💬│general

[crisp aurora]

Omg the setup is so goodddddddddd

[young hollow]

Btw Brace Yourself Guys Qwen 3 Max is Coming

[violet minnow]

anyone?

[hollow vessel]

LETS GOOOO

[distant aurora]

https://tenor.com/view/we-are-not-the-same-touche-round-round-broken-record-here-we-go-again-gif-1704936680092501125

[gloomy cipher]

Gwen 3

[young hollow]

https://tenor.com/view/e-boy-gif-112987497394211353

[young hollow]

Exactly "The Biggest Brainiest Guy in Qwen 3 Family" Only Indicates to Qwen 3 Max The Last Model of Qwen 3 Family in My Intelligence and Prediction

I Believe it Will be Very Promising

[hollow vessel]

How many parameters are you guessing it will have?

[young hollow]

https://tenor.com/view/dbz-dbs-goku-ssb-kaioken-gif-6995431762871256822

[young hollow]

Probably Already 500B-700B

Possibly Upto 1 Trillion

[hollow vessel]

Im guessing like 800b-900b

[young hollow]

Yes is Possible that's Why i also Added upto 1 Trillion

Man I Swear if it's 1 Trillion it Will be A Nuke

[hollow vessel]

In smartness and a physical nuke if you try to download it on your pc

[young hollow]

Just 235B is So Decent

[distant aurora]

Gpt 4.5 was beyond 2 trillion, lol.

[young hollow]

Oh Yeah it Won't be Normal Consumer PCs

[young hollow]

Those were Old Times

Parameters Work Differently Now

Is More About The Quality of Training Data

[hollow vessel]

Yeah

[young hollow]

That's Why i think it Won't be 1 Trillion

[hollow vessel]

But im guessing the will most likely have it on qwen chat. Liem they do most models

[hollow vessel]

Yeah fair enough

[young hollow]

Oh Yeah They Don't Restrict Their Flagship Models for General Consumers

That's Their Strategy

Free for Consumers for Better Training Data and Big Community

[hollow vessel]

I love qwen for making their models without message limits

[young hollow]

Businesses-Enterprises for Monetization

[warm sparrow]

idk if y'all knew this but PayPal is giving 1 year perplexity pro for free

https://www.perplexity.ai/join/p/paypal-subscription

[young hollow]

Yeah Man That's Why I Love these Chinese Companies

They are Democratizing Ai Access

= Faster Better Innovations

They Should Have Web Portals for Donations

So Anyone Can Donate if They Feel They Got Something Really Worth it

[hollow vessel]

Yk whats sucks for me tho about qwen. Webdev dosent really work for me. I think thats for me being in the us. Bc it just loads forever

[vague hedge]

well well well

[young hollow]

China is Certainly Going All Out in This Ai Race Despite Restricted CPUs/GPUs

[young hollow]

Strange Hopefully they will Fix it or Try Trouble Shooting it

[hollow vessel]

Yeah

[young hollow]

Mark My Words

Around 2026 These Chinese Companies Will Release Great Ai Models

Deepseek, Qwen, Kimi, Z glm

Probably More tbh

Even Chinese Food Dilivery Company has Launched A Ai Model Longcat

[hollow vessel]

Gpt oss was kinda mid compared to qwen, am I wrong?

[tame current]

hey guys

[young hollow]

Grok 5 Will Arrive Before 2026

And it Might Force A New Better GPT Model

[young hollow]

Idk tbh i haven't tried it

Well at least is Open Source

[hollow vessel]

Yeah

Why do most if not all gemma models suck

[young hollow]

Oss 120 stands 6th for Now Despite Being A Model to Run Locally

Is Open Source

So That's Kinda Impressive

And Chinese Companies Certainly Forced this for Closedai

[young hollow]

Gemma is Just Not In The Race for Now

Future Gemma Models Should be Better I Guess

[hollow vessel]

I wonder when the new qwen model is supposed to come out

Would be very cool if it was today

[young hollow]

1 Week or Maximum A Month I Guess

Is Not Far Since is Already Announced

[hollow vessel]

True

[silent ocean]

What're the limits for comet when using pro?

[thorny sedge]

Hi

Everyone has Perplexity pro

:)

I think i need to purchase max now

[warm sparrow]

true

why are they giving it out for free

like literally everywhere

telecom carries

[thorny sedge]

Huh? Free!

[warm sparrow]

now paypal and venmo

[warm sparrow]

ye

[gloomy cipher]

\

@young hollow

[young hollow]

I Already Know About it Haven't Tried it Tho

The Context is Double Now so That's Great

[unique kite]

How can i generate good videos using ai

[ancient inlet]

use veo3

this is one of the coolest things perplexity can do. sift through an entire github repository and identify issues and suggest fixes

well, if you prompt it to.

[green crystal]

lol sonar is at the bottom

didnt even see that

[ancient inlet]

I then pass the markdown to grok code fast on roocode

[gloomy sun]

ok 😂

ok..

[merry kettle]

mhh I'm not sure if I like Labs, it's kinda fun but also half baked? The app is kinda there but also not and in the middle pplx forgets its an app and just gives you a normal answer.

[vestal girder]

What is comet

Without insulting me

[green crystal]

at this point gpt-5 on chatgpt gives me way more "labs" prompts vs perplexity

[merry kettle]

Chrome with some PPLX agent capabilities

[vestal girder]

Is there a chance it’ll be available for pro users

[silent ocean]

Had anyone tried making full slides using comet agent (for example in google slides)

[silent ocean]

Built in ad block and different design

[brazen jackal]

bruhh i was 2 steps away from completed smth..

[green crystal]

@dry kestrel hey i wanna try something. can you ask gpt-5 pro to make the most complex trolley problem question it possibly can make?

like the solution shouldnt just sit in the open lol

[keen holly]

is there a chat only for pro users?

[brazen jackal]

#👑│pro-lounge

[keen holly]

yeah I can see that lol

thanks

[indigo idol]

happy friday fam

[keen holly]

do I need to verify somewhere or am I good now?

[lean hemlock]

no! it is place where people learn perplexity

[brazen jackal]

nah ur verified already

[lean hemlock]

i jus joined from my different account

[stuck pawn]

anyone know what the latest comet update included?

[lean hemlock]

no i dont know i opening after a while

[proven bluff]

Hello

Where can i get a comet invite, please?

[vague crow]

<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>Triple Pendulum Animation</title>
<style>
body { background: #232323; margin: 0; }
canvas { display: block; margin: 0 auto; background: #232323; }
</style>
</head>
<body>
<canvas id="canvas"></canvas>
<script>
// Canvas setup
const canvas = document.getElementBy

[wise nova]

Android app didn't show answer, but web show it well

[modern sinew]

Glitchy face the same

[distant aurora]

Lmfao

Ai diogenes

[green crystal]

lol

[toxic spoke]

Android app on comet is insane🔥 🔥 🔥

[green crystal]

i want to see how it handles other stuff besides math and scientific research etc

[toxic spoke]

Btw how many scheduled task we can have on pro/max?

[timber niche]

Do we have a date for comet on android

I saw the pre registration

[short narwhal]

Soon™

[timber niche]

Hm

[distant aurora]

Nah, she dipped on me for a better guy.

@young hollow I'm pretty disappointed with musk. I expected a more free platform.

[green crystal]

wots this about

[wise nova]

And same was with 4.1

[gloomy cipher]

https://tenor.com/view/dog-accepting-fate-tiktok-at-peace-gif-567488341419321713

[tired brook]

Weird

Are the api down?

They seem to work

[zenith shale]

Go to #💗│sharing

[wise nova]

I don't know, maybe token limit and my huge prompt 😄

[cold path]

Any timeframe on Comet coming to Linux users.

[green crystal]

@distant aurora bruh chatgpt can now use screenshots of sources as citation

[solemn hound]

https://music.youtube.com/watch?v=aoBcqpl8iaE&si=RV0yoqCWGgeswEL9

[tired brook]

Gemini 2.5 pro is best!!! Natively ofc

I just wish spaces were there in gemini

[wise nova]

Best for what?

[static osprey]

hey guys new here, signed up for a free year trial for perplexity pro with my paypal, trying to take advantage of everything perplexity liking the platform, but question i mainly use chatgpt havent used much perplexity or heard much on it till i received that promotion, how does perplexity compare to gpt do most use both? if you had to pay for 1 which would you pay for and why? for chatgpt i have literally hundreds of custom GPTs and am very integrated with chatgpt paying the plus. I use gemini for the new image banana and that is about it, so wondering how is perplexity better or do most use both? appreciate any help guys and nice discord as well, i do like that as well.

[distant aurora]

You mean it takes screenshots of web pages that it browses?

[distant aurora]

https://x.com/Alex_Oloyede2/status/1963885920165183888

Spetsnaℤ 007 🇷🇺 (@Alex_Oloyede2)

When being a war propagandist pays more than being a wh*rе. Such is the life in today's Ukraine.

Jay in KIEV owned p*rn websites.
Pekka was caught mаsturbаtіng at a war memorial

Those who support Ukraine are sick people, grifters who want the conflict to never end.

X · Sep 5, 2025, 8:44 AM

[tired brook]

For long context handling

[heady forum]

Hey guys, why is it that every time I schedule a task in Eastern Time and then, for whatever reason, I go back to the tasks and click on edit, it shows 3 hours later? For example, I schedule an 8 AM task EST, but when I go to edit, it shows 11 AM. Why? Can someone please clarify thanks.

[tired brook]

I would say perplexity is better to search through the web, you can also change between models which works great, i mainly use spaces for specific tasks

[obsidian urchin]

i have mostly switched to perplexity rather than the google ai mode, the google ai mode is frustating and is quite unreliable

[charred laurel]

hoho

[static osprey]

yea seems like perplexity is like all in one models which is pretty cool + the ai browser im about to experiment with, but even if it has all models built in one, is it really same models from chatgpt etc? technically can use one all in one?

[green crystal]

yep and can use them as sources

doesnt always do it tho

[green crystal]

yeah not covering it also doesnt stop bullets

[green sigil]

https://youtube.com/shorts/kTCVgNPu0fY?si=BN0Lj7MhVVR0V0B0

Please

Support

[low sable]

Hey I'm new to perplexity pro and used all my video credits... Is it possible to replenish my credits?

[haughty bramble]

anyone having trouble accessing perplexity at the moment? on mobile and web

[rain quarry]

I have issues with the api right now. Getting 500

[plain cipher]

no its working fine

for me

[south nexus]

A

[mighty sonnet]

Yes, I'm getting that as well. Looks like there's some type of issue going on, but nothing on the status page, yet.

[rain quarry]

Looks like its fixed. Works for me back again

[upper ore]

Welcome onboard! I use Perplexity primarily for fact checking and quick research as that is its primary strength - facts with citation and references. It beats ChatGPT, Gemini and Claude hands down.

My work context is high-stakes in legal and compliance and business. Even top models like GPT 5 pro and Deep Think gets some micro facts wrong (its a lot better now compared to o3 Pro). So perplexity stands out in fact checks. And comet is really interesting, gives one a glimpse of what the future is like in web surfing and work

[gritty parrot]

Yo guys, someone wanna comet invitation?

[upper ore]

No even with multiple models across ChatGPT, Claude and Gemini, it doesnt function like native ChatGPT etc. Its tuned to be a specialised search and fact check tool. You will not get the same output like the native app though it may sound like it to a certain extent

Treat it as a specialized search assistant, thats how perplexity is built. I love the research function, its awesome and it really digs into what I need, far better than any other platforms out there

[lament imp]

I personally think that the Research Function is nice, but Research is definitely not as in-depth as Gemini.

[indigo epoch]

yes please

[upper ore]

I would say its a rapid fast lite deep reseach. Its very helpful if you need quick but highly accurate fact checks. I always use Perplexity to do a factual audit on my work

[icy pasture]

Well I'm turning 19 tomorrow

[lament imp]

Yes very true

[upper ore]

Deep research on Gemini and ChatGPT are more long form massive depth and coverage. All these platforms design their tools differently

[icy pasture]

It's September 6 in a few countries

[upper ore]

Even Deep Think and GPT 5 Pro are designed differently hence the output of a same prompt defers quite a bit.

ChatGPT provides great nuances and very objective execution plans while deep think is more contemplative and strategic and flow

[icy pasture]

Here, it's September 5 8:52PM

[lament imp]

Gemini, I feel, is the best all-around model at the moment.

[quaint ravine]

are there any benchmarks of GPT5 Pro?

Got a friend to run a prompt and it got it wrong unfortunately

[lament imp]

However, GPT-5 Thinking is by far the best thinking model, for the tasks I use at the very least.

[quaint ravine]

that's what I thought until GPT-5 Thinking came....just for my use cases tho

[lament imp]

But Claude I feel is the best for writing/rewriting.

[quaint ravine]

but I really like Gemini 2.5 Pro style in PPLX still

[upper ore]

If you are not into extremely complex work, GPT 5 thinking is probably the best you can get as an all rounder

[upper ore]

Its a different tool. Not an apples to apples comparison

[upper ore]

Agreed. Claude has the best writing

[quaint ravine]

it can be apples to apples depending on how u use it

[lament imp]

I particularly use GPT-5 Thinking for very complex tasks in coding. Then Gemini 2.5 Pro is the best for most general tasks, even more complicated ones. When I see it struggles, I go to GPT-5 Thinking.

However, this is not on Perplexity. This is when working directly with their APIs or in Cursor.

[quaint ravine]

I think people get seduced by Google's verbose deep research platform etc...but the actual answer is often wrong in my use cases, no matter if it outputs 2000 words etc lol

[indigo epoch]

can someone share a comet invitation please?

[quaint ravine]

yes I've compared Google platform, AI studio and PPLX's Gemini 2.5 Pro - I prefer PPLX's tbh

the accuracy overall is similar tho

[upper ore]

Gemini 2.5 pro is really quite an all rounder. I personally like the canvas tool best among all the platforms, at least thats for me

[quaint ravine]

just styles are different

yes very good all rounder

was my default for months

until GPT-5 Thinking came

[lament imp]

Yeah responses are a bit long. But, if there is a case where I am asking it to provide a substantial amount of diverse information, it seems to work well. However, if I have a single question that can be answered within a 1-2 page statement, it becomes VERY redundant.

[upper ore]

Use mini or just use the regular one or restrain it to word output and what you need

[lament imp]

Is there a mini version on the output of Deep Research on Gemini's web app

[quaint ravine]

i don't mind verbose

but actually getting some facts wrong

is bad

[upper ore]

Nope

[lament imp]

Gotcha

[quaint ravine]

it's like...despite thinking for like 10+ mins...still gets it wrong

[lament imp]

I just don't think you can beat Google's Pro plan in terms of value

[quaint ravine]

while GPT-5 Thinking or normal PPLX pro search with 2.5 pro/GPT-5 Thinking gets it right etc in 1-2 mins

[quaint ravine]

only if u actually use stuff

[lament imp]

True. If you are only using it to chat definitely not.

[quaint ravine]

the free version of NotebookLM, AI Studio pretty much sorts it without paying imo

[lament imp]

Does NotebookLM have the Podcast creator for free?

[upper ore]

Google markets its AI product by throwing in the sink to make up for what it lacks as compared to more higher performing tools in GPT and Claude

[lament imp]

Or is that Pro only

[tame current]

is there a way to unlink your perplexity account from your paypal account?

[quaint ravine]

yes but limited

podcast, video gen...all for fun right?

[upper ore]

Its real powerful model is deep think. But one is limited to like 5 or 10 prompts a day

[quaint ravine]

i dunno...simply not my use case

[quaint ravine]

yes i'm skeptical with such powerful models tho....Grok 4 heavy, GPT-5 Pro all failed my prompts

but just took much longer to fail them lol

[upper ore]

But gemini is where all the fun stuff is. Video, bananas, audio overview, story books, integration with google and all that weird stuff lol

[quaint ravine]

hopefully we see big improvements in AI soon, but I doubt it

[upper ore]

Your case and my use cases are probably different. So I can't comment

[lament imp]

This is so true. But everything they've been releasing so far is really impressive. I definitely feel they have the best easy-to-use Image + Video Generator. NotebookLM is great. Canvas mode on Gemini is a fantastic feature as well.

[quaint ravine]

i think the point is that i not sure the powerful models actually solve significantly more in real life use cases

[upper ore]

+1

[quaint ravine]

Opus 4.1 Thinking certainly does not

in fact, ime, Opus 4.1 Thinking performs worse than Claude 4

[upper ore]

It does for me. I cant get those insights with the regular ones

[quaint ravine]

yea i not sure about "insights"...I am looking at scientific/medical answers

often based on PDF documents I upload

[upper ore]

So clearly different use cases

[quaint ravine]

but sometimes with web search

indeed

[narrow pier]

Hello

[quaint ravine]

probably with different prompting,u could match the pro version etc

just need to tell it to search at least 1 billion sources etc 😛

[upper ore]

In my context, I restrict it to a cascading effect. Cos my work is more into legal and compliance, so its different. I cant afford wide base

[narrow pier]

How to effectively prompt for creating a doc ( word ) to perplexity as it does a lot of mistakes when it comes to handling with data even if simple one whereas gpt is good at it.

[wise nova]

How to enable text wrapping in mobile web?

[lament imp]

I don't think Language Models are going to get much better/stronger in general (only by small margins). We are hitting a bottleneck in terms of model sizes and resources. We need new architectures that can be adapted to LLMs that are smaller and more efficient, or a substantial boost in resources without diminishing natural resources (electricity). To me, that is the only way these LLMs will substantially improve.

But, the way these LLMs are being utilized is just getting started. This is where the vast majority of resources will be applied to in the future for sure.

[quaint ravine]

yes it is disappointing...I was hoping by now a lot more

[green crystal]

those models work with parallel thinking

there's some limits to that

some yt videos explain it better

[quaint ravine]

i think incremental improvements are still nice tho

like Gemini 2.5 Pro to GPT-5 Thinking

Gemini 3.0 Pro might be a long way away though, given the bottlenecks

[modern sinew]

No way

[quaint ravine]

people aren't going to be impressed by improvements in benchmarks by 1-5% over 6-12 months

[uneven brook]

hello

[quaint ravine]

maybe Google should just release Gemini 2.6 Pro

incremental improvement

but even then...I remember seeing from like initial 2.5 Pro release to June's version

people complaining that 2.5 Pro regressed

[uneven brook]

no

[quaint ravine]

it probably didn't regress (just likely improved in some areas and got worse in others) - same as chess engines

but point is, overall Elo gain (progress) is near zero

[uneven brook]

hum

[quaint ravine]

in computer chess engines, u can gain overall Elo (be better at chess overall in a sense), but get worse in some positions

just that, u got better in more positions overall

but doesn't mean u didn't get worse in some other positions

the same principle applies in LLM

[steep totem]

hey guys would anyone be able to tell me why i cant download comet? i verify my subscrbtion and but no download button appears

[uneven brook]

ones i have asked it to give answers and explain it for 40 questions

and the explanation was perfect for me

[green crystal]

improved efficiency and cost while having some improvement would be a better outcome

like gpt-5

[quaint ravine]

true

[lament imp]

There's a lot of people who overhype AI's improvements unfortunately :I

While there's no doubt it's improved a crazy amount over the last 3 years, it's just that everyone started to jump on the bandwagon cause of OpenAI

[quaint ravine]

was watching Dr. Roman Yampolskiy on that Diary of CEO

[uneven brook]

check if it's available in your country

[lament imp]

He's an AI researcher right

[quaint ravine]

I starting to genuinely believe that people who develop AI etc don't actually know how to use it

I'm sorta jk of course

[steep totem]

Im east coast in the US

[muted wharf]

[quaint ravine]

but they need to look at people who actually use it in reality....and see how many problems the AI still has

maybe they have better versions of the AI than us I guess 😛

[nimble arch]

hiiiiii!

i got mine's pro vers because of an educational license, tysm perplexity!

[lament imp]

All they do all day is think about the possibilities of what can be done with the technology. But I think the thing AI doomers are forgetting is that we currently don't have the capabilities to create this level of stuff.

We're undoubtedly getting to the point where it is a possibility, and yes, we should prepare for it by putting on preventative measures, but I don't think it is entirely in the near future.

At least not until Quantum Computing becomes more of a mainstream.

[shrewd thistle]

guys what do you all think about the comet browser tbh

[modern sinew]

annoying and not useful

I will rather stick to my dear chrome

[spark scarab]

👍

[narrow mural]

guys when is comet releasing on android

[lament imp]

Good for certain things, especially with quick page summarization. Otherwise, having it do things for you is often substantially slower.

[shrewd thistle]

eh nah man tbh one thing i love about it , its wayy cleaner than chorme, looks fking great especially on light theme. and then perpexility homepage is just great

but yeah your choice

you might like chrome for some other reason

or different liking of styles

[spring agate]

Guys can anyone suggest ai tools for mobile application development

[lament imp]

With or without Coding background?

[spring agate]

With I have some basic knowledge on coding

[tame current]

Hi

[lament imp]

You can make Mobile Apps using a lot of different coding languages, including Python, C#, C++.

Each has frameworks dedicated to mobile app dev.

[spring agate]

I'm planning on using python, so can you suggest any tools that will improve my efficiency

[lament imp]

Cursor is a great IDE to streamline programming.

But AI-Tools specifically tuned for Mobile App Development is unlikely.

Generalized LLMs/Coders are your best bet.

[spring agate]

Got it thankyou wolf

[modern sinew]

I just deleted 3 months of work

https://tenor.com/view/idfk-gif-3761179052566277881

[serene sand]

welcome on board!

deep dive ➤ #💬│general message

🪼

Currently, the Pro plan offers 5 videos per month, while the Max plan provides 15 videos per month. Both plans are set to increase the number of video creations in the future.

[wise nova]

How to activate text wrapping in code-snippet on eeb and mobile?

[thin lake]

i need an +86 chinese phone number

[manic wedge]

test

[waxen jay]

Hello

[gloomy cipher]

And here iam complaining about lab temperature being too high 🤣
https://discord.com/channels/1047197230748151888/1411709096691761242

[green crystal]

@distant aurora i got gpt-5 pro for the remainder of my plus sub to try it

its so damn thorough lol

i gotta try your prompt again

[grand notch]

Just joined pro coming from chatGPT Pro -anything I should expect?

[serene sand]

welcome on board!

deep dive ➤ #💬│general message

🪼

[indigo epoch]

comet invite someone?

[serene sand]

https://discord.com/channels/1047197230748151888/1392544076527833188

[sleek blade]

Yo

[ebon orchid]

Bonsoir

Je fais comment pour lier mon compte Google ?

Depuis mon tel IOS

[mental kayak]

Yo

[glossy sail]

i'm a pro user and still waiting for comet in the waiting list.. how long does it usually take?

[serene sand]

invites —>https://discord.com/channels/1047197230748151888/1392544076527833188

[left stream]

All wrong

Except for GPT-5 pro

Gonna test claude 4.1

oof

[glass shale]

[left stream]

Cool comet gave me study mode

[distant aurora]

Wdym by "for the remainder of plus sub" ? 💀

[distant aurora]

I thought it worked on the website too.

[languid star]

It does

[distant aurora]

Now try grok 4.

Or O3 pro if available.

What's the correct answer, 70000?

[south briar]

which model do you guys use?

[distant aurora]

O3/gpt 5 thinking.

Grok 4 after that.

[left stream]

It failed

[remote kestrel]

Hey guys, I didn't get the invite for perplexity's comet yet and im a pro user.

[sinful sonnet]

grok 4 and GPT5

GPT thinking is good but it's too slow

[distant aurora]

I got 27500 on both O3 pro and gpt 5 thinking on you.com

[left stream]

Yes, which is why they failed

[young hollow]

Yooo Qwen 3 Max Preview is Here

[left stream]

Same response

[pseudo ferry]

Depends, I normally use Claude but this morning it failed my tests! Switched btw GPT-5 thinking and Gemini

[left stream]

There is one who got it right though

Qwen 3

[radiant trout]

En español

[left stream]

Ys

[lavish needle]

https://cdn.discordapp.com/attachments/996874817313710123/1364610147267117056/togif.gif

[distant aurora]

Try qwen 3

[left stream]

What

[young hollow]

Did you try it?

[distant aurora]

Qwen is strong in math.

[south briar]

grok and gpt are both good

[pseudo ferry]

https://tenor.com/view/im-blind-cheech-marin-blind-man-gif-14165338

[south briar]

i find claude to be kind of not usable when using for web search

[young hollow]

Are The Benchmarks Out?

[pseudo ferry]

True!

[distant aurora]

Nah, this is personal experience.

[heavy portal]

It was resolved 1.5 hours before it started?

[young hollow]

Ohhh

[last bloom]

Thoughts on grok 4

[young hollow]

But for fr the parameters are not out yet?

[sinful sonnet]

AGI

[limpid orchid]

this cracked me up as well

[gloomy cipher]

Lil bro is out but why is he so expensive 😭

[distant aurora]

Sonnet moment

Stay within 128k I guess.

[young hollow]

Probably Because of Context Window and More Parameters

[gloomy cipher]

No

It's still expensive even for the lower context compared to prev models

[distant aurora]

Qwen on it's way to join the paid services.

[young hollow]

Then idk why it's expensive

[young hollow]

Obviously they have to generate profits

[gloomy cipher]

Billy bones already explained why

Btw what is the parameters?

[strange arch]

p

[distant aurora]

e

[strange arch]

r

[distant aurora]

v

[strange arch]

e

[distant aurora]

r

[strange arch]

t

[dry kestrel]

1T+

https://x.com/alibaba_qwen/status/1963991502440562976?s=46&t=_NtP_RUn04yF_4hD_VEDkQ

Qwen (@Alibaba_Qwen)

Big news: Introducing Qwen3-Max-Preview (Instruct) — our biggest model yet, with over 1 trillion parameters! 🚀

Now available via Qwen Chat & Alibaba Cloud API.

Benchmarks show it beats our previous best, Qwen3-235B-A22B-2507. Internal tests + early user feedback confirm:

X · Sep 5, 2025, 3:43 PM

[young hollow]

Over 1 Trillion

[left stream]

Ys

[distant aurora]

Non thinking models in benchmark comparison.

[left stream]

It got it right initially

[left stream]

But then without thinking it got it wrong

[left stream]

Then thinking got it wrong

Not holding it as conclusive since no clue what the difference between the one from the qwen website and the one from openrouter is

[distant aurora]

I doubt your gpt 5 pro

I don't trust it.

[gloomy cipher]

Can't wait for qwen3 max 3B quantized version

[left stream]

It’s the official answer lol

[dry kestrel]

Yes

[distant aurora]

https://tenor.com/view/waltergotme-gif-18867690

[left stream]

Well off to study peace

Here you go if you guys wanna test other models

Bon voyage

[glad blaze]

i love how the issue was resolved before it started

[distant aurora]

They be time travelling.

[glad blaze]

fr fr

[vague pagoda]

We're just that good

(Might be an issue with the bot, looking into it)

?

[glad blaze]

he is going to sue you for the everyone ping

its so joever

[distant aurora]

Aravind doesn't even talk here.

Lol.

@left stream yoooo

[gloomy cipher]

And active? Or are there no information on this

[distant aurora]

This was free version btw.

[south briar]

grok and gpt are better

[vague pagoda]

We don't have a changelog for Comet just yet, one is in the works

[red nova]

How long does it take to verify student status?

[young hollow]

I think The Official Release has Time Yet

Is Preview for Now

[distant aurora]

No way joe dipped.

[weary abyss]

yall

i have 3 invites

don need them

[distant aurora]

What you are looking for is a distributor.

Or just drop them here or in #☄️│comet-general

[weary abyss]

oke

[weary abyss]

nah some people take multiple invites

greedy af

[distant aurora]

I mean, it's not like you can activate them without an account.

[weary abyss]

i know some who did create multiple accounts 😭

[distant aurora]

No way.

Do people actually buy accounts just for comet access?

💀

[weary abyss]

idk

prolly cuz limits too yk?

[distant aurora]

Huh

[digital scarab]

hi

[weary abyss]

comet has limited usage

[weary abyss]

wlcm

[distant aurora]

I know, I just can't think of using it generously.

[weary abyss]

same

[distant aurora]

Maybe people have more usecases than I expected.

[weary abyss]

maybe i am just lazy

[distant aurora]

I'm too worried about it grabbing my bio from wherever I log in.

[weary abyss]

whats your bio

[dry kestrel]

Didn't say

[distant aurora]

I meant stuff like name and address on amazon.

[gloomy cipher]

How do you know when iam online

I just got on the phone and you replied 😭

[weary abyss]

oh nah i don really care abt that stuff

its not like im the president

[dry kestrel]

I see everything

https://tenor.com/view/sauron-lotr-lord-of-the-rings-mordor-the-eye-gif-5619682

[green crystal]

@dry kestrel uh did canvas get removed?

[gloomy cipher]

https://tenor.com/view/big-bird-angry-door-fall-gif-11404358

[weary abyss]

what?

[dry kestrel]

No I dont thinnkso

[green crystal]

hmm odd i dont see it

[weary abyss]

show us some

[green crystal]

unless pro doesnt have support it

lemme check

ah yeah 5 pro doesnt

[dry kestrel]

Yeah gotta switch

You have a Pro sub?

[green crystal]

i got pro for remainder of my plus

so im gonna spam the crap out of 5 pro lol

[dry kestrel]

Niceee

[green crystal]

i got 2 weeks

[dry kestrel]

I didn't know that was possible

[green crystal]

yeah you can

next month is $200

if it helps me earn the $200 and more im letting it renew

[dry kestrel]

Nice

[green crystal]

else im going back to plus

[dry kestrel]

Lol

[green crystal]

5 thinking on plus helped me earn this

so its all good value for now

[runic notch]

React with horse in annoucements

[dusky yacht]

Does preplexity have any models of it's own ?

[solid loom]

Yo

[elfin ferry]

Sonar

[dreamy kraken]

Unfortunately

[solid loom]

https://cdn.discordapp.com/attachments/1047649527299055688/1413380606540120104/Blurrr_1757011970028.mp4?ex=68bc6183&is=68bb1003&hm=1608089f4087a0dbf5501958bf66d67e2dbdf0bf39a94f6a7b01b164477a7090&

[dusky yacht]

I ve never seen it on benshmarks

[elfin ferry]

It's just a custom trained llama 4 or something, nothing special

[solid loom]

A bit later ☠️

[green crystal]

its like eh not far from the bottom lol

[elfin ferry]

It's fast but it's dumb

[dusky yacht]

So it's more or less just a chat app ?

[green crystal]

its just enough for fast searches

anything multistep is better done with reasoning models in pro

[dusky yacht]

Is there anything special about preplexity ?

Like what did i gain by paying compared to something like t3.chat

[green crystal]

oh well its very focused on web search

higher limits aswell? someone here who knows t3 needs to help lol

[green crystal]

5 pro took 13 min with this

[distant aurora]

Ikr, lol.

[green crystal]

thankfully i can just leave it running and do other stuff

[distant aurora]

Also

[green crystal]

would be nuts to sit and wait

[distant aurora]

I think this particular question of joe needs web search.

And for some reason

The gpt 5 thinking on perplexity fails in it.

Same with you.com

[green crystal]

with web on pplx works?

[distant aurora]

It only passed natively for me so far.

[distant aurora]

Didn't.

[distant aurora]

Grok 4 failed natively too.

So I'm assuming it requires some complex research/understanding of some law.

[green crystal]

hmm odd

grok uses agentic search

so should've worked

[distant aurora]

Yeah.

Omw to try it on lmarena.

[green crystal]

seeing how some models just cant work well without their native search tools
i wonder how o3 pro on pplx actually differs from just using 5 thinking lol

[distant aurora]

Definitely requires some sort of detailed search.

[green crystal]

oof

try POE

seems like they have added a web search toggle

web search toggle for gpt-5 api

that should be agentic i think

[distant aurora]

I need to make an American poe account.

They give only 300 free credits per day to Indian users.

3000 per day for west.

[distant aurora]

Dayum

Poe cooked.

Too bad I can't see the steps.

@green crystal it's definitely the search. Gemini normie vs deep research

[green crystal]

yeah it seems like the web search toggle is the web search tool gpt uses natively so its agentic

[green crystal]

lol damn

for me its 3000 per day

[sudden pier]

hi

[green crystal]

yep agentic search is the answer

[distant aurora]

Perplexity research got right answer on second attempt.

[green crystal]

bruh isnt chatgpt agent supposed to be able handle massive amounts of work?

[abstract mesa]

2000 years later....

[distant aurora]

Ikr. I thought that's the entire point of agentic ai.

[green crystal]

i got a chatlog worth 10 million tokens for it to work through

[distant aurora]

10 million?

[green crystal]

yeah....

[distant aurora]

💀

[green crystal]

whatsapp group lol

[distant aurora]

Bro definitely works for the feds.

Does agent have any context limit?

[green crystal]

havent heard of any

its just a desktop that gets separate instructions internally

so the file would be able to be stored and worked on in pieces

ohhh

its working now

[distant aurora]

Was probably going through the material.

[green crystal]

first attempt it stopped working when trying to retrieve it from drive

now it got the file

ah crap spoke to soon

its ded

gg open ai

very enterprise ready

[weary abyss]

what phone is this

[distant aurora]

Android

[weary abyss]

there are plenty of android phones

[distant aurora]

Realme

[weary abyss]

ah good phones but crappy system

why dont you install a custom rom

or cant unlock bootloader?

[gloomy hornet]

https://www.youtube.com/channel/UC_e0UIxnbKna2U2Y_bZAEAg

[distant aurora]

Can't.

System is decent imo.

[weary abyss]

is it? maybe it changed

[young hollow]

Guys Drop Your Opinions on Qwen 3 Max Preview

[distant aurora]

It's basically OnePlus OS now with a bit of bloat apps.

[weary abyss]

no how are you?

[distant aurora]

Oppo, OnePlus, and realme share same os now.

[weary abyss]

YEAH those bloat apps

annoy me so much

huawei had them too

[distant aurora]

I debloated it.

[weary abyss]

how

[young hollow]

Im You How About U?

[young hollow]

He Threw The Phone

[weary abyss]

did you get rich

[young hollow]

If you did then i did

[distant aurora]

Wireless adb > shizuku > hail app
Wired adb on computer > adb app control.exe

[young hollow]

🤔

[weary abyss]

im far from that

[young hollow]

Im the same

[distant aurora]

Send money when you get rich.

[weary abyss]

shizuku is nice but wireless adb is pain in the ass

[young hollow]

Alright Paw

You Too Tho

[distant aurora]

How? You just need wifi.

[weary abyss]

nah you hustlin too much

[young hollow]

Send Money if You Become Rich

[distant aurora]

What you gonna buy?

[young hollow]

Im Gonna Buy and Erase Your Memory

[warm summit]

Does anyone know where I can get the link to download comet? I do have the pro subscription but for some reason I can't download comet

[weary abyss]

does realme allow booloader unlocks?

[young hollow]

That's A Bold Assumption You Don't Know Anything About My Hustle LOL

[distant aurora]

Bro is a menace 💀

[weary abyss]

you need an invite

[young hollow]

IS JUST A JOKE

[weary abyss]

oof

didnt expect that treatment from you

[young hollow]

If i get Rich I Will Build A Best Gaming PC and Buy Best Phones

[warm summit]

ah ok. I thought I could get access as part of pro. it says so on the website. Can you share an invite?

[young hollow]

Im Just Trolling Buddy Sorry

[weary abyss]

xxxtentacion sad music simpsons background

[weary abyss]

oke

dm

[distant aurora]

You know that episode of rick and morty where they exchange intelligence/iq and then the other guy kills himself?

[weary abyss]

why rick and morty fans think they are geniuses

like everyone i met who watched that show

[distant aurora]

Good taste ~ awareness

[weary abyss]

why cant comet browser be similar to zen or arc

why is it so ugly

[distant aurora]

Shit taste ~ bliss

[young hollow]

No i don't know anything about it

[weary abyss]

how many seasons, 4?

[distant aurora]

I haven't watched.

8 I think

[weary abyss]

bruh

[distant aurora]

I'm going through family guy for now.

Rick and Morty is serious though.

[weary abyss]

you into movies?

i want recommendations

[distant aurora]

Depends. I would be if I had enough storage.

[distant aurora]

Have you watched punisher series?

[weary abyss]

nope

[weary abyss]

you a movie hoarder?

i just stream no down

[distant aurora]

Yeah. Kinda. Internet isn't very good here.

And I am broke.

[distant aurora]

https://www.youtube.com/watch?v=CWO5skS1MjQ&t=19

He probably got overloaded with all the anxiety that comes with overthinking and overanalyzing the memories.

[weary abyss]

ya did a speedtest? how much you got

[distant aurora]

It's a sham.

It's fast in the speed test but can't even load twitter comments sometimes.

Carrier be doing some funny shit.

[weary abyss]

hmm seems sus

you get like 30 in the test?

cuz its megabit not byte. divide by 8 thats your megabyte speed.

[distant aurora]

That's bytes already.

I don't use bits.

[weary abyss]

you can watch 4k with that

[distant aurora]

In theory, yes.

[weary abyss]

wdym

you changed that in the settings ?

[distant aurora]

In practical experience, it gets stuck on loading.

[distant aurora]

Yes.

[weary abyss]

bruh

80 mb

thats awesome

[distant aurora]

I know. It just doesn't work when I open YouTube or prawn tub.

[weary abyss]

check your private dns settings

[weary abyss]

prawn tub?

[distant aurora]

Using adguard dns rn.

[weary abyss]

tf is that

[weary abyss]

try without it

it lags with me too

use adguard modded app works fine without dns

[green crystal]

@distant aurora https://x.com/Alibaba_Qwen/status/1963991502440562976?t=4KIMg1Tb3I8xnsPfhLnQ7Q&s=19

Qwen (@Alibaba_Qwen)

Big news: Introducing Qwen3-Max-Preview (Instruct) — our biggest model yet, with over 1 trillion parameters! 🚀

Now available via Qwen Chat & Alibaba Cloud API.

Benchmarks show it beats our previous best, Qwen3-235B-A22B-2507. Internal tests + early user feedback confirm:

X · Sep 5, 2025, 3:43 PM

I guess either gpt5 pro and standard is either way above 1 Trillion or scaling alone really doesn't carry a model

[twin grail]

@dry kestrel 300+ sources for pplx dr

[minor shard]

Thats awesome

[twin grail]

seggggg

[young hollow]

Well That Got Real Bad

[left stream]

If only we could use GPT-5 pro in codex cli

At least we can use gpt 5 high

[green crystal]

Oh you can't? I keep seeing posts how some people use it for planning etc lol

[lapis hamlet]

ola

guys i use fake addres to avoid tax, the card is a cashless master card, that how i get pro with cash

[blazing comet]

https://x.com/swishfever/status/1964064463717281992

fishy business (@swishfever)

DH3 on the Artificial Analysis Image Arena appears to be Seedream 4.0 from ByteDance

SoTA image generation and editing

X · Sep 5, 2025, 8:33 PM

DH3 is Seedream 4.0

[dry kestrel]

Thats awesome

What was the prompt?

[hollow vessel]

Chat just got home

Qwen3-max is incredible

[distant aurora]

Gpt 4.5 was beyond 2T.

[dry kestrel]

Source?

[distant aurora]

I believe scaling alone doesn't carry the model.

[distant aurora]

There's just guesses. No official confirmation.

[dim salmon]

hey guys does does any one have turn it in account and does anyone know where i can buy temporary acees or use to check some work

[distant aurora]

You get 3 free pro searches per day I think.

[lapis hamlet]

hi guys what is the best ai model for teaching?

[green crystal]

Really? Guess that's it's reason for being so sluggish?

Hmm

[lapis hamlet]

can you help me/

[green crystal]

Well I can use it now and woah I missed the high emotional intelligence of it lol

[green crystal]

Eh?

[distant aurora]

@left stream grok 4 deep search nailed it for me.

[dry kestrel]

Yeah, it's a bit annoying that we don't know the actual parameters

[lapis hamlet]

like what is teh best ai model/

[green crystal]

That depends allot on what your goal is

Gpt-5 as generalist and then it branches out

[lapis hamlet]

searching?

some ai always provide wrong shit so i want a good model

@green crystal

[green crystal]

Gpt-5 thinking

It's not 100% accurate but very close

[lapis hamlet]

i heat grok 4 is good at search but i will listen to you

[green crystal]

Grok is good natively

Pplx erh doesn't seem to be the same

[distant aurora]

Chat gpt edges out grok by just a bit in the search quality imo.

[lapis hamlet]

so they didn't 100% copy and paste grok 4 on perplexity

[green crystal]

That's purely because grok decides to mislead itself in the first 10 seconds during reasoning

[distant aurora]

Fr bro

[green crystal]

Nope it doesn't have access to it's own tools

[distant aurora]

It's assumptions are too strong.

[green crystal]

It doesn't even assume that stuff you tell it might be past it's cutoff date and just says that you're dumb and it'll use it's own knowledge instead

[lapis hamlet]

ok whatever i ask deepseek say for teaching go claude think, for searching it says grok 4, but you say is chatgpt 5 think

[final plank]

👋

[lapis hamlet]

sonar is the worst ig

[green crystal]

Sonar is fine for simple fast searches

Like a Google search

Beyond that any other model will be better

[lapis hamlet]

you unlock the 20$ plan so you should not waste any pennies so always gof ro the best

[distant aurora]

Sonar > sonnet retarded.

By retarded, I meant non thinking.

[green crystal]

@distant aurora ok time to test how gpt-5 pro is at cooking

Told it to search a bunch of recipes and make a single recipe out of them

Hope it doesn't make me puke

[distant aurora]

Burrito pizza? Lol.

[lapis hamlet]

oh google says that for newest information go for grok, for history or b4 this week go for gpt 5 think

[green crystal]

Just a simple tomato chicken stew
But it had to lookup complex recipes and draw tasty parts of them and figure out which parts fit best for my requirements

[lapis hamlet]

how can you apply a job in perpelxity?

[dry kestrel]

Hard disagree

[distant aurora]

Alright wait.

[dry kestrel]

@onyx hornet Please don't self promo

[lapis hamlet]

he just delete a introduction

[left stream]

how about we let o3-pro and gpt-5 pro fight it out

[lapis hamlet]

why you logo is pro not max?

[dry kestrel]

You can still use o3pro on ChatGPT btw

[distant aurora]

Bro, it's more about the search quality than the model iq.

[lapis hamlet]

whatever i am not spend 200$ on a ai, i can saved to buy iphone 17

[distant aurora]

Gemini nailed it in deep research.

[green crystal]

I only see o3

[dry kestrel]

Why do you need to buy the newest iPhone?

[dry kestrel]

change the model url

[green crystal]

[distant aurora]

#💬│general message

[green crystal]

Ahah I see

[lapis hamlet]

i am 14 i have no phone i need to wait new one

[left stream]

No clue

[dry kestrel]

you can still use o4 mini high btw

[lapis hamlet]

teach me how to glitch max

[left stream]

I heard comet becomes next level when on maxx

Was curious

[dry kestrel]

Let's not try to abuse the service...

[green crystal]

Any behavior difference between 5 pro and o3 pro?

[dry kestrel]

try it out and lmk what you think

[left stream]

Time to find out

[dry kestrel]

5pro is a bit less consistent in terms of reasoning time in my testing

[green crystal]

See if the models can use vision

[lapis hamlet]

i will turn the screen to extremly black so teacher won't see, and post the anwser in humanize AI

[green crystal]

Like it thinks too long/ short for some tasks?

[dry kestrel]

please don't do this, || humanize AI isn't full proof||

[edgy stream]

[dry kestrel]

Yep

same prompt and sometimes it will think for 10 minutes, sometimes up to 20+

[lapis hamlet]

i will do multiple time untill it detect below 10%

[green crystal]

It did take 10 min to make a tomato Chicken stew recipe

[dry kestrel]

Doing this isn't helping you at all. Someday you are going to have to write a paper exam

[lapis hamlet]

well i know those public school sending some boring and pointless work for us, i will ask ai to give me fake sat question and i just keep doing it

[distant aurora]

Use o3. Thank me later.

O3 has high odds of passing those ai detection tests.

[distant aurora]

Or you could insert white text.

Invisible adulteration.

[lapis hamlet]

i remember ai is start at end of 2022

[distant aurora]

People often use this method to hack ai powered resume screening.

[left stream]

Wdym vision

[dry kestrel]

Very interesting, this appears to be true

[lapis hamlet]

this is a good idea i always copy and go to another site and paste it, your is mroe faster

[distant aurora]

I have banged my head for hours on originality ai.

The entire point of the humanizing prompt I was working on was to make it work on cheap models.

O3 price drop makes it accessible to all.

[dry kestrel]

I gotta look into the mechanics of these originality detectors but there has to be better ways of detecting LLM output

[lapis hamlet]

there is a important reason i always use perpelxity is differnt model and school wifi didn't banned

[distant aurora]

I ran a Google deep research on this.

[lapis hamlet]

pls don't set the website to like game so firewall will detect and cook

[distant aurora]

Lemme see if it's still there.

[dry kestrel]

I'll do some research on my own first. I have a rough idea but let me see if there's anything else I can find

[lapis hamlet]

so o3 is the choice to do classwork

[distant aurora]

For now.

No guarantee it'll stay that way.

[lapis hamlet]

ik

[dry kestrel]

This was an interesting read https://arxiv.org/pdf/2502.12150

I have a feeling that advanced plagarism services such as TurnItIn would still detect LLM output to a certain degree

[lapis hamlet]

first time that i am use ai they don't provide online searching they use information b4 2022 august

it was chatgpt

[quaint ravine]

PPLX + Claude 4 Sonnet Thinking (with web search enabled only) gets it right too
ChatGPT + GPT-5 Thinking also gets it right

interestingly on PPLX, if u enable all 4 sources (not just web search), will get it wrong lol

[dry kestrel]

I've never use Edge on my phone. It's still there for me

[distant aurora]

Also, make sure you check for and remove ai watermarking. It's a real thing.

[chilly saddle]

I purchased the Pro subscription via PayPal as part of the one-year free offer, using an email address to sign into Perplexity. This account is not associated with my Apple ID or any Apple ID.

Because of this, I’m unable to cancel the subscription through the App Store or Perplexity’s website, as it isn’t linked to an Apple account.

Emailing support hasn’t led to any help. What can I do? I’m on iOS

[dry kestrel]

In reality if you really wanted to detect AI outputs, even with these newer models it should still e possible.

[left stream]

Interesting

I thought study mode is included in max lol

[dry kestrel]

Only for EDU atm

[distant aurora]

You actually read this whole 32 page thing?

[quaint ravine]

meanwhile Opus 4.1 removed from Claude Pro sub

[left stream]

Cool, I’ll have my second profile for edu then

[chilly saddle]

How can I get someone to help with the cancellation? The assumption is my email is tied to an Apple account and Perplexity doesn’t seem to have an actual support team

[quaint ravine]

wait nm https://www.reddit.com/r/ClaudeAI/comments/1n9hbv6/opus_41_temporarily_disabled/

[dry kestrel]

It has nothing to do with your subscription status

https://status.anthropic.com/

[dry kestrel]

when emailing [e-mail] start it with "I am requesting for human support"

replies take 3-4 business days

[quaint ravine]

I know look at my post after

[dry kestrel]

l a g

[quaint ravine]

restat

[dry kestrel]

I know, I'm saying I saw your post after I sent my message

[quaint ravine]

I know 🙂

[chilly saddle]

I’ll try this — Sam (an AI) has been saying it would forward to billing and nothing happens

[quaint ravine]

tbh, amazing experience with Claude sub so far....not

very good experience tho

[vale bridge]

Hi all

[quaint ravine]

eye opening

[dry kestrel]

I read through the key parts

[vale bridge]

what is the model guys to regenerate the images

[distant aurora]

Gpt img 1

[dry kestrel]

They have a wide range of models

[distant aurora]

You can select it in preferences.

[quaint ravine]

image gen has been pretty solid lately

[chilly saddle]

How do I get pro status on discord? Is there special support here for Pro subscribers like the support docs state?

[dry kestrel]

[distant aurora]

Use the invite link on the website.

[dry kestrel]

Also you need to email [e-mail] for billing support

[lapis hamlet]

how do I remove

i never heard this

[chilly saddle]

I did— I get an AI called Sam that says they forward to billing and billing never got back with me

[chilly saddle]

I don’t see this anywhere

[dry kestrel]

#💬│general message

[dry kestrel]

Check the pinned message

[dry kestrel]

Upon further testing, o3 isn't completely fool proof

[oak meteor]

<>

[chilly saddle]

testing

[distant aurora]

I know. I just haven't come across anything better yet.

[dry kestrel]

cough cough rewrite it yourself

I don't see what good it does when you just copy the output anyway

[chilly saddle]

testing now

[distant aurora]

Wouldn't mind if same words were approved.

[distant aurora]

I feel like a lot of academic grind simply gets in the way of learning instead of encouraging it.

More of a time sink.

This ofcourse depends on the learning style of the individual.

[dry kestrel]

Yeah well you gotta learn these skill sets anyway

[distant aurora]

No

I'm a diy person.

[dry kestrel]

HR doesn't care about that

[distant aurora]

Homeworks are for people who can't diy.

[dry kestrel]

and for that particular user's case, universities don't care either

[distant aurora]

Hr probably won't ask me to write long authentic pages. Depends on the field of work.

I have worked below people who use ai like crutches.

I'm assuming by hr you meant human resources/recruiter.

[quaint ravine]

Opus 4.1 back

[quaint ravine]

on Claude platform, Sonnet 4 and Opus 4.1 alternate answers between 30,000 and 70,000

sometimes giving one or the other only, sometimes giving both dependent on tax

This might be the best answer perhaps (Opus 4.1), but it doesn't give this answer consistently as I described above:
If it's a discretionary trust (where you as trustee have discretion over distributions):
As a trustee of a discretionary trust, you're treated as still owning property for SDLT purposes GOV.UKGOV.UK, triggering the 5% additional property surcharge SDLT Additional Property Surcharge 2025 | Policy Guide UK:

£70,000 total SDLT

If it's a bare trust or life interest trust (where your son has absolute rights or right to occupy/income):
The beneficiary (your son) is treated as the owner, not you as trustee GOV.UKTax Insider, so no surcharge applies:

£30,000 total SDLT

The calculation breaks down as:

Without surcharge: £0 (0% on first £125k) + £2,500 (2% on next £125k) + £27,500 (5% on remaining £550k) = £30,000
With surcharge: Each band increases by 5%, totaling £70,000

Most likely scenario: If your son is under 18, it's probably a discretionary trust = £70,000 SDLT due.

[distant aurora]

Yeah, I got something similar on grok deepsearch.

I'll try again though.

[quaint ravine]

it's actually not following instructions in a way, cos ur prompt said to "answer succinctly" lol

okay another Opus 4.1 try gives 2 compeltely different answers lol

[left stream]

I’m trying to see if comet performs better with study mode or with extended reasoning

[quaint ravine]

I tried that with ChatGPT platform

doesn't make much difference

in terms of accuracy

[left stream]

Wdym

I’m talking about this

[quaint ravine]

yea just talking about the concept of study mode

[left stream]

Both can’t exist at the same time so gotta figure out which is better

[quaint ravine]

it's still at mercy of the actual AI model intelligence/RAG etc

[left stream]

Not when both can control the webpage and get the transcript lol

But ngl

[quaint ravine]

wat u mean

they still rely on the actual undelying model

[timber niche]

Why your iPhone camera ass

[left stream]

Sweat

[timber niche]

Lmao

[quaint ravine]

lol

[timber niche]

Wipe it off

Lol

Your phone camera looks like 2014 Samsung a3 camera haha

Also why do you have hp and McAfee pinned 😭

[distant aurora]

HP is fine

Mc afee was probably default settings.

Microsoft store and copilot bother me more.

[tough heron]

Hey all, has anyone tried "Comet"? Is it any good? Is it really invasive?

[solar widget]

Bro it played a full game of wordle out before my eyes and I didnt even give a propt, its just cool

[timber niche]

Lol

[left stream]

Interesting, juice number in codex cli is high asf

[dry kestrel]

You can manually switch the reasoning setting

[left stream]

Yes

[dry kestrel]

its great

I might actually unsub from Claude

for Codex

[left stream]

If only I could use GPT-5 high on chatgpt

[dry kestrel]

yeah that really sucks

Someone posted that they were working on a reasoning slider

which I found pretty cool

[left stream]

At least it can do cool shit and get internet access

I know what to do now

[dry kestrel]

Yep

[left stream]

With chatgpt pro it’s awesome ngl

Good thing mine’s bugged

[dry kestrel]

there are still rate limits though

afaik its weekly

[left stream]

Yes, but the number is gigantic

[dry kestrel]

Mhm

[mental ocean]

Just subscribed to perplexity pro any feedback advice ?

[dry kestrel]

Check out this guide on Perplexity #💬│general message

[dry kestrel]

Wth

Agent?

[left stream]

That’s codex via cloud

[dry kestrel]

oh

thats super weird

[bitter ore]

quick question

[dry kestrel]

Yeah, issue I have with this is the large range. requests and then they don't specify the weekly limit for $200 is insane and I'm not too sure how they rank queries

[left stream]

Inr

Ikr

[bitter ore]

perplexity told me that it uses collected data only for answering my questions, and to train ai only if the checkbox is enabled

i checked privacy policy and it seems like true, can anyone confirm

[left stream]

300-1,500 messages every 5 hours

[dry kestrel]

You can turn off data retention and training here https://www.perplexity.ai/account/preferences

[left stream]

It doesn’t mention the weekly limit

It does mention a weekly limit, but no specific number

[dry kestrel]

Oops I worded that kinda wrong

Brain isn't braining

[left stream]

It’s a monster

[heavy crystal]

sorry but my messages keep getting deleted

why is that

[left stream]

Bro

I just realized something

[left stream]

256 might be legit

I’ll take a look at usage rates because it said something interesting

@dry kestrel

It doesn’t say a limit for cloud tasks

It only says generous limits for a “limited time”

What if 256 was actually deadass legit?

I would use it more on cloud but it looks ugly on ios

Cool windows seems beautiful

[dry kestrel]

Gonna test a bit later, so far 200 is the highest I could get over the API

[left stream]

Yeah, I’m gonna test it as much possible because no clue when that limited time will end

[dry kestrel]

I've def seen people hit limits, not sure about the specifics though. I hope they manage to make money on their Pro accounts instead of what Anthropic did

[left stream]

It might explain why codex on cloud has ask and code

[lapis hamlet]

perplexity need to be more popular than chatgpt, cuz you can choose differnt models(btw are they 100% copy and paste model)

[left stream]

Ask uses fast mode

But code

That is a different monster

[dry kestrel]

True

[dry kestrel]

They use the same underlying models

[lapis hamlet]

🇼

gpt 5 thinking need to take a long time