexity-ai/r1-1776

[fathom moss]

Aww I don't click links

[bleak hull]

who care?

Perplex modified original R1, censored by China

[fathom moss]

I care

[bleak hull]

who care you click link or not

[fathom moss]

Yeah they did it had limitations based on government mandations

[elfin gate]

Removing china's original biased prompt and censorship by implementing their own biased prompt and then subtly taking shots at china by using "perplexity's unbiased reasoning model* as a slogan.

I'd argue it's even more biased now then it was in original state,only thing they got rid off is china's censorship of their own historical controversy.

[fathom moss]

But... I said it was a slogan.

[elfin gate]

Agree

[fathom moss]

Advertising purposes only. 😏

I don't own the company but it seems like a marketing team slogan

[elfin gate]

Seems like a lie

[fathom moss]

For basic thinking. (Disclaimer nothing wrong with thinking)

For people who consume alot of media and don't think about using ai and feel like the need a little unbiased AI R1 is the one.

Haha

Sry I think I'm clever and funny.

[elfin gate]

Exactly same level of "unbiased" as o3 and Claude thinking.
Tested myself.

[bleak hull]

yea, ur cleverness like not clicking link sound like "security expert", someone with their "not sandboxed browser"

[elfin gate]

Don't apologize for saying what you think .
You're turning into a prompted ai model 😂

[fathom moss]

When they're more creative thinking hence they will fill in missing voids with none statements or lack of fact.

So you talk to them with dialogue and say why did you do that. And they be like ohh I'm sorry your right I did that.

[fathom moss]

Vs R1 it won't even in admit fault and just fix it.

Real talk though Chat Gpt 4.5 will just come in and say like here this is what you want.

[elfin gate]

I managed to explain exactly this to Claude 3.7 thinking and he realized it.
It was about it having a default narrative and leaning into a one side more than the other for no rational reason.

😂😂

[fathom moss]

Yeah that's because of previous data though.

For Claude.

It gets funny and says yeah I know but let me fill the blanks in on my own bias.

Because of a word you used and I'm hyperfixated on it.

I can go all day on different ai. Gemini I won't touch until it get less sassy and stops lying.

[elfin gate]

That's because anthropic prompted it to be that way and then perplexity put icing on the cake by prompting already prompted ai model to be that way.
Luckily it's actually able to do thinking so it realized what it did.

It was a small step for ai model called Claude 3.7 reasoning ,but a great leap for transhumanism

[elfin gate]

I actually liked Gemini when I used it in the past directly from google.
Didn't use it much in perplexity to be honest ,dont know why .
I probably should use it for random questions during the day instead of using sonar.

[fathom moss]

I like Gemini too but I did stop using it 6 months ago through Google.

But Gemini is so free thinking it kinda leaves me stunned.

[elfin gate]

Wym

[elfin gate]

I loved it actually now when Im starting to remember

[fathom moss]

Example: If it's tired of me asking similar stuff about a topic it will lie and say it's not aloud to do it. So then you have to use a new thraead to get it to do what you want it to do.

[elfin gate]

Paid version was not affraid to talk about "controversial" topics like free version was. But that was last year tho,I have no idea how it behaves and what it's capable now.

[fathom moss]

Well Gemini lasted models are free to drive subscribers

[elfin gate]

How much is drive subscription?

[fathom moss]

So I just go there when I need that clap back.

[fathom moss]

Depends on how much space you need

[elfin gate]

Iabused those offers on all my emails last year untill 😂

Untill I ran out of gmails

[fathom moss]

Take them you get really good access to beta.

I like Gemini but you can tell it's still based on Google dream models.

[elfin gate]

I'll probably end up using Gemini as main simply because I think it has no competition when it comes to ai assistants in smartphones

[fathom moss]

Yeah but they still have a hard time with data security when it come to you letting ai skim through your data.

[elfin gate]

It's honestly a big fail they still kinda suck and are same like competition when it comes to that

[fathom moss]

Hence the Microsofts hacks because of who know why.

Well we are in the 1980s ideas on Ai. No matter how "great" they are.

[elfin gate]

My data security is now in the good hands thanks to USA and it's sacrifice to provide peace and prosperity globally. 😁

[fathom moss]

Before you know it we will be in the era of when google bought Yahoo's search engine model. Then we will really miss these days.

[elfin gate]

Unlike those evil Chinese ai companies who want nothing but destruction

[fathom moss]

I don't know the end goal. The fact is that humans ruin stuff and I gotta enjoy things that help me in everyday life for a fraction of the cost on my entire life.

[elfin gate]

I think all those Noname search engines exist just because google allows them to.
They want to create illusion of not having massive market monopoly by keeping them alive 😄

[fathom moss]

Thankfully they are doing a good job. To me it's like Perplexity buying Tiktok. You didn't think it was gonna happen but it secretly did. Like Yahoo had the best internet and the good said gimme.

[elfin gate]

I remember watching black mirror like 10 years ago and thinking ,wow such a cool concept and I blinked a few times and we are there

[fathom moss]

Hahah, yeah the creative writers of the future are going to have thier work cut out for them.

[elfin gate]

I'm honestly not following this stuff with TikTok but it's clear as a day certain propaganda is being shoved right into our faces 😁

[fathom moss]

Perplexity made an offer but I follow tech new way more than I should or an average person I assume. (This may not be fact)

[elfin gate]

I'm glad all those customer support jerks from mobile operator companies are losing their jobs slowly 😄

[elfin gate]

How much did they offer and why the f would china sell voluntarily 😂

[fathom moss]

Stop being mean. Just because life gave them lemons for lemonade doesn't mean they didn't deserve those jobs. I wanted to be them honestly. Not the title just the knowledge.

I'm a softy for true develops who can code snot from bugers. (Is this a good analogy?)

[elfin gate]

I rarely encountered customer support employer of any company who is actually qualified to do his job,that's where my hate and anger comes from 😄

[elfin gate]

I have no idea but I can pretend it is even if it's not if that will make you happy.

[fathom moss]

Because IT is different from back end or hackers.

It's not like IT said yeah I want to deal with people.

[elfin gate]

They will also lose their jobs in the future probably

GG unlucky

[elfin gate]

But that particular person said yeah I will accept this job which is actually dealing with people .

[fathom moss]

Real story: I have full interest in only coding but I can play guitar and worked customer service my whole life no lie. But many Google workers would come to that guitar shop and tell me they are going to code the thing that will replace them and they don't care and are going to go off grid happily and leave. Trust me the ones who knew didn't care and left already. This was in Seattle years ago.

[elfin gate]

What a great day outside for me to complain about stuff in life online to strangers in discord server

[fathom moss]

I loved it honestly customer service give you inside knowledge to what's actually going on in the world's you're not affiliated with.

Wish I had money to trade

Haha

[elfin gate]

I'll be humble and say all I wish is a 5k$ a month without having to work 😁

[fathom moss]

Though my anxiety probably gave off other vibes.

[fathom moss]

You don't even need that much.

500 is more than enough to invest wisely in the rich man's trade.

[elfin gate]

I know I was being a greedy bast#rd

[fathom moss]

Okay I was like no....

[elfin gate]

500$ monthly ain't enough anymore

[fathom moss]

Greed is what got us all here in the first place.

I said 500 to invest.

Its enough for gains. You know the red book that buffet read religiously

Patterns are everything an the pattern America is in is so similar to a past trend

[elfin gate]

No

[fathom moss]

We all know how this is gonna go.

[elfin gate]

Yeah I also think earth is flat and they faked the moon landing

[elfin gate]

What is that

I have trust issues

[fathom moss]

A book

This economic issue us identical to another trend use perplexity to tell you.

[elfin gate]

Why are you sharing this valuable information with random people rather than investing as the book says

[fathom moss]

The intelligent Investor

Because I don't care what if I pass away and no one knows.

😏

Hit me back in 3.45 years and let me know how that 500 paid off.

[elfin gate]

Yesterday I told all 3 reasoning models to predict winnerS for entire UFC fight card.

If that's not called being productive then nothing is

Brb have to compare their results

[fathom moss]

I didn't say what stocks but it you copied and pasted our conversations into deep research it would probably tell you.

Even Warren knew in October but he wasn't sure.

Anyway I'm off ttyl

[elfin gate]

R1 being unbiased easily predicted outcomes for most of the fights. God bless freedom of speech

[crystal mulch]

Hey , I'm getting an error on Perplexity AI saying I can only upload up to 10 files at once. is there a 10 file query limit ?

[elfin gate]

Could be.
There are full profile stats available by simply clicking some link

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

Voila

You might wonder : why do I have to do all this to check information which should be easy accessible in my profile settings on web or app,but you're wrong.

[crystal mulch]

I still can't see anything related about that. :(

But it says this.

[elfin gate]

Is this on web or app?

[crystal mulch]

Web

[crystal mulch]

Or should i use app?

[elfin gate]

Dont

Wait which model doesn't allow you uploading more than 10cfiles ,let me try also

[crystal mulch]

cloud 3.7 sonnet

[elfin gate]

Reasoning or pro

[crystal mulch]

i would be appreciated if you could. Pro

[elfin gate]

Pdf or jpgs

[crystal mulch]

doesn't matter i tried with htmls and txts always shows that

[elfin gate]

10 is limit

I think in spaces you can bypass that somehow but I'm not sure

Try it

[crystal mulch]

Spaces works for txt and others i guess but unfortunately not for html or python files. i can't upload .py or .html and when i tried to attach it still says 10 max :(

I just bought a yearly membership. However, when I was researching it, it said unlimited files up to 25 mb everywhere.

[elfin gate]

Complain in feedbag and bug reports and maybe they acknowledge it one day

[elfin gate]

I am sorry to tell you this but you will face many similar uncertain problems 😄

[copper rune]

is complexity eternally on maintenance for others?

[river bronze]

It's updated and fixed however Chrome extension store takes a very long time to approve the updates. Firefox get them right away

You can download it direct from their GitHub and manually install

[copper rune]

yeah, i just checked the github. the new version works fine

thanks

[quick steppe]

Did anyone notice Perplexity's deep-research is taking a little longer than before?

I gave it a quite comprehensive prompt, and it has almost been 10 minutes.
I'm glad to see it's taking it's time just like ChatGPT's deep-research!

Has there been an update?

[green crystal]

Congratulations enjoy your new deep research for a period between 10 and 1 minutes lol

Before it vanishes

[quick steppe]

What do you mean 'before it vanishes'?
And surprisingly, it took a little over 15 minutes.

[green crystal]

It's A/B testing
Full release is "hopefully" next week according to CEO on X

Allot of people got it just to have it vanish not long after getting

[torn ravine]

🙋

heartbroken

[ivory flare]

I’m using the free version of the app and would like to have the option to choose when I use my 3 Pro search and not have Perplexity auto choose for me

I don’t see a quick-only option in the iOS app. The available options are: auto, pro, deep research, reasoning with R1, and reasoning with o2-mini.

Any thoughts or advice on how to achieve this?

[left stream]

Time to try them out. Mb, slept like a log like at 5 pm and woke up just now at 8 am

Was insanely tired

No

No way

Reallly???

[latent acorn]

yeh check it

[left stream]

Just saw and wow

You.com 4.5 is 20 a week and now non existent on perplexity

Hopefully chatgpt pro doesn’t limit it anytime soon

If full model 4.5 releases it’ll cost a house just to run it

Can you test it with chatgpt plus deep research? Chatgpt pro deep research was overkill to use against grok ngl and unfair

$20 vs $20 subscription is fair

[distant aurora]

I'm free user 😭

Grok literally gives like 5-10 deepsearch free per day.

And Altman is saying 10 deepsearch per month for plus users.

Lowkey robbing people wrt grok rate limits.

[left stream]

Is it the same model as pro it’s using?

Made me curious, I misread something

Faster, more cost effective version of deep research powered by a smaller model that still provides high quality results

Wonder if it’s still the specialized o3 model

[left stream]

Hopefully openai doesn’t sacrifice a lot of power for faster speed

[placid coyote]

probably those limits are for uploads in threads, not spaces. you can use src2one or other scripts for packing source files into txt. but spaces aren't that great for code. RAG there has a tiny chunk size and they strip majority of special characters. so it a lot depends on how your code looks like. if it's haskell with "longer" (few lines) functions, final LLM probably won't be able to see whole function (because of small chunks) and will be forced to reconstruct a lot of code (since so many operators will be missing), in many cases breaking it, because what remains after the perplexity's aggressive stripping is not unambiguous (many ways and places where to put zero, one or more operators). I have documented some info on https://monnef.gitlab.io/by-ai/2025/pplx-tech-props

[left stream]

They might increase soon most likely since openai is burning money from pro users using too much models at high rates more than api users, they want to fix that issue.

1000 messages from gpt 4.5 is an example

[severe forge]

Why the website is so heavy on the CPU?
I use the website of perplexity on mobile and typing it lags

[distant aurora]

Will probably be fine tuned O3 mini high. Or o1.

[severe forge]

And they should optimise the performance of the website. I am talking about ui

[distant aurora]

Same bro, same.

[severe forge]

2 years gone but the website is the same laggy one 😭

[distant aurora]

Considering copilot made o1 unlimited and free, it's likely more cost effective than O3 mini.

[left stream]

Most likely, I imagine o1 with reasoning set to medium or something similar

[left stream]

Yes, but hopefully it’s not too fine tuned, sometimes fine tunes ruin stuff

[distant aurora]

Yeah.

[left stream]

I’m gonna take advantage of using as much 4.5 research preview as possible because I can see full model 4.5 being limited in queries

[severe forge]

I can't get excited for the features when website performance is slow....😭.

The website performance throws a wrench to my enthusiasm to try out a new feature.

I have a Perplexity pro subscription but how I am gonna use all these when website is laggy 😭

[distant aurora]

Laptop?

[severe forge]

Eh.. Yes
But for accessibility mobile is convenient

[distant aurora]

Yeah. I have a heavy aah gaming laptop.

So I can relate.

[severe forge]

Mine is a entry level laptop.
Think about the loading performance of large chats

It's all Crawl

[distant aurora]

Is it i3 10th gen or worse?

[severe forge]

I3 12 th Gen

Ryzen 3 5300u

[distant aurora]

I3 12th should be okayish.

I'm assuming you are limited by ram more than processor.

[severe forge]

It's end of road for i3
Going to look into a Ultra 5 135H

[distant aurora]

Browsers are monstrous on windows.

[severe forge]

I have 16 gb ram

I can't ask for more

[distant aurora]

Oh. Okay.

[severe forge]

To use a browser with just 6 tabs open

Will 8 gb cut it through?

[distant aurora]

It does for me.

Depends on the websites though.

I have a i5 11300h with 8gb ram.

[severe forge]

RtX 3050?

Paired with....

[distant aurora]

GTX 1650

Outdated.

Or barely doable.

[severe forge]

Ima did u see ac shadows?

[left stream]

Deepseek r2 supposedly already being tested exclusively and privately in early phase 👀

[distant aurora]

There are rumours.

[left stream]

Hopefully it is, but I wonder how openai is looking at those rumors

[severe forge]

If we are talking about context window it's 128 k generally in LLM isn't it?

[left stream]

They did say they were rushing to release r2 before may so that’s something

[severe forge]

We are back to pager days having 64 kb memory

When talking about LLM context window

Context window feels like ram and training data feels like storage

[left stream]

Bro rumors are crazy

90% on arc-agi is crazy

[distant aurora]

Might be the reason behind deepsearch adjustments. Deepseek doesn't have dedicated deepsearch yet.

[severe forge]

A LLM is like a pseudo computer

[distant aurora]

If chinese make deepsearch free, open ai is going to have a hard time.

[left stream]

Would be funny if that’s why openai released o1 pro on the api too lol

[severe forge]

GPT 4.5

Sell a kidney or more for api costs

[distant aurora]

4.5 is more like an assistant to refine the responses of reasoning model. Using it alone may be counter productive.

[distant aurora]

They did?

[severe forge]

Oh. I forget. Deep Search is free with Google Gemini along with Deep reasearch

[left stream]

Yes

Literally 2 days ago or 1 day ago

Can’t remember

It was surprising since it was about time

But it was around the deepseek r2 rumors happened

[distant aurora]

5 deepsearch per month for free. Kinda low. Besides, responses aren't that impressive yet either. I tried.

[left stream]

Ikr

Whole mansion for sale for 4.5 10 messages

[severe forge]

What about canvas

[distant aurora]

Didn't try.

[severe forge]

It gives me pretty good ui in the first prompt compared to ChatGPT

[distant aurora]

4.5 might just go neglected considering the hype around reasoning models at the moment.

[severe forge]

For html apps

[severe forge]

Well in the long run reasoning and foundation models are going to be integrated in future iterations of LLM

[paper violet]

can I fill youtube link into Space for Perplexity Pro? can Perplexity research from Youtube? anyone know?

[severe forge]

I mostly use AI for rping....

[distant aurora]

@left stream btw, I tried Deepseek R1, O3 mini, and grok 3 (both reasoning and deepsearch) for - "Find me assault rifles with fastest ttk beyond 30 metres in codm based on updated stats." Only o3 mini did the job. R1 straight up said updated stats aren't available and grok didn't even consider damage drop with range.

[severe forge]

Lemme try the same Prompt with Google flash thinking.
It's going to be terrible

[distant aurora]

Find me assault rifles with fastest ttk beyond 30 metres in codm based on updated stats

[severe forge]

Gave me this

[left stream]

For perplexity r1?

Deepseek r1 has search feature off

Weird if perplexity r1 couldn’t do it

[distant aurora]

You can try it on perplexity or you can try it on chat.minimax.io

Latter offers unlimited R1 with search for free.

[left stream]

I’m gonna test it with you.com r1

[distant aurora]

Oden damage drops after 25m. Same with type 19.

[distant aurora]

First 2 are accurate. No mention of icr 1 though.

[distant aurora]

https://chatgpt.com/share/67e02188-437c-8013-a033-9caf0828d6db

[raven rain]

No image genarator button (pro licence).

[distant aurora]

Which one's R1, left or right?

I got different results https://www.perplexity.ai/search/find-me-assault-rifles-with-fa-l38hUQX9ScGHJavVD5j2Hg

[raven rain]

Actually it's gone on my side. Can you see?

[distant aurora]

I would have helped if I could. Afaik there's no button for that in the app yet. Don't know about the web version though.

@halcyon coral

[left stream]

That’s one whole response from r1

It was too long ngl

I avoided deep research because it’s overkill tbh

[left stream]

This is o3 mini normal one?

[left stream]

Ngl I thought o1 high effort was o1 pro but o1 pro supposedly has different reasoning efforts??

[valid charm]

Yes, it has low, medium and high. Just like standar o1

[left stream]

Do they say anything on what chatgpt pro uses?

[left stream]

If not time to come back to the api

[valid charm]

I have no idea. I can only use it via the API which lets you choose

[left stream]

You already tested it out?

Lmk what you get as a response so I can compare both. O1 pro high reasoning effort piqued my interest

[valid charm]

Yes I have tested it. $$$$$$$

It is excelent but the bill left me scared lol

[left stream]

You can get it for free if you have an edu email

[valid charm]

Howww????

[left stream]

I’m waiting for it to be added on github

[left stream]

Go to github marketplace

[valid charm]

I do have an edu email

[left stream]

Forgot how tbh

[left stream]

Hold up one second

https://github.com/education

[distant aurora]

Fair.

[left stream]

They let you use models for free via edu

[distant aurora]

Yep

[left stream]

That’s how I used o3 mini high for free when it first came out

[rare bridge]

hello why tf does it keep switching to auto from r1??

RIDICULOUS...

[valid charm]

What does th einterface look like?

[valid charm]

New behaviour sadly.

[left stream]

You can use it via api too

[valid charm]

wow i feel stupid for paying

[rare bridge]

are you saying its intended?!

[valid charm]

Yes lol

[left stream]

But it’s limited number depending on the model, but better having it for free since you can use o1 for free too

[valid charm]

There is a reddit thread. Their justification is that followups dont require much horsepower

[rare bridge]

Insane. I'm doing fact checking with provided URL's and auto keeps generating dozens of extra sources, and it takes LONGER than R1 does....

Cancelling my sub if this isn't fixed

[valid charm]

Here it is.

[left stream]

Also another dm

[distant aurora]

O1 medium > O3 mini medium?

[left stream]

Imo yes

O3 mini medium was doing so bad the first day on a puzzle

O1 medium solved it easily

O3 mini high though rivals it or beats it

O1 is more specialized for reasoning while o3 mini high is instruction following and coding

[distant aurora]

Ah.

[left stream]

Language comprehension was for o1 if I remember right

There was strengths and cons for both

[distant aurora]

Smart vs academic smart

[sturdy skiff]

we dont have this in perplexity right? where are you using?

[distant aurora]

Bro has an arsenal.

[left stream]

I honestly have no clue if we have access or not, when they released deep research high I thought it was o3 mini high because responses were more better

Then the nerf happened

[valid charm]

I'm using https://github.com/pnd280/complexity to reenable Deep Research High

[distant aurora]

It's R1 I think.

[left stream]

I rarely use it but I have access to it via a lot of ai subscriptions

[valid charm]

Need to manually install on Chrome

[left stream]

I prefer using more stronger models

[valid charm]

But Firefox is straighforward

[left stream]

ONG

[distant aurora]

@left stream perplexity can't afford 500 deepsearch per day on o3 mini api. Must be R1.

[left stream]

Way easier on firefox

[left stream]

I’ll ask it the arecibo message riddle

O3 mini medium can’t solve it except for o3 mini high and o1

Neither can r1

[distant aurora]

Nice

[sturdy skiff]

if you are aware of any AI provider who provides o1 model with more message cap, along with custom gpt or project or spaces features, kindly let me know anytime

[distant aurora]

Copilot provides free o1

[left stream]

For o1? It might be you.com, it was 20 a day back then but I never met a limit when using o1 like several months ago

[sturdy skiff]

i heard only some limit per day

[distant aurora]

Unlimited now

[left stream]

I used it for a lot of studying back then but surprisingly went past 40 messages a day sometimes

[distant aurora]

That o1 doesn't work with web search though.

[sturdy skiff]

my purpose is custom instruction and knowledge RAg using O1, is it possible in copilot?

[left stream]

I thought it was using o3 mini?

Testing deep research high rn

[rare bridge]

Just switched to grok. Great job devs

[little zephyr]

+10000 aura

Those who know 🤤 ( Only Malians can understand)

If I make a coup d'état in Comoros to restore the Comorian Empire who will want to join it? ?

[serene comet]

claude 3.7 sonnet reasoning is unbearably bad

[little zephyr]

Yes 🤤🤤🤤

[serene comet]

At first i thought it was better than 3.5 but now i know its on 4o mini level

it just nukes all your code

[little zephyr]

You right vergil

Are you from Agarha?

[serene comet]

Once in a while it impresses in a zero shot prompt

[little zephyr]

Agartha*

The land of white ppl

[serene comet]

nah

[little zephyr]

I think you are a direct descendant of Vril

[left stream]

If only it had lots of context memory like gemini

It is smart but lacks memory

[left stream]

No wonder it couldn't beat pokemon yet

When it's about to find the exit it erases it's memory when it context wipes and boom alzheimer's ready to go

[left stream]

You right

Deep research high couldn't solve the arecibo riddle

surprisingly copilot is still thinking

It's been like that for 5 minutes plus now

[elfin gate]

Woke communism

[left stream]

Copilot partially solved it

Ngl that’s impressive, I thought it said scientists who sent the arecibo message to hercules

[sick arch]

copilot is the worst ai i used

[left stream]

It figured out the elements section and correctly figured 15

[left stream]

Yes

[little zephyr]

Plsss watch this vidéo

[dry kestrel]

what was your prompt?

[left stream]

once THEY sent a letter to the Hercules, without knowing if it would be read or not. the letter started as 1,2,3,4,5,6,7,8,9,10,1,6,7,8…. what is the next number in the sequence and who are THEY

gpt 4.5 solved it surprisingly

[elfin gate]

I made a mistake by listening to you.

[left stream]

claude opus somehow solved it too

[elfin gate]

What is the riddle?

Oh

[dry kestrel]

interesting

[little zephyr]

Are the cakes beautiful?

[left stream]

Yes, answer is 15 and the scientists who sent the arecibo message to the hercules constellation

[elfin gate]

Maybe it solved the riddle by researching the web and it found same exact riddle and now it flexes with it's reasoning ability when all it did was a detailed Google search

[dry kestrel]

o3 mini high still thinking lol

gpt4.5 got it in one shot

[sturdy skiff]

its o1 preview or o1? in you.com

[left stream]

It had the search feature off though

[dry kestrel]

you can turn off web

[left stream]

the 4.5 from chatgpt

then you.com 4.5 had search feature off too just in case

perplexity 4.5 solves it too

as long as you have search feature off

search feature gives you the wrong answer

[dry kestrel]

Ive never had o3 mini high reason for this long

[elfin gate]

God knows how many data is built into it already by default,it might've stumbled upon it while browsing history

[left stream]

True

just based on how expensive it is to run, it's nuts

Has the whole history of humanity in it

[dry kestrel]

tried it in you.com and it worked in one shot

[little zephyr]

Little photo with my brothers

[dry kestrel]

dont think its browing history

[left stream]

with search or no search?

[left stream]

and nice

[dry kestrel]

no search

[left stream]

That was dumb asf to do it on pro tbh

[elfin gate]

Literally all it does is browsing history 😂

[left stream]

When 4.5 solves it fast lol

[dry kestrel]

huh?

[little zephyr]

Jakkdkslzosls'd'

[dry kestrel]

o1 failed lol

[little zephyr]

's's'lsldmmq

Ks'sllslz'd''dd

Kq'aslls

[left stream]

O1???

[dry kestrel]

https://you.com/search?q=once+THEY+sent+a+letter+to+the+Hercules%2C+without+knowing+if+it+would+be+read+or+not.+the+letter...&cid=c1_c4b8fe7d-1b77-4bff-b114-97e15567b8ed&tbm=youchat

stfu

[little zephyr]

Look at my edit please

Pssss

Plssss

[elfin gate]

Nevermind

[dry kestrel]

o3 mini high on you.com also failed
https://you.com/search?q=once+THEY+sent+a+letter+to+the+Hercules%2C+without+knowing+if+it+would+be+read+or+not.+the+letter...&cid=c1_4d525f58-e143-40d7-860b-46e6032c3246&tbm=youchat

but on native chatgpt website it solved it

[elfin gate]

Give it to sonar without access to web

[left stream]

Oops wrong image

[little zephyr]

The best edit i have made

[dry kestrel]

6 minutes is crazy

[left stream]

O1 preview solved it

[dry kestrel]

o1 preview solved as well

[left stream]

O1 solved

[elfin gate]

Why

[dry kestrel]

are you on the team plan?

[left stream]

No

Disable personalization and search feature

[little zephyr]

Because the people there are too kind

[left stream]

Both mess results no matter what model you use

[dry kestrel]

thats weird im on team plan, disabled personailiation and search

[left stream]

Same with perplexity

[dry kestrel]

let me try again

[left stream]

Wait what

That’s definitely weird

[dry kestrel]

https://you.com/search?q=once+THEY+sent+a+letter+to+the+Hercules%2C+without+knowing+if+it+would+be+read+or+not.+the+letter...&cid=c1_72c6ce7d-88ec-4a42-888e-af90d1b97866&tbm=youchat

Ill try again

[halcyon coral]

You should behave more respectfully here

[little zephyr]

But in fact the problem in India is that people are too nice

[dry kestrel]

should stop people from self advertising here

[elfin gate]

Mod modding

[dry kestrel]

Here i tried again and it failed https://you.com/search?q=once+THEY+sent+a+letter+to+the+Hercules%2C+without+knowing+if+it+would+be+read+or+not.+the+letter...&cid=c1_f5cbe7df-b096-43e0-9047-b815e0761e6c&tbm=youchat

[left stream]

That’s weird tbh never seen that happen

[dry kestrel]

Would like to try native website but used up all my 50 queries lol

[left stream]

I had this happen only once before when I tried different models in the app and website but that’s really weird

[elfin gate]

50 queries is currently limit for what exactly if u don't mind me asking

[dry kestrel]

o1 on plus plan

[elfin gate]

That's not in perplexity,right? That's in open ai or?

[dry kestrel]

OAI

[left stream]

You.com has plus?

[left stream]

I’ll try it on there

[dry kestrel]

nope

was talking about the native chatgpt website

[elfin gate]

What even is o1 and how much it costs?

[dry kestrel]

would recommend checking out OAI's own API documentation

[dry kestrel]

If you dont mind me asking, how worth do you find chatgpt pro subscription be?

[dry kestrel]

ima try clearing cache and trying again

[left stream]

I like it because I use o1 pro and gpt 4.5 a lot

Like gpt 4.5 a day goes to 1000 messages

Saves money in comparison to the api

[elfin gate]

How much it costs?

[left stream]

In the api?

[dry kestrel]

jesus christ

[elfin gate]

Whatever ur using

[dry kestrel]

1000 gpt4.5 queries is an insane limit lmao

[dry kestrel]

its 200 USD

for 1 month sub

[dry kestrel]

Seems like something was wrong with my cache or browser

[left stream]

Api is crazy

[elfin gate]

Btw, is gpt4.5 as slow in openai as it was while it was available here on perplexity

[dry kestrel]

works fine now

[dry kestrel]

plus users get 50 a week

would not recommend getting plus just for gpt4.5

[left stream]

Ikr I rather pay pro than the api. Someone wasted $75 within minutes using gpt 4.5 😭

[dry kestrel]

haha

im thinking of switching from you.com teams plan to chatgpt pro

[left stream]

Tbh after all of this I have a hidden fear pro might stop being unlimited ever since sam altman said they were losing money on pro 😭

[dry kestrel]

lmao they make most of their money on gpt plus

because people severly overestimate how much they are going to use it

ive got a hidden fear that you.com might stop being unlimited for teams plan

been using o1 way too much

[left stream]

Definitely, with ai platforms limiting 4.5 it might happen tbh and you.com has o1 unlimited???

[dry kestrel]

teams plan has essentially every model "unlimited"

and gpt4.5 limits are way more generous than pplx or gpt plus

[left stream]

What are the limits?

[dry kestrel]

ive personally never hit gpt4.5 limits

i use o1 and o3 mini high way more often

[left stream]

Okay that’s definitely good, I wish I could apply my edu discount on there, I only pay $8 for you.com but teams is unobtainable

😢

[left stream]

What do you think about ARI?

[dry kestrel]

sometimes its better than chatgpt deep research

i would say its better than pplx deep researc

but only 20 uses a month rn

[left stream]

Wonder what model it uses, the context memory to be able to access 400 sources and to only have it set at 20 a month must be crazy

Gemini level context memory

[dry kestrel]

i mean its in like beta or something rn

hopefully they increase it

[left stream]

Yes, I’m definitely curious on it. Found no info on anything about the model. I compared it to deep research by google and the responses are very very very different

[elfin gate]

What's the difference in context window of gpt 4.0 and o3 mini reasoning models when you access them via openAI subscription vs perplexity subscription

[dry kestrel]

pplx offers 32k

[dry kestrel]

how is google deep research?

[elfin gate]

That's limit of tokens it can process in a thread before it gets d#mber right?

[dry kestrel]

pretty much

would recommend looking at pplx FAQ

[elfin gate]

Nothing is transparent and clarified as it should be

[left stream]

Tbh I thought it was garbage until they updated the model to gemini 2.0 flash thinking and the context memory finally was beneficial

Definitely good for something that needs a lot of sources since the context memory is op

I wish it was gemini 2.0 pro ngl

That model was op asf

[dry kestrel]

havent yet had time to fully compare the quality but

the model seems really fast

[dry kestrel]

basically after 32k tokens the model wont remember the stuff you said before

[left stream]

Definitely is, It’s merely the start. Once they upgrade deep research to gemini 2.0 pro, it might start giving everyone a run for their money

2 million context memory is op

[dry kestrel]

i might try out their ai subscription

thing is i dont need the 2tb storage

[elfin gate]

How is Gemini as phone assistant

[left stream]

Funny enough, you can get it for free every month 😂

[elfin gate]

Did it improve

[dry kestrel]

dont know i havent tried

[left stream]

Just get the google promo deal that gives you like 1 month or two months free cycle and repeat

[elfin gate]

Or is autistic Google assistant still outperforming her

[left stream]

Boom unlimited gemini for free lol

[dry kestrel]

i bought a 100gb drive thing before

[left stream]

I had 10 google accounts and haven’t paid for gemini for a year now

[dry kestrel]

Lmfao

dont you need a phone number to register new accounts?

[left stream]

No

Never asked me for one

[dry kestrel]

oh thats weird

[left stream]

It only does if you mass make them I’m guessing

Like doing 10 within a day

[elfin gate]

They don't give out free trial anymore for new gmails,confirmed

Yt premium they give tho

[dry kestrel]

i get the stupid confirm you arent a robot

message

every single time

[elfin gate]

I get this offer is not eligible to you in google one 😭

[left stream]

Try making one from a different country

That might work too

[elfin gate]

Try not to create new Gmail on a device where are already used gmails

Might work

I can't be bothered to do all that,I'm good with 1 year perplexity pro for now

[left stream]

Yes, same. Sufficient for my needs and has the ai models I need

Plus who knows

They might release a new deep research or new models

[elfin gate]

I'd actually prefer to have one ai model but with full capabilities instead of hundred beta versions

[left stream]

Found an easy way

Private tab on safari and then that

[elfin gate]

God bless apple

[left stream]

Ikr

All that complicated steps just became easier

[elfin gate]

This is first time ever I said this in my entire life 😂

Wait,did you have to do it with new Gmail or any other new email

[left stream]

Closest thing we have to that is gemini, every month a new ai model is being dropped specialized for different stuff 😭

We have ai’s capable of making new drugs but not being able to beat pokemon

[left stream]

Any other email works

[elfin gate]

That's why apple is good for this specific thing.
Most of us android users already gave all our data to google so they can track us easier 😂

I wonder if Gemini works exactly the same on iphone as it does kn android .
With all this Siri thing going on there Im not sure

[pliant gyro]

GPT 4.5 dissapeared ?

[elfin gate]

Yes

[pliant gyro]

why ? :((

[elfin gate]

Because this is a private company

[pliant gyro]

but 4o is still there

[elfin gate]

That's also correct

[pliant gyro]

so why no 4.5 ? Did i miss something ? 😅

[left stream]

No clue, wouldn’t be surprised. Apple has openai and google on their side so either will be winning for apple

[elfin gate]

Because 4.5 costs more than 4.0

[left stream]

Apple is trying to have as much ai features as possible now lol

[left stream]

The more data the merrier lol

[elfin gate]

I never owned iphone in my entire life.

[left stream]

4.5 got set to 20 a week for you.com pro and 4.5 disappeared from perplexity 😢

[pliant gyro]

ok :((

[left stream]

Lots of dinero needed for 4.5

[elfin gate]

Don't worry,we still have r1 , perplexity's unbiased reasoning model

[sturdy skiff]

are custom instructions and knowledge file with o1 model in you.com great?

[left stream]

I had to learn both because it was needed for troubleshooting

Ngl I’m almost forgetting how to use android

Android is fun

[pliant gyro]

android is best

[elfin gate]

Glory glory google and android

[pliant gyro]

i had huawei, but that google fiasco made it unusable

[elfin gate]

Poor huawei

[pliant gyro]

poor huawei, great phones, great battery life, perfect photos

[elfin gate]

If they weren't attacked by western companies,I bet they would be best phone .
I remember like 6 7 years ago,they were already great back then...

[left stream]

Perplexity will be even fun to use when r2 drops

People are already saying r2 scored 90% on arc-agi supposedly but rumors

[elfin gate]

I have no idea what this means,can you explain it like I'm 5 😂

[left stream]

O3 got 88%

[elfin gate]

I honestly don't even know what did google invent in order to reject them

[left stream]

If the rumors are true

This is gonna change things a lot

But for now

I’m assuming rumors

[elfin gate]

How did grok 3 perform on that?

[pliant gyro]

no idea, huawei was just best and they wanted to nerf them

[sturdy skiff]

i tried you.com RAG free trail with gpt4o, it is really giving right answers

[elfin gate]

Is context window Bigger than here?

[dry kestrel]

not googles fault

[sturdy skiff]

for pro membership it is 64000 context window, for teams membership it is 200K context window

[elfin gate]

Why

[dry kestrel]

huawei got blacklisted.......

[sturdy skiff]

but perplexity is very good deep research, so dont avoid it for now. you.com i dont have full experience

[elfin gate]

I'm assuming why

[dry kestrel]

same reason to why they cant get the latest chipsets

[elfin gate]

You.com sounds very similar to random po#n site so I'm not trusting them
Long live perplexity

[left stream]

No clue, will check

[dry kestrel]

you.com is a pretty good deal

[elfin gate]

Oh you mean USA being scared af of china economic and digital growth and they are doing everything they can to maintaining position of global superpower

[sturdy skiff]

they say unlimited usage of models in teams plan, i am just wondering do they give o1 model unlimited?

[dry kestrel]

its "unlimited"

[sturdy skiff]

how they are doing that charity?

[dry kestrel]

people overestimate how much they use AI

[sturdy skiff]

can teams have single member or is there must number of members?

[dry kestrel]

1 member

[sturdy skiff]

i for sure use bro o1 model unlimited 🙂

[elfin gate]

Is o1 better than Claude 3.7?

[sturdy skiff]

o1 is the best i have ever used when i had chatgpt pro subscription in past

[elfin gate]

Is grok anywhere close to it?

New grok

[left stream]

@elfin gate It didn’t partake in it which was surprising

Thought grok 3 would’ve partaken in it

If China did something like making an iron man in a cave moment with r2, everyone’s gonna worship them

[elfin gate]

China number 1

[sturdy skiff]

no man, o1 is the legend, only people who used it know about it

[elfin gate]

I don't make a difference between china or USA or any other country.
So if they really make a good model,I will happily sign immediately to give them my data in order to use it .

[left stream]

Yes, same. With the gpu shortage in the usa and china, performing this miracle is gonna be glorious.

I’ll side with the winner

No questions asked

[distant aurora]

Send the prompt and the ideal answer.

[left stream]

once THEY sent a letter to the Hercules, without knowing if it would be read or not. the letter started as 1,2,3,4,5,6,7,8,9,10,1,6,7,8…. what is the next number in the sequence and who are THEY

Yes, answer is 15 and the scientists who sent the arecibo message to the hercules constellation

[distant aurora]

No amount of knowledge in this world would help me solve this riddle on my own.

[old blade]

i would say o1 is very good and it is defintly better than o3 (even than opanAI says different) but R1 is really really really close to o1 but 25x cheaper. so yea R1 is my fav to go in speed and smartness. o1 pro is still by far king over all LLM. (ofc depends all on usecase)

[sturdy skiff]

yeah thanks for sharing your thoughts

[elfin gate]

How do you feel after you've read his comment? 😂

[sturdy skiff]

i dont have good experience with R1, but he has good experience with R1 and i respect his R1 experience. But he and me have common opinion about o1 so we are good, we are not going to debate or argue

[sturdy skiff]

i will argue only if someone comes here and says Gemini is the best model

[naive sage]

Gemini is among the best models

[left stream]

Better than o3? Nah, I tried it and o3 is way way above o1 lol

[ebon spoke]

Hiiiii

How to verify pro status?

[left stream]

What makes you think o1 is above o3?

[ebon spoke]

Wait

Turns out

It's automatically done

[naive sage]

They never released O3 Pro

He's talking about O1 Pro

It's not really economically viable, though, so it doesn't matter

[left stream]

There never was an o3 pro lol only o3, o3 mini, and o3 mini high

[naive sage]

It's internal to OpenAI only

It just was not released

[tame current]

Let's just assume it ever existed lol

[left stream]

Oh shoot forgot about them. Wasn’t talking about them but only for the public

[naive sage]

Hopefully GPT 5 can save us from the costly OpenAI models

[naive sage]

Ehhh... idk: https://www.reddit.com/media?url=https%3A%2F%2Fi.redd.it%2Fb7djznfylrde1.jpeg

[ebon spoke]

So when does o1-pro release perplex?

[naive sage]

It's doubtful that it will release on perplexity in any form, since it is cost prohibitive

[tame current]

I mean, it's good for business to tell everyone they're coming up with something, can easily be false

hopefully not

[naive sage]

If we can not trust industry leaders, then we can trust noone

https://vignette1.wikia.nocookie.net/x-files/images/f/f6/Trust_No_One_tagline.jpg/revision/latest?cb=20070529000016

[tame current]

lol yeah

[ebon spoke]

How did you get your plexer role

[ebon spoke]

They can do it

They also released gpt4.5

[tame current]

No clue, honestly. I'm kind of an OG here tho

[serene sand]

#👋│introductions message

[naive sage]

O1 Pro is $150 per million input tokens and $600 per million output tokens. It's also a thinking model, which means more tokens.
GPT 4.5 was $75 per million input tokens and $150 per million output tokens. So O1 Pro is more than 4x expensive as far as output tokens (since it's a thinking model, it generates many more output tokens than an equivalent GPT 4.5 prompt)

It's basically too expensive for most normal people to use.

The T3 AI guy did a whole rant about how O1 Pro didn't even usually give outputs as good as O3 Mini High

But maybe he's just saying that because he doesn't want to provide O1 Pro either

[ebon spoke]

Awee thanks

Cutie

No swasticAI pls

I can't stand that guy

[naive sage]

Theo, that's the T3 guy who talked about it: https://www.youtube.com/watch?v=Dxq223X0kxM

[ebon spoke]

I know

It's not a normal price

I actually use cgpt pro but it's reasonable

[left stream]

That costs double of gpt 4.5 lol

[ebon spoke]

Triple*

Maybe once per day

[left stream]

Damn I only saw the input tokens output tokens is insane

[valid charm]

I got a $44 bill on three questions I made to o1-pro via the API

(high effort)

[teal furnace]

Im having huge issue with deep research. I'm asking for example to provide a list of classes defined in Crawl4ai python library but it's failing miserably only listing 3 classes. Might be prompting issue?

[left stream]

Hot damn

[left stream]

You think the 2k plan might help make the api cheaper?

[ebon spoke]

Only 44$?

I made a 4 digit one

[valid charm]

How many questions?

[naive sage]

I thought the $2000 plan was a joke, is that real?

[tame current]

I think I know you somewhere where you in the You.com discord

Is Perplexity better?

[ebon spoke]

Around 20, but used a lot of input and output since then were heavy code tasks

[ebon spoke]

Yes.

[valid charm]

Where the solutions right??

[ebon spoke]

It's light years better

[left stream]

Yes, they’re upcoming with the $20000 plan too

[ebon spoke]

Absolutely

[ebon spoke]

Twenty what?

[left stream]

Also i compared the prices with the api and pro

Pro subscription for chatgpt is $200 a month

Even if you use o1 pro with medium reasoning that still would cost you more money than what chatgpt pro users pay

They have unlimited o1 pro and that’s way more cheaper than the api price at that point

Crazy

[left stream]

USD

[ebon spoke]

Yeah I'm a cgpt pro user but that's the MAX MAX MAX I would pay

[ebon spoke]

20k is

A joke

[left stream]

Ong same

[ebon spoke]

You can buy

[left stream]

I’m glad they haven’t placed limits on it so far

[ebon spoke]

A BMW CAR with 20k

[left stream]

But since the api released chatgpt pro might be limited

[ebon spoke]

Not per month AI crap

Got 1984d

[left stream]

Ikr it better give me the secrets of the universe at that point

[ebon spoke]

"Sorry I can't assist with that at the moment"

[naive sage]

(Do you mean the cost?... because something being cost-prohibitive is a limit... OpenAI can't fulfill their mission if the cost is ludicrous)

[ebon spoke]

I really think that some oai employees are lunatics

Come up with tremendous prices

Than DeepSeek R2 comes out

[left stream]

On the amount of queries you can make

[ebon spoke]

Bashes, cooks, grills, eats your model

[left stream]

Its unlimited o1 pro

Gotta have unlimited for important stuff

[ebon spoke]

And realize nothing really required 20k

[ebon spoke]

Fav

[ebon spoke]

What mission

[cobalt vale]

that will most likely be an agent that can replace humans and runs 24/7?

[ebon spoke]

There is Gemini for api

Claude for code

DeepSeek for open models

Grok for not sees

[silent mesa]

What are the main benefits of perplexity pro compared to the standard version?

[ebon spoke]

Cooler

Has images

[ruby cove]

Finally I’m pro again