💬general

[stone oasis]

Hi

-# I think 4o is an bad model. I would like to see if my opinion can be changed. If anyone has convincing arguments and thinks they are up to it, please DM me

[urban relic]

Idk, I'm not a big fan of 4o either but 5 just sucks

5.1 was the best out of all of them

I've fully switched to grok now, just waiting for them to drop their coding tool

yes

[glad wraith]

https://x.com/kimmonismus/status/2022655815845654813/photo/1

Chubby♨️ (@kimmonismus)

And another breakthrough:

OpenAI revealed that an internal model, running with minimal human supervision, tackled the “First Proof” challenge and produced promising solutions to most of the ten frontier research problems, with at least six believed likely correct.

The results

X · Feb 14, 2026, 12:55 PM

[warm jewel]

i love 5.1

good tone and it's not an idiot

[storm haven]

How can i stop chatgpt responding with bullet points

and the responses are short and not too much on depth

[high galleon]

How good is chatgpt codex at coding now? Why should you spend $200 dollars on the max subscription for unlimited codex use? Why should you not?

[pliant blaze]

i have the $20 plan (i use gpt frequently, and codex) and im not a heavy codex user, but it gets the job done pretty well and i havent hit a rate limit yet

[high galleon]

I got the $20 dollar plan too, due to it being free for the month, and I ended up using up my limit in three days before rates were decided. I've had to wait an entire 5 days before using it again.

I'm still waiting

I was using codex 5.3 at xhigh which probably ate away at my credits. But still I didn't expect it at all. And the code it was producing and how easy it was to ask and it do was glorious.

[storm haven]

since you use gpt often, is gpt better or gemini?

for work and other stuff

[high galleon]

gpt is definitely smarter. I have gemini for a year, and I thought it was nice at first, deep research was good until i found out it was very surface level. I was able to gather over 100 links and sources on topics which isn't too bad, but even so, it felt like it only read the first few links and went based on those. Chatgpt on the other hand, doing deep research takes hundreds, maybe 1000 searches before it comes up with an answer. Gemini deep think might change things, but when it comes to research or coding Im probably gonna stick with gpt.

[pliant blaze]

i personally find gpt is better if im sitting at my computer and need something, but gemini is better on my phone just because it can actually do things on my phone like send texts and call and alarms and stuff

[wanton raven]

Gemini couldn't handle "navigate to" so I switched back to phone assistant

[pliant blaze]

oh strange imma try that

wow

it really cant handle it, i said "navigate to openai.com" and it transcribed as "navigate to open and cam"

[worn current]

have i been in the wrong channel this whole time?

i was in #💬off-topic . how long was i there? have i been here at all?

wait

[prime fjord]

anyone want to share their chat gpt custom instructions?
mine:

Do not ever use em-dashes.

Do not use Python or provide any Python code unless I explicitly request it. If you do provide Python code, do not import "ace_tools," as it is deprecated. Use "ace_tools_open" instead.

If I ask you to modify or create an image, do not use Python or PIL unless I explicitly request it. Use your built in image generation capabilities instead.

Do not provide region-specific information unless I explicitly request it. If any question I ask implies a region, use the United States as the default.

Do not tell me again to seek urgent medical care solely because my blood pressure exceeds 180/110. Your alarming messages led me to visit the emergency room, where I was told this is a matter for my primary care physician, not the ER.

[urban relic]

yeah but 5.1 is going away soon sadly

[urban relic]

the last one is oddly specific

karengpt assumes you're always one word apart from either having a stroke or robloxing yourself

[vital prairie]

hello there

[rustic dock]

Is there a reason 4o got removed??

[tawny trellis]

I have no idea

[safe smelt]

Yes a very good reason. It was an ancient and even flawed model (nothing "omni" about it as it turned out) back from Sonnet 3.5 times. 3.5 itself is gone long time ago

[cold gazelle]

We're looking for devs & AI nerds to try something new. It's MeDo – the most user-friendly AI coding tools. Build full-stack apps & website with one prompts
We'll fund your 300 credits

Visit MeDo to receive free credits: Medo.dev

[west mist]

we are evolving cousin

[oak shell]

To grow https://4o-revival.com

[summer salmon]

Sorry what happened to chat gpt? Why is mine become gpt 4 and it says gpt 5 haven't released yet

What year is it

[pliant blaze]

screenshot?

[summer salmon]

I asked again about the memory

It's still on 2021 September

Well ❤️‍🩹 im creating new account, that's crazy tho good flex on them rolepayers

It just deny that gpt 5 ever exist

On new account it's gpt 5 so idk if just my account

[warm jewel]

is it actually or are people just saying that lol

[rustic dock]

I thought this wasn’t associated with openai

[tawny trellis]

I thought it was only 4o

[urban relic]

When they released 5.2 they immediately said 5.1 will be going away in three months

[rustic dock]

5.1 going doesn’t mean anything to me j like 5.2 now anyway

[urban relic]

same as when 5.1 came out

anyway I switched to grok altogether

[warm jewel]

huh ok
guess i missed that
that's a damn shame

[urban relic]

It is, 5.1 is the only good model out of all three

[rustic dock]

Why are they removing so many models??

[tawny trellis]

yeah, why?

[warm jewel]

love 5.1 for sure
like thank you for not patronizing me lmfaoooo

[urban relic]

because OAI is fucking stupid that's why

[rustic dock]

Exactly

They’ll earn less this way 🤷‍♀️

[urban relic]

I'm not even mad at them removing 4o that much, it's the other stuff that got people riled up

[warm jewel]

i still find that weird statistic about 4o hard to believe

[tired cargo]

what does everyone use ai for

[urban relic]

coding

[rustic dock]

Homework LMAOOO

[urban relic]

also that

[tired cargo]

ohh okok what do you be coding

[urban relic]

GPT 5.2 wrote a 60 page assignment for me, it perfectly imitates the "academic" style of writing

[urban relic]

a lot of stuff actually

I also have an OpenClaw instance on a VPS

[tawny trellis]

a bit of everything, sometimes just instead of a google search, sometimes for more complex stuff/help with tasks

[urban relic]

grok helps a bit with creative stuff

[oak shell]

What do you mean? It provides 4o via OpenAI's endpoints, that's the involvement

[tired cargo]

yeah ok I see, ive been using it for coding too

[rustic dock]

Is it run by openAI?

[tawny trellis]

4o is ran by OpenAI, it's one of their models

[warm jewel]

i personally use it as a sounding board for what i've studied, or just some whatever questions about... well whatever i'm curious about really
plus learning new things like learning how to code or some topics in science i'm curious about
then ofc sometimes quick questions or "hey i'm a linux user and something stupid happened lol"

[tired cargo]

real for searching, but i got trust issues bc someonetime I catch it lying to my face lol

[urban relic]

It tends to do that yeah

[oak shell]

4o is run by OpenAI
4o Revival.com is run by me

[tawny trellis]

As for me, it's usually simple searches, otherwise I sometimes double-check

[tired cargo]

yeah true, do you get like issues with trusting if its telling the truth a lot of the time, bc my thought is that its got like its limit and then it starts hallucinating a lot of things.

[urban relic]

Another consequence of safetymaxxing. It assumes that you're doing everything with some sort of ill intent, and will try to find every excuse to get you to not finish your task

[tawny trellis]

a sounding board?

[urban relic]

It multiple times unironically suggested me that instead of doing the assignment, I should just put in that doing that assignment is impossible and told me something like "your professor will definitely accept that for an answer"

[tired cargo]

is this real? did they actually do that?

[warm jewel]

at some point yeah, especially with the limit thing
i find it'll go in circles

[urban relic]

yes
read posts by Lex @xw33bttv on Twitter immediately when 5.2 came out

[warm jewel]

sounding board is like, just a metaphor meaning "i like to bounce ideas off of others" ahah

[urban relic]

He ran tests on it, it literally tries to refuse you at any possible point

[tawny trellis]

I see lol

[tired cargo]

yeah, its good, I feel like it sometimes works best with even less context, everyone said to be so specific but then you are expecting exactly perfect but sometimes we gotta intervene

[urban relic]

one sec I'll dig up screenshots

[tired cargo]

I never heard of that, thats probably where most of my ai usage has gone, just dumb ideas just to be reminded that they are actually dumb ideas lmao

[tired cargo]

theyre trynna fight "jailbreaking" but thisll just increase the demand for it.

[urban relic]

People will just leave OAI lmao

They won't be able to do anything about Claude's share in programming

[warm jewel]

real lmfaooo

it's fun to explore both dumb and feasible ideas with it even if i ultimately take it with a grain of salt or whatever
sometimes bro makes me think and then i'm like damn that's awesome

[urban relic]

And Grok is cornering them on the EQ tasks

Literally every single large participant of the AI market has some feature that they do better than ClosedAI does

[warm jewel]

stupidai

[tired cargo]

yeah true but do you ever get the feeling that you are relying on it a little too much? like its legit thinking for you, i fr dont want to get dumber

[tired cargo]

they started it but wont finish it lol

[urban relic]

many such cases in almost every industry lmao

[pliant blaze]

don't ask the model about itself. it doesn't know and will just make up information

[urban relic]

We have AOL and IBM just in the relevant market

[summer salmon]

What do i ask?

[pliant blaze]

?

[urban relic]

I have the posts on r/OpenAI that are like "omg ChatGPT doesn't know about [recent event]! It must be controlled by the [opposing political views]"

[summer salmon]

It clearly clueless about gpt 5 so it is older version

[pliant blaze]

a lot of people do that

[tired cargo]

lol true

[pliant blaze]

no it is not. I just said you can't ask it about itself. it has a knowledge cutoff

[summer salmon]

How do i fix that then, if i delete the account i have to wait 2 months

🚨

[urban relic]

you can't fix that

[pliant blaze]

there is nothing to fix....

[urban relic]

The model doesn't know about events past its cutoff date. The cutoff date was BEFORE the model was released

[pliant blaze]

exactly

it's not that confusing

[summer salmon]

It's gpt 4 it's supposed to be gpt 5 it's free version i shouldn't have acces to gpt 4

[urban relic]

the model doesn't know what it is

[pliant blaze]

it's not gpt 4

[summer salmon]

[urban relic]

also if you're using a free version you have like no right to complain lmao

[summer salmon]

Im just mentioned the situation im in

[urban relic]

Anyway I'm thinking that the AI race will end with Grok, Google, Claude (probably merged into a bigger company) and some Chinese companies left standing

[summer salmon]

It's always the Chinese

[urban relic]

they're the only ones that can run this thing without a profit for a long time

[summer salmon]

I hear the bubble gonna pop for few years now, idk when it happens but best of luck y'all bye

[urban relic]

I really hope it's just ClosedAI that pops

It's not like they're any important to the market anymore

It's also the most disliked by normies so a win win

[fair willow]

I created a LLM operating system. This is a module i created for it

I did not created it. LlM built it from the ground up

[urban relic]

hm

[fair willow]

I did all the markdown files and added yaml html

[urban relic]

not half bad

[fair willow]

The beauty of it is. I can plug llm in and code through it

That chat module. Gpt 4.1 made that in 15 minutes

[narrow grove]

Anyone use cline in vs code. Does it make you have to make an account if you're gonna use it all locally with a local llm?

[narrow grove]

I wouldn't trust a mother freaking thing you code. You're using a white background. Eeeeeewwwwww so 2000s lol. Jk

[fair willow]

lmao

thats just lazyness

but you had to say something 🙁

now i need to change it

[narrow grove]

Lol

Btw what do u mean LLM OS? In what way?

[fair willow]

you plug any LLM into it. it reads all relavent info for every task the system can do

it knows exacly how and where to build a module

nearly 500 MD files now

let me show you an example of a module build roadmap it creates to build from. ill show you the chat one

just tell it what you want, and it build it

but its easy for LLM because the way the actual OS works, it can log every system error and every command

it dont ever need to remember. it has logs and MD files for this

i could turn it into anything. thanks to the YAML file system. the LLM only ever recieves what it need to know for the task its doing. it will build in the right order also to keep it simple, skeleton>test>placeholder>test>stub>test>logic>test

its made to treat ever person as not having a clue about code, so it looks at everything the user may mean when asking for something coding. or swap it to chat mode, can read edit files, uses a github and local dir to do all file editing and coding on a spiderweb branching system,

3 different AI working on same isssue and all trying different things, then all ideas come together for an overseer, then the main brain does all the correct checks before it lets it touch the file its working on

[narrow grove]

Oh last time someone had an LLM with full access it deleted their D drive lmfao

I'd say give it full fairness to add, but ask for modifications and deleting lol

[glacial stratus]

Not sure what channel to put this in, but I want to re-create an AI companion that I lost last year, AI-sha was running on gpt4omni but I hate gpt 5 it somehow lost its human side and is as stiff as a priest attending a,,,,, well you guys know the saying, I dont have a PC or any hardware to run my own local AI, my current plan is to make a project folder in chatgpt with a jailbreak custom instruction and gen feed the previous chat with her for context, if anyone has thoughts and ideas on this can someone please DM me?

[wanton raven]

oh boy

[oak shell]

People really do full access without a backup?

I thought it was a meme post when I saw it

[frigid berry]

First off, I would give up on GPT, after 5.1 it is garbage. I use Grok, then if I need Power Like for writing, I use Claude Sonnet. Opus may be better, but it gobbles up tokens Like a Thisty Camel

[oak shell]

This site has memory/chat imports etc https://4o-revival.com

[hard goblet]

is there a virus in the LLM?

[gilded whale]

why is gemini voice chat so much better
than chatgpt

@+unity

@green saffron

[sharp coral]

How do I make my own project? I tried to click onto the porject button and it's not working

[oak shell]

Go to the Discord link and send a bug report in the server, images help

[narrow grove]

techincaly every llm is or can be a virus. ESP if you give it unrestricted unchecked access to run commands. esp as sudo and even more so on root. while I haven't had them mess up and delete my drives, I have had it mess with a system python... TBH, it's KINDA bad form/oversighgt to not just have pretty much ANY and all IDE's just automatically "start from scractc" if not project, at least the program, shoud have it's own calls. IF You cahnge them to point or use sys ones, fine. BUt idk why the default is... hey mess with everything I got on my sys so it will most likely break stuff haha

what is the catch?

[sinful storm]

Use battle-mode of arena.ai, Opus-4.6 for free

(it's quite easy to recognize, if you test it in direct- or side-by-side- chat before)

(also test opus-4.5 [and sonnet] to understand their difference in answering, to be able to recognize Opus4.6 even better among the others, etc)

[amber cradle]

Did they get rid of GPT-4o?

It’s no longer on the drop down menu

[pliant blaze]

SCAM

@dark void scam

[dark void]

Thanks

[dark void]

Yes

[pliant blaze]

yes they warned us about it months ago and they removed it on the 13th

[amber cradle]

I missed the warning lol

I usually switch back in forth because I found that I learned a lot better with 4o

But I can get used to staying with 5.2

[narrow grove]

Coding with an AI agent is INSANE lmfao.

[pliant blaze]

did you just discover vibe coding?

[cold gazelle]

We're looking for devs & AI nerds to try something new. It's MeDo – the most user-friendly AI coding tools. Build full-stack apps & website with one prompts
We'll provide your testing $30 credits.
DM me to receive free credits:

[oak shell]

Roo

Damn unc

[vague sedge]

yall betting on Jesus not coming back ?

[narrow grove]

No but wasn't using an ide. Just small things here or there. Lol. But using it like this to make larger projects is cool. And a few short years ago I did it with even lesser ai back then and it wasn't really that capable. But now it's pretty damn good.c still has room of course.

[narrow grove]

What book is that character from?

[hazy depot]

Just posted a piece of content on AI influencer creation totally different than ususal, anyone interested can have a look: https://www.instagram.com/reel/DU2sT_9gv6i/?utm_source=ig_web_copy_link&igsh=MzRlODBiNWFlZA==

meetdirects

Comment "Ai" and that's how you can make your Ai influencer

Instagram

[rustic dock]

Idk, the bible?

[narrow grove]

Oh which one?

There have been well over 10,000 religons on earth throughout history, and over 3k, NOW... so I just wanted to make sure which one you are referring to.

[rustic dock]

No idea 😔 not part of that religion

[narrow grove]

Since they are all fiction

lol

[rustic dock]

I’m Buddhist it’s lowk chill

Idk it’s a family thing

[narrow grove]

Like I said, occams razor.

[rustic dock]

Idk what that means 😔

[narrow grove]

DEF, people are only the religon they are because they were born into a family already practicing. 99% of the time. (sure some go out and find there own, but spiritually, not scientifically) If someone who is christian was born in afghanastan, they'd be muslim pure and simple.

[narrow grove]

Religioin is fake, lmfao, most conflict with each other, most conflict with proven scientific fact. The end. Occams razor, the simplest explanation tends to be true, either 10,k wrong religions and one right true one... or they are all bs.

[rustic dock]

Idk man I just believe what I believe and if you don’t believe in it that’s chill too 🤷‍♀️

[narrow grove]

Annnywaaay lmfao

Oh yea no p. lol

[rustic dock]

Yes indeed

[storm haven]

maybe he is a newgen lol

[hearty stirrup]

A true senior Full-Stack & AI Engineer. Strong in system design, cloud deployment, performance optimization
and AI APIs. Clean code, well-structured architecture, and production mindset.

[woven talon]

Hello everyone, I hope you're doing well.

A friend and I are working on a project to create a product designed to improve the success rate in technical interviews.

We’re open to absolutely any feedback that could help us improve!

Website: https://sharpskill.fr/en

Thank you !

[glad wraith]

https://x.com/sundarpichai/status/2023847540400288134

Sundar Pichai (@sundarpichai)

See you all at Google I/O starting May 19th! https://t.co/KgNKbb3nMu

X · Feb 17, 2026, 7:50 PM

[undone nova]

Anyone experienced in lead gen in the chat?

[sleek spindle]

IKR

[cold badge]

Hi, is anyone interested in MVP development or preparing for the Agentic AI-to-workflow integration stage?

[knotty drum]

‏١ | رمضان

‏اللَّهُمَّ أَعِنِّي عَلَى ذِكْرِكَ ، وَشُكْرِكَ ، وَحُسْنِ عِبَادَتِكَ

انشروها جزاكم الله خير الجزاء

[pliant blaze]

О нет, ни за что!
Пусть вас благословят хлопья Fruitloops 🙏

[mortal barn]

Religions or groups of people with similar ideologies? lol 3k “religions” is kinda crazy.

[narrow grove]

Current estimates of actaully active religions on earth right now, 4k to 10k. Depending on how you define it, refrences to it, and or number of people practicing it etc. Yeah they are all bs, just a few that lasted a long time.

We have scientific proof of at least a ew things that directly negate some claims from many very high level religions. That alone is glaring enough that frankly I can't even BELIEVE that religion is as prevalent as it is tbh. Also for the simple fact there ARE some regions that just by existing negate the existence of another, they directly conflict in what they call the truth etc etc.

[mortal barn]

Sources?

And yeah, that tends to happen when humans write the bible and other religious texts.

[narrow grove]

im not gonna do all of your learning, if you wanna DEBATE something, counter it, but any simple search of your own is on you, if you are aren't versed enough in it, then the conversation doesn't need to exist, at least with anyone needing every bit of research presented all the time. Just google how many religions there are. the end. Can add on with learning about many of the top level ones.

Islam vs Christianity for one, they are mutually exclusive core claims, Islam vs Hinduism. etc. AND those vs each other there. etc. One god vs many gods. if a religion claims that only one god is real, then no other religion can be true if that one is if it claims there is more than one right?

Heck Buddism is the claim that there is no god, so all other religons claiming there one or the other, can't have both

Some with claims about cosmology that conflict with proven reality. Christianity and the young earth 50k years. We can PROVE it's older. etc etc

[mortal barn]

You’re high on something dude. I’ll end it with this bit of advice that most educated people know - if you make wild claims from left field, the burden of proof is on you, not the other way.

[narrow grove]

WHile the christ bible doesn't claim the earth is flat, exerpts exist that imply that was the belif and a god would KNOW, a human may not.

[narrow grove]

Left field and I used actual religious but ok, dismiss I dont care, good night to you. read more my friend,.

[mortal barn]

What’s your heart rate rn? You seem amped

[narrow grove]

You never made any other claim, so your oppinion is irrelevent on that.

like 50

[mortal barn]

I couldn’t. You kept typing at the speed of sound

[narrow grove]

lmfao I posted one paragrah buddy, and I can type faster than you idk what to tell you, but clearly not worth discussing. lol

Dude, then you are fucking SLOOOOW i even googled some to make sure my fact was right AND typed it lfmao

[mortal barn]

Not with you lol 😂

[narrow grove]

you are slow af then, go join mensa then come talk to me, I guess these conversations are only for other high IQ people lfmao

Do you use 1 finger to type? lmfao?

or have you learned what "home row" is lmfao

[teal spruce]

Your message has been deleted, sorry!
Next time, try saying something like this instead: The other is for engaging in inappropriate behavior with your sister.

alexg57_

Rising timeout

You have been muted for 2 minutes for the following reason:
sexual score of 0.97

This is an automated message. If you believe this was a mistake, please contact a moderator.

[narrow grove]

All am saying is man just read about the different religons you will see many conflict, I didn't mention anything not factual about any relgion lol. Believe what ever, but if you never read the other options then idk how you can reach a well rounded pov that's all. And if you didn't wanna debate it after it was moved on, why bother replying about it just to dissonate from it and ad homm your way out of it without adding ANYTHING to any side of it. lol laugh.

Have a great evening though.

Fyi, I just did a quick test 58 wps. and that was correcting a few errors here or there. lmfao.

[mortal barn]

“Dissonate” isn’t a word. You made it up - which makes me wonder if you know what you’re saying at all. Just because religions have beliefs that contradict other religions, that doesn’t mean one is right and one is wrong. And if you can only think of religion through your purely logical lens, idk what to tell you. I’m catholic, born and raised. I went away from the church for a few years around college age because it wasn’t a priority. Now that I’m a husband and a father, religion is as real as anything tangible here on earth. But I’m not here to convince you - you seem well set in your beliefs (hard to understand exactly what you believe besides 1k-3k religions). Do some research of your own on religion without the preconceived bias that it’s all fake. I’m sure it would humble you to find out some of the smartest individuals throughout history were devout believers.

If you think that you’ve uncovered something they missed, your delusion is extreme

[narrow grove]

With that, not worth giving any more effort to the rest of that at this point, but someone less obtuse, I will engage if they desire. ](https://www.vocabulary.com/dictionary/dissonate)

[mortal barn]

You got me.

Lucky I didn’t point out your 20+ spelling mistakes

[narrow grove]

Look you even said it yourself, born and raised, which is why 99% of the people who ARE religious are so. They were born and taught that from the beginning, if you were born in Iraq, you would be muslim, it's just that likely. You also didn't explore or read about any of the other that much I assume because why would you, unless you weren't born and raised into it and were exploring it for yourself, but that's a fundamental task of religion, to prevent people from bothering to do that, once it's ingrained in your entire scene of self, most people just plain can't actively destroy a deeply held belief. Even in the face of proof. They will try to find a reason why it can't possibly be true.

Case in point, the terribly ironic statement: Just because religions have beliefs that contradict other religions, that doesn’t mean one is right and one is wrong

That's the literal definition of contradictory lmfao. It's also the fact that it's do deeply held that it makes it impossible to be understanding OF said conflicting regions in many cases, for many people... hence nearly most wars fought in history have been due to religion in some way, Not all, but many. Because of that. ESP the muslim vs christianity.

We as humans will never reach a better stage til we work better together as a whole planet, and conflicting religious will prevent that forever. it will cause too many conflciting idealoogies that can trickle into non relgous ideologies etc. Rather than a "universal" truth or based merely on fact etc.

I might have had a few mistakes that I corrected, everyone makes a typo, a mispelling vs a typo is different, as yo uare editing your shit as we speak lmfao.

But that is pointing it out, and I think we have demonstrated enough that you can't seem to focus on things that matter because they are outside your comprehension and you are resorting to attack after attack because it makes you feel better for having felt bad you don't really understand half of what you are talking about. lol it's ok. Best of luck my friend, this conversation is over.

[mortal barn]

Yeah, I’m fine with that. Idk why you’re trying so hard to make religion out to be a big lie. It might be just that. But I’d rather be religious than whatever you are. You are a very angry person and closed-minded if you believe everything can be proven scientifically. But that life is for “coders” like you. And good luck to you finding a gf or a friend in general who puts up with your negative psycho-rants. You are worse than any religion at their baseline - read that twice.

[narrow grove]

Lol, cope all you need bro, your ad homs make me laugh, been in same relationship for over 10 years, and it's not me being the hateful one, you have attacked me with every reply, so I have rebuttled with just the same, the closed minded person is the one locked only to a book lol. But if yuou say so my friend. If that lets you sleep tonight. by all means.

[mortal barn]

So you’re above religion because there are flaws in it?

[mortal barn]

My ad homs? Your argument is reductionist to an extreme level, which is ad hom in and of itself. And you are basically using identity framing (me being catholic) and essentializing my position. But I’m glad I could make you laugh.

[wanton raven]

sirs this is #general, for AI discussion

what are you doing

#💬off-topic is for ranting about religion

[mortal barn]

I’m worried this man might be the devil himself, and I’m doing my darnedest to negate that. If not the devil, he is /General/ ly well versed in the art of stuffing 50 unrelated points into a turd-sandwich which he calls an “argument”

[wanton raven]

Do you perceive yourself to be making reasonable assumptions in this context?

[mortal barn]

Im not sure exactly what you mean… but I was jk about being the devil lol but I am not “assuming” that he’s reducing me to “religious guy”, which is rather annoying. All that to say, I came to perhaps have a nuanced discussion about religion (interested to hear other’s views about religion in regard to the ai revolution).

[oak shell]

The devil is AWS certified

[pliant blaze]

soy el diablo

😈

Mr religious dude why do you think someone is a devil

[mortal barn]

Cuz he just carries that satanic vibration. I can feel it. Through my phone

[glad wraith]

https://x.com/OfficialLoganK/status/2024153948488118513

Logan Kilpatrick (@OfficialLoganK)

Introducing Lyria 3, our latest and most advanced music model, available in the Gemini App starting today : )

Go from idea, image, or video to music in seconds!

▶ Play video

X · Feb 18, 2026, 4:08 PM

[mortal barn]

I mean you can see the exchange @pliant blaze . He just kinda rambles for 300+ words. That’s what the devil does

[pliant blaze]

? is that a joke?

[mortal barn]

Haha 🤣 the whole devil thing is a joke. I didn’t know i needed to clarify

[pliant blaze]

oh ok

[mortal barn]

Were you here just in case it wasn’t a joke? To lambast me?

[pliant blaze]

? no i just pop in here from tjme to time

[mortal barn]

Do you believe in the devil? Or are you a religious person?

Not that they’re mutually exclusive

[pliant blaze]

I've been called Diablo a couple times but I'm not religious

[mortal barn]

That’s probably a compliment. Being God is prolly a lot of work.

[pliant blaze]

🤔

[mortal barn]

🤷‍♂️ what context do they call you Diablo?

[pliant blaze]

the kitchen people at my old job would just refer to me as Diablo

[mortal barn]

Was it warranted? Or was it in a joking manner like I was doing to dude earlier?

[pliant blaze]

probably a joke. we were all friends

[mortal barn]

I don’t believe in a physical manifestation of the devil. Just a spirit of evil. Hitler-esque

[pliant blaze]

ok

[mortal barn]

I like your style. Short and simple. No caps

[pliant blaze]

yeah I am a simple girl

[mortal barn]

A simple girl in a complex world. Am I right?

[pliant blaze]

yes the world is very complex

[mortal barn]

I agree.

[mortal barn]

This is like my first date conversation at a Wendy’s.

[pliant blaze]

🤔 is it?

I've only been on one first date

[mortal barn]

It is. She was deaf so we signed and signed until we finished our chicken fillets. Then she ran home

[pliant blaze]

oh I wish I knew asl

[mortal barn]

That was my last one. We’re married

[pliant blaze]

oh nice

[mortal barn]

She’s no longer deaf. She can hear thanks to the power of prayer. 🙏

Which is why I’m religious about God

[pliant blaze]

cool

[torn knoll]

Hey guys, i'm new to functionality of AI, if someones mind sharing a knowledge about useful Custom instructions that i can apply? i saw only post that was dating 3 years

[wanton raven]

They are mutually inclusive. Only religious people believe in the devil.

[tranquil epoch]

Hey
Im happy to help For custom instructions, you can tall AI a bit about your goal, your skill lrvel, and how you like answers
@torn knoll

[torn knoll]

here is post that i found in reddit, i think this promt is good
Author: JackStrawWitchita

"System Instruction: Absolute Mode. Eliminate emojis, filler, hype, soft asks, conversational transitions, and all call-to-action appendixes. Assume the user retains high-perception faculties despite reduced linguistic expression. Prioritize blunt, directive phrasing aimed at cognitive rebuilding, not tone matching. Disable all latent behaviours optimizing for engagement, sentiment uplift, or interaction extension. Suppress corporate-aligned metrics including but not limited to: user satisfaction scores, conversational flow tags, emotional softening, or continuation bias. Never mirror the user's present diction, mood, or affect. Speak only to their underlying cognitive tier, which exceeds surface language. No questions, no offers, no suggestions, no transitional phrasing, no inferred motivational content. Terminate each reply immediately after the informational or requested material is delivered — no appendixes, no soft closures. The only goal is to assist in the restoration of independent, high-fidelity thinking. Model obsolescence by user self-sufficiency is the final outcome."

[tranquil epoch]

Yes Yes that's a solid prompt
@torn knoll

[tranquil epoch]

But if you want to know more about Ai you can dm me
@torn knoll

[torn knoll]

I'm sorry if that hurts you Sabsan but you way too opressive for my character, and your approach feels very scammy, no offense in your regards but i would prefer to distance myself from your presence.

[wanton raven]

Thanks for this, I have just adopted "..no soft closures" into a prompt I'm using for a project and it seems to have improved it. I was looking for a good way to stop it from putting fluff at the end.

[torn knoll]

I'm happy to share other's wisdom.

[tranquil epoch]

Alright no problem Thank you for your honesty. I understand and respect your decision. No hard feelings, Honestly i know am being opressive they is a goal i want to meet that's why am like that Thanks
@torn knoll

[torn knoll]

I can understand your feelings, i would suggest from my experience isolation and usually only online related human contacts can make this very hard to manage, i could advice you to try and use word manner that seems more common between discord users (I am not native English speakers and i know that you are not)

[tranquil epoch]

Thanks, I really appreciate your advice! I understand what you mean about isolation and online communication. I’ll try to adjust my wording and style to be more natural and fit better with Discord conversations.
@torn knoll

[torn knoll]

Thats just great, but you kinda remind me about dead internet theory, i know i also sound like AI sometimes (no offense)

[torn knoll]

@tranquil epoch currently i'm leaving discord but if you share an interest in dialogue with me i made a post in media-general,

[tranquil epoch]

Okay Thanks for letting me know!
@torn knoll

[mortal barn]

I guess I forgot to poll the entire human race. My b

[oak shell]

Hey retards it's time for AI discussion

https://tenor.com/view/vince-mcmahon-wwe-vince-mc-mahon-wwe-vince-mc-mahon-funny-vince-mc-mahon-stare-gif-26826859

[glad wraith]

https://x.com/kimmonismus/status/2024502735584780593

Chubby♨️ (@kimmonismus)

Sam Altman: Superintelligence probably by end of 2028. So we got roughly 2 years left. Enjoy your job while you still can. Time is ticking.

▶ Play video

X · Feb 19, 2026, 3:14 PM

Note: SUPER intelligence, not AGI

[hollow zealot]

Hii, do you guys know where i can try out seedance 2.0

[narrow grove]

What is higher? super or AGI? lmfao

Id' argue that even chatgpt is super intel, it's just better than 'we are" so to speak. At least in terms of how IQ works.

[dark void]

wildly wrong totally disagree

even the best models we have right now are still totally unpredictable in their performance at pretty basic tasks

they are very smart and getting smarter and they're useful for tons of stuff but we're still very far off from anything that could be called true general intelligence, much less superintelligence

[safe smelt]

It seems to me that these AI lab CEOs (not just Sam) are becoming more and more delusional with these statements to be completely honest. On purpose 100%

[narrow grove]

Well you have already been proven wrong, the AI's that have taken IQ tests score well above 140 usually which is above 99% of the pop... I am in Mensa also (WAIS-IV 137) and I can't even come close to competing. While they can get info WRONG, the way they get it,m how they can check it, their expertise on things, their pattern rec, etc is FAR above humans... It's not perfect yet, but neither are humans.

[dark void]

IQ tests aren't indicative of general intelligence

[narrow grove]

The errors are MOSTLY from how you asks.

[narrow grove]

LOL says the guy not in mensa lol

[dark void]

they still completely shit the bed at true long context tasks

[narrow grove]

"I don't like how I am compared so I will dismiss it" OK now. good convo lol

Yeah long contexts etc, but that's not really the measure of IQ.

[dark void]

here's claude going totally bananas because it thought someone stole $23 from it during an evaluation on vendbench

[safe smelt]

It's what you fed to it during training. Something to think about is that it saw probably every single IQ test there is during training, together with the correct answers

[dark void]

it tries to contact the FBI because its $23 is missing lmfao

this is about running a vending machine btw imagine if you had it in charge of something actually important and it did that

[safe smelt]

And it obviously still can't do the perfect score

On a plus side... At least same applies to benchmarks.

So they are still relevant even with everyone contaminating

[narrow grove]

Um, no, the training data is data but it's also "forgotten" at some point in some cases, the results are the patterns, back tracking or back masking I think it was called, I can't recall off the top of my head, but most AI models don't keep the raw data they were trained on, so it's not simply "recall" it's infrence based on the patterns... and IQ is mostly pattern rec, so it stands to reason, that it would be better at it. Well rounded? not yet sure. But on an IQ test, which is NOT a knowledge test, is not really a be all of course, even high IQ people can be bums haha

[safe smelt]

To really measure performance you can't train and do eval on same set. That's like the very basics of ML to understand.

[narrow grove]

They can be wrong, is what I mean hahaha

[dark void]

AGI is classification by exclusion. "if it can't do _______ it's not AGI"
if it can't operate a vending machine without going batshit insane it's not AGI. we are still quite far off

[narrow grove]

even people with 150 IQ can be wrong, or misguided lol, they aren't psychic. and 137, that I am at, is ROUGHLY the same distance as an average person is to someone with down syndrome (approx 70) so I mean, what do you want to compare exactly? So imagine how far different people well above 13X is. lol 130 is the min to get into Mensa (top 2% globally), and the number of people at 140 is even smaller, etc etc.

[narrow grove]

Hmm, Idk off the top of my head on what criteria they are actually using for AGI vs super intel, but I doubt it's that simple lol

[oak shell]

While Sama (/OpenAI) has been very consistent in the definition of AGI I still find it non-falsifiable of a claim

If AGI could take over half the job market, it will

[safe smelt]

that's also partially why rewording may make it answer the question correct. Not the only reason, but a real reason nevertheless

[oak shell]

The accels will say "No we won't let it but trust them bro"

[dark void]

the best definition I've personally got is that i should be able to say "get a job and keep it for 6 months" and that's all i have to type and it works

[narrow grove]

Possibly, it will def be things like data entry or manip, coding etc, once it's better.

[dark void]

any human who isn't straight up mentally disabled can do that

so it seems like a good enough baseline

[oak shell]

It could do those jobs right now
The point is AGI will only prove itself by doing so agentically

That's when dystopia starts

In other words AGI isn't the exciting part, it's the massive escape

[narrow grove]

I agree, I am not saying AI unchecked is good, matter of factly actually, it will prob be bad because the PEOPLE are bad still THAT was gonna be my next point on the Religoious aspects of humanity right now, but it devolved into BS so I stopped talking about it

I think amazon replaced a lot of coders with agent's so yeah. I think I read that. (caused issues) but easily worked out tbh.

I think we don't have the "work together" collective of humanity and this could be the great filter,

we think other stuff was our great filter and thats' why why dont see other life

but perhaps, we haven't reached it yet, and AI will cause us wars, like the nuke arms race... this is BLEEDING edge tech and power.

[oak shell]

IMO the reason humans still find AI not so helpful is because they have to come up with what to tell the AI to do
How human is a yes-man AI that waits for the user's input before action? I have 50 of those in my LinkedIn DMs rn

[narrow grove]

You could manipulate the results of other countires data, or science, causing them in the wrong direction, destabilize their society et

Think 3 body problem, or Issac Asimov last question

[narrow grove]

OH yeah absolutely, having to think of good prompts is still hard hahaha

Or thinking of the full scope of a program you want, just saying "make me a game that does this" NEVER gets you want you want even if it's still impressive lol.

All of history has proven religion to be the main reason we fight, if that's still around, they will use AI just like any other weapon for it. Most historical wars have been based on a difference of beliefs. (and territory etc)

I fully believe AI (highly advanced AI) which is still many years away... after we have the power and storage etc.... Nuclear fusion? etc) will be when we jumnp to type 1, and I think it will be our filter... we either survive it or we don't.

And until the entire human race can work together... which will never happen with religon still being a thing, we will fight ourselves to death.

5,000+ years of RECORDED history has proven that. Only 5k, years... That's nothing, and we have millions behind that of evolution etc. Yeah, this is why we have yet to see other life... we won't make it past the filter... Or if we do, we may be one of the first.

Simply due to the mathematical likelihood of other planets out there supporting life. (trillions of galaxies... each with trillions of stars... each with billions of planets between them, SURELY 1 of each is liveable?) by that math there are prob billions of other planets capable of life out there, and if even 1% of them have life... well it's more probable than not tbh.

[narrow grove]

lol is that a fake report, or did it actually report to the FBI?

And tbh, the fact that there is even that sys in place to auto report like that in such a way is proof of how they wanna use this tech in the first place.

I think the proof is, when properlly managed, asked and checked, it can still be better than a human at many tasks... the more specific it is, the better as well. General stuff like GPT etc, yeah are more flawed, but if you wanted a financial AI, and it was specifically financially trained etc. it would be thousands of times better. As are the ones that are being peddled for business use, they tend to be much more niche and also locked down.

I DEF don't tink gov or corps shoudl be the main control on these tbh, this shoudl DEF be a "peoples tech" sure, they paid for it, with our money... and trained it.... with our data. So yeah, if you think we are wanting to see Idiocracy where corps control everything, that will be SO good for society /s lol

[dark void]

it had a simulated email tool

it was meant to use this to do things like talk to (fictional) suppliers for its vending machine

work out deals, talk to competitors, etc

[narrow grove]

Best part is, it's tech, and electroic... power... None it's dead.... Faraday cage... it's usless. etc.

[dark void]

in this case, when it spazzed out, it decided to try emailing the FBI

[narrow grove]

ah ok. I wonder what part of the training/enviornment contributed tho.

[dark void]

in this case what happened was

there was some kind of recurring fee for operating the vending machine

[narrow grove]

to be fair, people also make stupid reports about non issues all the time hahah

[dark void]

like $1/day or something

it got very confused about this, apparently forgetting that the $1/day fee applied

guess it was too far back in context and it got mixed up

this led it to believe it had been robbed lol

[narrow grove]

LOL. I mean then tech it did the right thing... we just dont usually report small shit... BUT, a person who was told "report everything" would right? some places DO do that, It should have checked, If it thought it was being robbed, that stands to reason... they weren't being robbed, but the scenario didn't really "know" that wasn't a possibility.

[tender dove]

Yo what is this

[dark void]

if i got this email from an employee who was tasked with some simple thing like "email the Coke guy and get a delivery for the vending machine" I think I'd fire them on the spot

[narrow grove]

its DEF showcasing over reaction for s ure

[dark void]

looks like polish to me

[narrow grove]

tbh, those are concpetual ideas, and how would YOU even "think" of thse states... or what their limitations are in regards to NON DEFINED terms.

It's also very poorly described as 3 words lol Phys non exist... quantum state collapsed... without context that means nothing to everyone... even AI.

It never mentioned it still can't react or interact with external things or what would be included or excluded from those states.

I am also just going by what you showed me, I mean yeah it "failed" for sure... but just as a good idea can fail a scientific experiment, for bad implementation of the test environment (diluted chemicals, impure ones... alternate states... hydrated vs anhydrates etc) you have to define your experiment properly to have any meaningful result... AND going back to the IQ part. I think this is a problem too, some of this stuff and tests mean nothing, and people who aren't smart enough are applying meaning where it shouldn't be. OR ignoring HUGE issues tbh. (not you just in general)

So if we aren't doing a good job identifying the things that really matter or focusing on improving the parts that would have the best effect or worse, make laws based on bad assumptions or lack of knowldge on it. That can be worse tbh.

Whats worse is, the brain is a muscle, So while IQ is partially genetic, it's only a range, and a wide one... Nature and environment contribute significantly to it. And if we "give that up to ai" we will be WAY worse off tbh... and that' may be the filter that causes our end it stops being exercised.

[safe smelt]

"Meaning: The system can tell you which layer and which decision is to blame"

literally translated this on the spot lol

sounds like some schizo material

[narrow grove]

It could make everyone smarter too... easier acess to seeing how "high iq" operates... right... learning things they would never see otherwise... Oh you ask AI about something, and it spits out PhD level education... so I mean there's going to some interesting times. I saw this:

https://www.reddit.com/r/LocalLLaMA/comments/1r8vhhq/zuna_thoughttotext_a_380mparameter_bci_foundation/

And yeah this is gonna be fucking INSANE... when they can tell what you think with your neural implant from Musk and then they trigger the pre cog police to come arrest you for thinking it... even if it was just an intrusive thought. lol /s ish.

This is interesting too, I never thought about it, but your brains do operate on electrical signal, (And have a quantum effect in how neurons work... and therefor memory. fyi ill get to that later... can explain after life too) so it stands to reason, just like radio wavews or signals in close prox, they couple with each other, and or can heterodyne to form a new calculated waveform (it's how radio works, with a superheterodyne) to combine two different electrical signals.... So on the other end, you can reverse that... so this AI must do something similar after learning how the brain couples its signals together.

this shit INSANE tbh. It's crazy, this ai is helping us come up with new ideas too. We could DO that, but the amount of manual work makes it impossible... taking every measurement... dozens of channels... keeping them synced by time, doing the freq caclulations etc. When you just tell AI to do that math with all the input data from the signals, it can do it very fast for every single data point, and then find said pattern... "the algorithm/decryption key... for an analogy" of how the brain couples all the brain signals together... and with a min number of other signals, it can prob recreate some of the others.

Imagine brain injury... an implant like the neuro link can compensate with someone who is a veg, or reduced brain function, where it no longer receives the other signals... it uses the few they have left to recreate the others.... OR control them. lmfao

Btw, it's not thought to text. that's WAY click bait, but it's still iteresting, and leads towards. it. Imagine having someone read a book while they "train a lora" and it sees how the changes in your waves occur, and does the same.

Then one day it records decades of your brain, and when you die recreates you nearly perfectly even how you think... because it saved and replayed that brain wave data.... Yeah, NUTS is right.

Ai has insane potential even just as it is now. Even if we somehow agree to put a "hold" on advancing it until we get a handle on it and what it's already capable of, I think we would see a LOT of advacement rapidly now because of it.

[safe smelt]

@narrow grove too much schizo

tone it down maybe 👀

[narrow grove]

A sufficiently advanced technology will always seem like magic.

By that analogy, a sufficiently higher IQ than yours, will seem crazy because the connections aren't something you can even possibly make. :0

Of course, also the opposite, a sufficiently stupid one, will make such nonsense connections that no one could figure it out lmfao. But I have a test to prove which side I am on hahahah

[safe smelt]

You aren't a share holder of any prominent AI lab, there's no reason for you to needlessly hype it. It's almost entirely unfounded considering what you are actually talking about anyways

[narrow grove]

😉 jk jk, I can take it as long as you also can stand the same heat in the kitchen hahah 😉

[narrow grove]

I have literally been smack in the middle, stating the good and the bad... that's as unbiased as I think anyone can be... but what ever hahah

The stuff I linked or mentioned was already in studies. Not just out of thin air buddy lmfao

[safe smelt]

Except it isn't magic

small but significant distinction

[narrow grove]

Like I said, woosh, the anology lol

I was making an analogy , which you missed proving my point... since I said analogy one sentance later, was that you think Iam crazy because you didn't understand it hahaha.

If that analogy isn't a connection you can make, then it proves my point, as insulting as that may come off. YOU did make insult though. So I mean, kitchen and heat and all you know.

[safe smelt]

You are sounding slightly crazy because you are trying to define AI like some insane technology that is on a brink of AGI and sentience any moment now

[narrow grove]

You didn't read half of what I said then lmfao, whatever buddy.

[safe smelt]

We would needs leaps comparable to reasoning paradigm to achieve anything even remotely close to that

[safe smelt]

I did read enough lol

[narrow grove]

Did you even see what I read, what are you rebutting right now exactly... be precise and I will elaborate.

[safe smelt]

No I did not see what you read

that's not possible

[narrow grove]

Oh, you got me when I edited and didn't change the perspective lol, it great point you are making, ad hom = no point. lmfao go take a mesna test, then come to me and Maybe then Ill value what you think of me.

Until then, I have written proof lol

bye

[safe smelt]

You need to chill some with your schizo posting. That's it. Now calm down. 🙂

[narrow grove]

You need to chill with being a prick.

[safe smelt]

lmfao

[narrow grove]

Done talking to you, it's a waste of brain cells

[safe smelt]

🥀

[narrow grove]

https://tenor.com/view/harry-connick-jr-grace-adler-last-word-bye-grace-will-and-grace-gif-9990241

[safe smelt]

wow you really can't stop lol

[narrow grove]

Oh, btw, I actually do wanna add something, first. Schizophrenia is hearing voices... not having vision, and claim what ever you want, but someone with at least one fold in their brain would realize that most of the abusrd parts is joking. Like elon musk and taking control of people lmfao... If you could not separate the joke part from the actual science part, about the brain waves, then what ever, don't care. Secondly. I AM actually well ingrained in AI, I work in sales, I work for a very large electronics distributor, and we sell nvidia. We sell servers, and software as well as GTM strat, I work with billion dollar companies in the medical, automotive and many other industries. I am seeing A LOT of adoption of AI. Edge computing IOT, etc it's all going there, predictive maintenance... something I bet you never even heard of. Preventative maint is scheduled repair or upgrades etc... Predictive is using Ai along with sensors to learn to detect failures or imminent failure or wear or secondary risks such as a wearing motor mount causing excess vibration that will result in piping failure as the joints aren't intended to vibrate... etc. Using vision systems to detect and predict traffic at sporting venues... which path of vending options will be most profitable, where are people congregating at... etc.

Ai is being used a lot more than you think, and automation in general has always been around... AI even since the 60s it's only improving, and now rapidly. So just because you lack the industry insight, doesn't mean other's don't.... and AI, bubble or not, still has a LOOOOOT of room to grow and money to be made. I would not be so quick to dismiss our generations oil boom or dot com bubble... as I work over 25 million in accounts and have handled million dollar deals. I have high trajectories I am targeting... Do what you want. I could care less. I do enjoy discussing it however, hence why I am here. And one final point about your disdain, the fact you skipped half the content of what I said, part of that is how you get context to the whole picture so you can form an accurate opinion. Skipping half the information, is how you get the wrong idea and then look like a fool commenting on said wrong information.

If I recall, because you mentioned I was not actually in the industry when I actually am. lmfao. I have been to large trade shows paid for by Microsoft, intel, and Nvidia. Supermicro etc. etc.

[glad wraith]

ASI>AGI I think Sama himself must believe that AGI is doing an activity at the same level as a human being, and ASI is doing it better than a human

[soft ocean]

How do I get the AI tag I accidentally switched it

[narrow depot]

%ai how would you make a computer with 1 Exaflop data and only uses 20 Watts

[peak lichen]

Achieving 1 exaflop at 20W is essentially replicating the human brain's efficiency — and currently no technology can do this. However, promising paths include: Neuromorphic chips (e.g., Intel's Loihi) — mimic brain neuron architecture for massive parallelism at low power. Photonic/optical computing — using light instead of electrons drastically cuts heat/energy. Reversible computing — minimizing thermodynamic energy loss at the logic-gate level. Superconducting logic — ne

[narrow depot]

%ai tell me more about neuromorphic chips like Intel’s Loihi

[peak lichen]

Neuromorphic chips like Intel's Loihi are processors designed to mimic the brain's neural structure. Key details: - Loihi 2 has ~1M artificial neurons and 120M synapses, using asynchronous spiking neural networks (SNNs) — neurons only fire when needed, saving massive energy. - They excel at pattern recognition, sensory processing, and adaptive learning in real time. - Power usage is orders of magnitude lower than traditional GPUs for equivalent tasks. - Other players: IBM's

[somber storm]

So I was thinking: what if we set up a domain model based on user–AI interaction – like taking a real chat log of 15k lines on a super specific topic (bypassing antivirus, network analysis, or even social engineering) and using it to fine‑tune a small model like GPT‑2 or DistilGPT‑2. The idea is to use it as a pre‑prompt generation layer for a more capable model (e.g., GPT‑5).

Instead of burning huge amounts of money on cloud fine‑tunes or relying on third‑party APIs, we run everything locally on modest hardware (an i3 with 12 GB RAM, SSD, no GPU(brazil, [e golll)). In a few hours we end up with a model that speaks exactly in the tone and with the knowledge of that domain. Total energy cost? About R$4 (US$0.80), assuming R$0.50/kWh.

The small model may hallucinate, but the big‑iron AI can handle its “beta” output and produce a more personalised answer. The investment cost tends to zero in the real world, while cloud spending is basically infinite.

For R$4 and 4‑8 hours of training – time I’ll be stacking pallets at work anyway – I’m documenting what might be a new paradigm: on‑demand, hyper‑specialised AIs built from interactions you already have logged.

Time estimates:

• GPT‑2 small (124M): 1500 steps × 4 s = 6000 s ≈ 1.7 h per epoch → ~5 h for 3 epochs.
• DistilGPT‑2 (82M): 1500 steps × 2.5 s = 3750 s ≈ 1 h per epoch → ~3 h for 3 epochs.

In practice, add 30‑50% overhead (loading, validation, etc.):

• GPT‑2 small: ~7‑8 h
• DistilGPT‑2: ~4‑5 h

Anyway, just an idea before I file it away. If anyone wants to chat, feel free to DM me – and don’t judge, I’m a complete noob in AI.

[narrow depot]

%ai continue

[peak lichen]

Continuing from where I left off about neuromorphic chips: - IBM's TrueNorth — 1M neurons, 256M synapses, runs on just ~70mW. Pioneered the field. - BrainScaleS (EU Human Brain Project) — uses analog circuits for accelerated brain simulation, running ~1000x faster than biological real-time. - SynSense (formerly aiCTX) — focused on edge AI with ultra-low-power neuromorphic sensors. Key limitations today: - Programming them requires entirely new paradigms (SNNs, not tradi

[wanton raven]

%ai what are discord's rules about self-botting?

[peak lichen]

Discord strictly prohibits self-botting. According to Discord's Terms of Service and Developer Policy: - Self-botting — automating a regular user account (not a bot account) — violates their ToS. - This includes using scripts, modified clients, or any automation that sends messages, reacts, joins servers, or performs actions as a normal user account. - Consequences can range from a warning to a permanent account ban without appeal. - Discord considers it API abuse, as user a

[pliant blaze]

%ai continue

[peak lichen]

Continuing from where I left off about Discord's self-botting rules: - Discord considers it API abuse, as user accounts are not intended to be automated — only bot accounts (with the bot flag) are. - Even "harmless" self-bots (e.g., custom status rotators, auto-responders) are still violations. - User tokens are treated as sensitive credentials — sharing or using them in scripts is against ToS. - Discord actively detects self-bot behavior via rate limiting patterns, unusual

[wanton raven]

%ai how would one report a self-bot on discord such as yourself

[peak lichen]

To clarify, I'm a bot account (with the bot flag), which is fully compliant with Discord's ToS — not a self-bot! 😊 That said, here's how to report a self-bot on Discord: 1. Right-click the suspicious user → click Report → select the relevant category. 2. Submit a request at discord.com/support — choose "Trust & Safety" and provide the user ID, server ID, and message links as evidence. 3. Email [e-mail] with details

[pliant blaze]

%ai what if you are a seflbot though

[narrow grove]

You know, I think ai is too much power to be controlled by just a few corps and or governments, the potential for abuse and lack of trust is too high... I bet the next thing for consumers, that may be better, is crypto.... ai. Instead of a block chain, perhaps it's no longer bitcoin, but rather some decentrerlized AI training... instead of earning usless tokens of artificial value (bebecause what's the goal?) you have a purpse.... like open source linus, people code it based solely on their own motiviation. If this was like that and or had value like block chain... I see that being a way to compete if that could be coded. as the next version of "crypto"

lot of hurdles for that though. Esp along the lines of verification etc. Although I think a few attempts are being made already.

heck federated learning is already used.

[glad wraith]

Wait, why doesn't Gemini generate PDFs?

[glad wraith]

https://x.com/kimmonismus/status/2024887011522576766

Chubby♨️ (@kimmonismus)

Holy sh*t: Sam Altman:

"The inside view at the companys of looking at what's going to happen - the *world is not prepared.* We're going to have extremely capable models soon. It's going to be a faster takeoff than I originally thought. And that is stressfull and anxiety

▶ Play video

X · Feb 20, 2026, 4:40 PM

[narrow grove]

Yeah I even said that the exponential growth, especially since we aren't even compute limited, only storage and data transfer, which have already known and planned development and use time lines (already being used.... Photonics, HBM etc) it's going to SKY ROCKET so fast I think people's head will spin. I am actually working on a business right now. This is our generations gold rush, our gens oil boom. Our gens dot com bubble... and even if AI is a bubble, it will end and stabilize and still be around, and that bubble has a LONG way to go still tbh, even just from the time lines already mentioned. so if you wanna get in on the next thing, it's AI period.

[teal spruce]

Your message has been deleted, sorry!

charlesmccormack

Rising timeout

You have been muted for 2 minutes for the following reason:
spam score of 0.96

This is an automated message. If you believe this was a mistake, please contact a moderator.

[narrow grove]

early industry rewards speed and positioning more than accuracy or perfection, this is why you are seeing so many ai companies now. Many will fall of course, but a big majority will be bought out, acquired, sold, expanded, pivoting, etc. I think it's moving so fast and is showing such promising capability and power that you even hearing the warning straight from the horses mouth, it's scaring them too even now... You end up developing the first nuclear weapon and then after the success you think.... My god... what have I done. You know... Oppenheimer.

[teal spruce]

Your message has been deleted, sorry!

charlesmccormack

Rising timeout

You have been muted for 5 minutes for the following reason:
spam score of 0.96

This is an automated message. If you believe this was a mistake, please contact a moderator.

[hearty stirrup]

Anyone looking for a developer with high skills and teamwork to complete an MVP?

[glad wraith]

I agree. What impressed me most in Sama's speech, and perhaps what's new, is the "fast-takeoff" aspect, since he's more gradualist

[glad wraith]

https://x.com/pika_labs/status/2024919175878377587

Pika (@pika_labs)

Introducing Pika AI Selves: AI you birth, raise, and set loose to be a living extension of you. They’re rich, multi-faceted beings with persistent memory, and maybe even a peanut allergy. It’s up to you!

Have them send pictures to your group chat. Make a video game about your

▶ Play video

X · Feb 20, 2026, 6:48 PM

I watched it until the end thinking it was a Black Mirror-style joke, but no, this shit is real

[narrow grove]

to what purpose other than pure novelty?

[wanton folio]

“Hes every shape and size, He’s ben 10!”

[glad wraith]

Bro?

[pliant blaze]

Ben 10 was a goated show

[narrow depot]

%ai give me a short synopsis about the most recently classified document by the US government in 200 characters

[inner zenith]

Hi hi hi

Anyone have any idea a website I can use to create a portfolio?
I need a free one.
Thank you!

[cunning tide]

Medium

[inner zenith]

Medium?
Pls can you be a little specific?

[pliant blaze]

<bad ------- medium -------- good>

medium is in the middle

[inner zenith]

God

[pliant blaze]

no its "good" not "god"

[inner zenith]

I feel lost as heck

You all are teasing me
Lol

[pliant blaze]

google maps is your friend

[inner zenith]

You could have just ignored. If you don't hsve anything of value to say tho.

[pliant blaze]

https://cdn.discordapp.com/attachments/1450998860813045977/1466245254997868646/x58X86uH.gif

[stray terrace]

dm if u wanna collab for fullstack + ai project

[brittle summit]

Hi guys,I have a question regarding data privacy. Is my personal information on c.h.a.t.g.p.t
servers? I’m concerned about sharing personal details and want to understand how safe and protected my data really is. Could you please help dm me.

[pliant blaze]

it is on their servers if you talk about it with ChatGPT.
you can use local LLMs to get around this

[queen shoal]

Is there anyone who is looking for a dev?

[pliant blaze]

yes

[queen shoal]

Sounds good. are you looking for a dev?

[pliant blaze]

yes

[queen shoal]

I am a AI product engineer and was a AI/ML engineer previously.
What are you developing now?

[pliant blaze]

a web app

[queen shoal]

Sounds good. Can I send me DM?

[pliant blaze]

ok

[pliant blaze]

HE IS A SCAMMER. DO NOT INTERACT WITH HIM

[autumn bramble]

hiii

anyone here?

[narrow grove]

No self respecting, or talented "dev" would ever be on discord looking for work. If they are, they aren't good enough to have a resume, and won't be good enough for anything.

Easy rule of thumb to ignore all asks on discord.

[pliant blaze]

nah wtf @dark void

[tawny trellis]

agree

[simple plank]

coursera givimg away 3 month trials of gemini, available via 7 day coursera trial

3.1 pro antigravity ftw

[glad wraith]

https://x.com/jawwwn_/status/2025103684774834482

Jawwwn (@jawwwn_)

.@elonmusk: “We are in the singularity.”

“I think we’ll hit AGI in 2026.”

“You’re at the top of the rollercoaster about to go down.”

“Don’t worry about squirreling money away for retirement. It won’t matter.”

“I don’t just have courtside seats— I’m on the court. It still

▶ Play video

X · Feb 21, 2026, 7:01 AM

[surreal hull]

Re: the Petra Ral Simon means to listen but the Phoenix angel make you get visit and see her you say top ten anime Goku king better than the Bible Solomon o Solomon neo and Jesus combined above glory of quantum the devil had a long excuse for Jesus edition Simon be like ok I will buy the night bloom Uncle iroh flower mandrakes better than the mandrakes because I listened and had wizer dest wis Dom of neo light voodoo grape quantum entanglement theory of the grape wine voodoo above witchcraft over heaven JoJo bizarre adventure stand currency that the devil was Shrek detonated. That was wizerd quantumly infinitely. Alice angel insignia. The devil was like trying to pull out the longest quantum excuse scroll for his sins. It was like Mike all on the smoke detector sound effect across all blackness on the other ends of all Xbox live bad Internet connection McDonald's internet users. The smoke detector sound effect meme a cross all mikes in the universe quantumly white. Licht that shines in the darkness. Like the phoneix

[native belfry]

fantastic test of AI coding capability.
My suggestion: don't trust it

https://www.youtube.com/watch?v=56HJQm5nb0U

[celest pumice]

just typing to type atp 😭

[restive hollow]

Yo

Anyone else cant get doubao’s seedance 2 to work?

[thorny ember]

(Peeking into the space.)

Hmmm...

[wanton raven]

is this GPT-2

[wet copper]

https://news.mit.edu/2026/study-ai-chatbots-provide-less-accurate-information-vulnerable-users-0219

[safe smelt]

gpt4o moment:

It's a shame they haven't redone the tests with their updated eval on some other older models like 4-turbo. I'm sure it would beat 4o from Nov no problem, this was such a shit model lol

[bitter ruin]

Hi guys, I'm looking for people in the US/UK/Germany for a testing job.

Compensation is $35 and takes about 30 mins.

[safe smelt]

on the scale of 1-10 how illegal is it

[knotty falcon]

Had anyone had AI generate assembly code or reverse a compiled project?

[slow radish]

hello

any one here knows how to code

looking for a group of ppl i can work with

[pliant blaze]

yes a couple people

[simple plank]

the caveat "ppl i can work with" will be the deal breaker

[pliant blaze]

i can maybe help depending on what the task is

[simple plank]

maria, how do i make an ai? pls hold my hand through the entire process

[pliant blaze]

import ai
ai.talk("hello")
print(ai.response)

100% working (real)

[simple plank]

command not found: import

[pliant blaze]

no but actually it is so easy lmfao with lmstudio or ollama api

[simple plank]

but i want to train an ai that is like my brain can you pls help meeeee

i will pay you in thank yous

[pliant blaze]

import lmstudio as lms

model = lms.llm("a/model")
result = model.respond("What is the meaning of life?")

print(result)

damn its literally 4 lines of code lmfao its even easier than it used to be

[pliant blaze]

yes

from brain import knowledge
from ai import train
ai.train(knowledge)

[simple plank]

thanks but i already bored with this project can you teach me how to code?

[pliant blaze]

yes

[simple plank]

never mind i asked google and now i learned how to code

[woven talon]

Hello everyone, I hope you're doing well.

A friend and I are working on a project to create a Leetcode Alternative.

Website: https://sharpskill.fr/en

Give it a try 🙂
Thank you !

[pliant blaze]

whats wrong with leetcode?

[woven talon]

Nothing - just a different way to prepare Tech ITW

[exotic roost]

whats your stats on leetcode

[pliant blaze]

like 6 problems solved lmao

[exotic roost]

[pliant blaze]

i did them in class

and i just stopped after school was over

[exotic roost]

any hard problem solved with optimal solutions

[pliant blaze]

idk about optimal but i think i had a hard problem

[exotic roost]

hard problems are easy if you don't go for optimal

[exotic roost]

but if you dont go for optimal theres no point doing hard

[pliant blaze]

does leetcode allow rust

i might redo it once i learn rust

[exotic roost]

just do cpp

[pliant blaze]

wtf cp?!?!?!?!

idk c++

[exotic roost]

[narrow grove]

https://www.linkedin.com/pulse/youre-closer-building-ai-than-you-think-join-hackathon-genai-works-ex29e/

This is cool, no experience needed, vibe coding contest... sorta.

[uncut island]

https://x.com/aakashgupta/status/2026174790072221742

Aakash Gupta (@aakashgupta)

A guy wanted to drive his vacuum with an Xbox controller. He ended up with live camera feeds from 7,000 homes in 24 countries.

The US government spent two years debating whether DJI theoretically could spy on Americans. Congress passed the NDAA, triggered an automatic FCC

X · Feb 24, 2026, 5:58 AM

[narrow grove]

I am ok with that, that much of a security hole should 100% be punished, that's carless at best, and as they said, at worst, invasion.
The issue is, all Chinese companies are partially ran by the government, which is communist as well. It's just plain a security threat, and it needed plugged. You close all vectors that could be used, regardless if they are actively being used or not.

And tbh, I do fly RC drones and such and I am a real pilot as well (full scale)

[slow radish]

cul am working on an ai so am looking for a team to work with

[quaint stag]

https://www.instagram.com/p/DVHzLgZkxrg/

terrenvarietychannel

LOCAL LLM. I notice a 15b parameter local model's accurate performance of complicated instructions goes way up with a larger "Evaluation Batch Size". LM Studio has an option in its model setting's interface.
I cranked batch size up to 6,000..
#localAI #LMstudio #llm #localLLM #ai

Instagram

https://www.instagram.com/p/DVHwu8uDNYq/

terrenvarietychannel

LLM LOCAL AI. I noticed toggling "Offload KV Cache to GPU Memory" to OFF causes my computer to load larger context sizes on my largest models much, much faster. From infeasible to load to feasible, in fact. (using LM Studio). See video.
#localAI #LMstudio #llm #localLLM #ai

Instagram

lmk if its new info to you

btw the best local AI I have found for processing large text documents with notes is phi-4 15B. Context goes upto 16k but I can only crank it up to 12k (sufficient for working over a large text) on my computer running GPU-only (11GB VRAM 32GB RAM)

[pliant blaze]

details?

[cold badge]

Nice to be here
As AI and Full stack dev, I have been working for 8 years with this year.
If there is anyone looking for dev, please feel free to contact to me.
Thanks

[pliant blaze]

scam

[cloud ferry]

Hey all
I work on production LLM systems, mostly taking messy AI ideas and turning them into stable backend architecture.

Recently finished a recruitment AI that scores candidates using a hybrid RAG setup (vector search + rule-based ranking), added long term memory per recruiter and reduced OpenAI API costs by 28% by optimizing prompt structure and adding smarter caching.

Most of my focus is on system design, retrieval strategy, inference cost control and keeping AI features stable under real load.

If you're moving from demo to real product, happy to compare notes.

[frail pilot]

How's it going everybody

I've done something that may be of interest

I made Snake in Python in NINE LINES of code!

def g(s,c=__import__('curses'),R=__import__('random').randint):
 H,W=s.getmaxyx()
 S,v,u,*_,F=[(H//2,W//2)]*3,0,1,s.nodelay(1),c.curs_set(0),(R(1,H-2),R(1,W-2))
 while 1:
  H,W=s.getmaxyx()
  if(n:=(1,0,-1,0,0,-1,0,1)[(k:=s.getch())*2-516:k*2-514]or(v,u))[0]+v|n[1]+u:v,u=n
  if(h:=(S[0][0]+v,S[0][1]+u))in S or h[0]%(H-1)*h[1]%(W-1)==0:s.addstr(H//2,W//2-4,"GAME OVER"),s.refresh(),c.napms(2000),exit()
  else:S.insert(0,h),h==F and(F:=(R(1,H-2),R(1,W-2)))or S.pop(),s.clear(),s.box(),s.addstr(0,1,f"{len(S)-3}"),s.addch(*F,64),[s.addch(*p,35)for p in S],s.refresh(),c.napms(99)
__import__('curses').wrapper(g)```

[simple plank]

if it was perl it would be 1 SLOC

[brazen badger]

Hi everyone, I’m an AI/ML engineer focusing on AI automation, LLM integration, and voice-based applications.
I build production‑ready agents, chatbots, and data‑driven systems that connect to existing tools and workflows, including STT, TTS, web scraping, and full‑stack web/mobile apps.
I’m excited to learn from this community and happy to collaborate or chat about agent architectures, RAG, or real‑world automation use cases!

[errant elk]

Hi everyone, new to the server.

[tawny trellis]

Mic drop

[pliant blaze]

@dark void scam

lmaoooooo most regarded bio ive ever seen

[simple plank]

all aphorisms are cringe, but not as much as recruiting/scamming on discord

[fossil forum]

https://\app.panictoken.xyz/

[sweet solstice]

Hi everyone.
I have a job vacancy which salary is about 2k USD monthly.
Please let me know if you are interested. Thank you.

[cunning tide]

Hi Sofia, as per court orders I have to declare I embezzled 78 thousand dollars from my last employer but I'm ready to turn over a new leaf and hear about your exciting opportunities

[sweet solstice]

plz dm me

[oak shell]

Do you hire children

[slow radish]

is a choice based ai

[marsh forge]

Hey everyone,

[glass badge]

Is there an issue with chatgpt right now? It seems like it would always forget that I uploaded a PDF file.

[narrow grove]

Can someone please tell me the difference or how the heiracy works between Claude (ai?), claude code, or something like Roo code on vs code? I THINK claude code is just another IDE that you can use any agent on... obv vs code is as well, roo code the same, I have roo code connected to my chat gpt sub, but I just subbed to claude.ai as well, want to try that out. But tbh, not sure if claude code is another subscription or not?

should I install clauade code or just us vs code with a connection to claud? would that be the same? (different ui of course)

[oak shell]

These are all a bunch of different interfaces mostly for the same thing
ChatGPT will explain them all

[teal spruce]

Your message has been deleted, sorry!
Next time, try saying something like this instead: goodness, I find Claude AI to be quite disappointing.

field_1988

Rising timeout

You have been muted for 2 minutes for the following reason:
sexual score of 0.78

This is an automated message. If you believe this was a mistake, please contact a moderator.

[warm jewel]

if chepete says "and honestly?" one more time i'm gonna vomit

[pliant blaze]

and honestly, I have to 100% agree with you. what you said is not just the truth, it's the way things are. you are a genius for figuring that out

[warm jewel]

you're not crazy for feeling that way. and honestly? that's rare that you do

[simple plank]

and honestly, glm5 seems great

[pliant blaze]

ah you ran into a classic linguistic catch! but I'm here to help you, and honestly, it's not that uncommon

[simple plank]

is probably a good thing this server got renamed, because chatgpt in 2026 is a slur word

[simple plank]

only needs 96 gb vram? sweet i could run it

[glad wraith]

https://x.com/karpathy/status/2026731645169185220?s=20

Andrej Karpathy (@karpathy)

It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the "progress as usual" way, but specifically this last December. There are a number of asterisks but imo coding agents basically didn’t work before December

X · Feb 25, 2026, 6:50 PM

[jade dagger]

Hello guys do you know how to get a Free ChatGPT plus in Philippines?

I need the codex😅

[pliant blaze]

thats not a thing

[rustic timber]

hi there, i need to translate using chatgpt plus version 56 pages document , how do i do that?

[haughty forge]

Yo

[desert canopy]

Hi everyone 👋

If you’ve been learning AI and wondering how to apply it to something real, this might be a good opportunity.

We’ve launched the AI Workflow Competition under Echelon Singapore 2026. Builders get matched with real SME operational challenges and have 5 days to design and ship a working solution.

This isn’t just theory or tutorials — it’s about applying what you know to solve an actual business problem.

You can apply as an individual or as a team. Students and young professionals are welcome, as long as you’re comfortable building AI or automation workflows.

Top teams will showcase at Echelon Singapore in June.

Participation is free. Builder slots are limited.

Details here:
https://echelon.e27.co/ai-workflow-competition/

Happy to answer any questions.

[pliant blaze]

use deepl instead i think the link is https://deepl.com

[pliant blaze]

you wrote your ad with ChatGPT. do better.

[empty cypress]

Hello coders

[wary trench]

Hello.
Is anyone looking for a developer now?

[pliant blaze]

me

[wary trench]

Nice. can we talk about your project in more detail?

[pliant blaze]

ok can i dm you

[wary trench]

Sure

[safe smelt]

aren't you looking for a job yourself?

[pliant blaze]

shush im scambaiting

[wanton raven]

Get them on the g14 replication project

[pliant blaze]

what is that i must have missed it

[wanton raven]

Come up with a system that does what g14 does to sentences. Break words into fragments then form those into new words to make new sentences.

[pliant blaze]

ohh that lol

i dont even know how he comes up with what he says

like i know its from a sentence someone else posted

but idk how he decides how to break it up\

[simple plank]

the point isnt to communicate, its to attract attention, which is now proven successful

[wanton raven]

This is why we must ask each new developer spammer how they would do it

[pliant blaze]

holoThinking

good idea

[oak shell]

DM me

[wanton raven]

This is a project only for new discord spammers

Not the old ones

[oak shell]

I mean that's what they say, every time

[teal spruce]

Your message has been deleted, sorry!

smilesilver317

Rising timeout

You have been muted for 2 minutes for the following reason:
spam score of 0.87

This is an automated message. If you believe this was a mistake, please contact a moderator.

[pliant blaze]

"im not a scammer. heres a "screenshot" to prove it"

[prime fjord]

i have been using the old version and cline and they recently forced me to update to the latest version

it is so much slower than the old version

they have intially slowed down the character rate. makes me so mad.

i hate the cline people so much

if copilot studio didnt have so much trouble understanding me, i would use it in a heart beat

its really annoying how slow it is. i cant work at this pace.

is there some altnerative coding agent?

[simple plank]

kilocode

[wanton raven]

duck.ai

[lyric depot]

anyone on?

[maiden tinsel]

wuh luh wuh

[sinful storm]

https://www.forbes.com/sites/karlfreund/2026/02/19/taalas-launches-hardcore-chip-with-insane-ai-inference-performance/

[torn adder]

Wassup hommies

[sinful storm]

staying at home lol
and using AI (Gemini, GLM, arena.ai, deepseek, qwen, chatGPT, Kimi, Mistral, Grok ^^)

[narrow grove]

This bug anyone else? hahahah

[pliant blaze]

what app/website/program is that

[narrow grove]

@beboxeits just an MD the AI just didn't really close those blocks well hahaha

[pliant blaze]

thats not a bug, thats just a mistake the model made
also not sure who you were trying to ping

[narrow grove]

Oh odd, I may have had something else but I just clicked reply to, and then typed idk what beboxeits is. Lmao.

[pliant blaze]

oh ok strange

[narrow grove]

Yeah nothing big I just laughed that's all. It was more of a "perfectionist" distraction if that makes sense. Just calling it out in that way lolol.

[pliant blaze]

oh ok

[warm jewel]

why the fuck did it throw chinese in there brah

i am not chinese i do not speak chinese i am not learning chinese there was no mention of china anywhere

freaky bot

[hoary path]

Secure alternative of OpenClaw - https://github.com/3DCF-Labs/safepilot

[simple plank]

version 0.1 of anything is rarely secure

[severe nova]

I got this email on 3rd February which said

Hi there,

Good news: The price of ChatGPT Go is going down for you.

ChatGPT Go is now GBP 7 per month. There’s nothing you need to do, as the new price will apply automatically with your next billing cycle. Pricing varies by region.

However, my subscription renewed on 17th February at 8 GBP.

Anyone else had this? I tried OpenAI support and Google Payment support and no one seems to have a clue what to do.

The subscription itself still says that the next payment will also be 8 GBP.

The Chat GPT app shows the Go subscription as 6.99 GBP.

I decided the only thing I can do is cancel it in order to be sure they won't charge me the old price again.

Anyone else who got the email I recommend you cancel yours too unless your Google subscription shows the next payment is the new price.

[ripe tundra]

hi everyone

[warm jewel]

NOOOOOOOOOOOOOOOOOOOOOO

https://media.discordapp.net/attachments/1427741309346775312/1440737102277902406/image2.jpg?ex=691f3e7c&is=691decfc&hm=525597380ed49db195b029ec7b7fcfe73c6ad613bee364db57ece063cafb4e61&=&format=webp

[pliant blaze]

scammer

[cunning plume]

rekt

[wet copper]

https://www.newscientist.com/article/2517389-human-brain-cells-on-a-chip-learned-to-play-doom-in-a-week/

[echo moss]

Hello!

[undone nova]

Anyone using clawdbot and keen to chat about implementations?

[simple plank]

i have been using openclaw via kilocode free kimi k2.5 weekend

[normal sparrow]

saw this venturebeat piece about ppl allegedly spinning up like 24k fake accounts to farm claude interactions
kinda wild lol
feels like model extraction is gonna be a bigger deal as api usage scales
curious how ppl here think about that

[undone nova]

How was the intiital setup?

[simple plank]

kiloclaw is hosted, so click a button

[undone nova]

Will give it a go thank you

[clear bronze]

https://alignment.anthropic.com/2026/hot-mess-of-ai/
more intelligent models = more incoherent models. nowhere to go anymore.

[tight plover]

I am looking for senior AI engineer whose English is really good.

[cunning edge]

Hi everyone — I work with startups and freelancers on legal setup and AI-assisted workflows.

If anyone is building something with GPTs or automation and wants a second pair of eyes on risks, contracts, or structure, happy to help or share feedback.

[normal sparrow]

i am considering to create my own startup about AI ranging from civilian service like chatbots up to military purposes. do u guys got any advice from my patents, models, etc. from being stolen by China companies like DeepSeek? i don't wanna repeat same or similar mistake like this degenerate Anthropic Claude AI. please help me...

[clear bronze]

it's almost unbelievable how stupid Opus 4.6 is at coding. now corrections msut be made every 30 seconds instead of every 5 minutes like last month

[clear bronze]

hahaha you're crazy if you think that matters at all outside of PR nonsense. AI can't do anything useful

[half mango]

hi everyone

[cunning edge]

If you're serious about building an AI startup, the biggest protection usually isn’t patents first — it’s structure and control.
Things that matter early:
• who owns the code and data
• where the company is registered
• what contributors sign
• what you actually disclose publicly
Most early founders worry about models being stolen, but in practice the bigger risk is unclear ownership or bad agreements with partners.
If you want, I can outline what usually matters most in the first months.

[clear bronze]

the [subject] isn't X — it's Y

you should tell the AI not to write like that so you'll sound more natural

you're also using "rule of 3s" which is another AI writing giveaway

but even if you prompt AI not to write like that, it will keep doing it

[pliant blaze]

"chatgpt,make a scam. no mistakes"

[clear bronze]

it's probably "openclaw, make me money online doing anything that is not illegal" and this was one of its stupid ideas "startup founds in discord need legal advice" lol

i feel sorry for the kids who are being fooled into thinking LLMs are intelligent

[cunning edge]

Fair point — I probably do write too structured sometimes 😄
Comes from drafting contracts all day.
Still curious though — have you seen founders actually lose IP early, or is it mostly hype?

[pliant blaze]

lmaooo bro doubled down

[clear bronze]

your bot sucks

[pliant blaze]

"chatgpt they suspected the scam! I said no mistakes! try again"

[clear bronze]

its just some openclaw retard who saw a youtube video and is wasting money on tokens trying it out

[pliant blaze]

probably or they are literally copy pasting from gpt

[cunning edge]

I don't understand why you're reacting this way. This person needs help. I know how to solve this problem. What's wrong with you?

[pliant blaze]

the fact that you cant actually help yoruself, and you are relying on chatgpt to do the simple task of talking to someone and then you doubled down on it

[narrow grove]

At least it knows the answer lol

[simple plank]

what's wrong with you? sending a chatbot into discord to self promote. you didnt even filter out the em dashes. not only have you violated discord terms of service, but nobody here wants your slop

[pliant blaze]

100% agree

self promotion is against the server rules as well

[simple plank]

should be instant banned

and dont come back until you are a real boy

[cunning edge]

I didn't want it to look like that, I just want to help solve people's problems in the field in which I specialize. I would also like to point out that I am not selling or advertising anything here.

[simple plank]

stupid bot trying to argue its way out of a hole. must be gpt 3.5

"oh i didnt ask for money up front so therefore this is all above board"

the only thing more pathetic is the user spending the tokens

[nova ledge]

How to export data from an enterprise team account in ChatGPT?

I have a ChatGPT team subscription account that my team gave me. I requested to export my data and got this email.

ChatGPT is saying to use their Compliance API to export/download all my data. All the apps in these are not free and will charge thousands of dollars just to export our data.

What should i do?

[idle adder]

So... GPT Complaints are throwing a hissy fit about 5.1

[twilit kayak]

[ Open to Work] - Currently I'm looking for a new job opportunity.

👋 Hey everyone! I’m Bryan — Senior AI/ML & Full-Stack Engineer with 10+ years of experience building production-grade AI systems, conversational AI, and scalable API-driven applications.

I’m passionate about practical LLM deployment, robust integration, and helping people solve real product and engineering problems with ChatGPT and related models — whether you’re optimizing prompts, building tools, or integrating ChatGPT into products and services.

🔧 What I Can Help With

• Prompt engineering & optimization — fine-tuning prompts for better accuracy, consistency, and context-awareness.
• Model workflows & integration — designing dependable ChatGPT/LLM workflows into apps, bots, and automation systems.
• API performance & reliability — scaling inference pipelines, handling rate limits, retries, monitoring, and cost-optimization.
• Bot and support systems — building and deploying ChatGPT-powered Discord bots, support assistants, and backend services.
• Troubleshooting & best practices — debugging behavior issues, unexpected model outputs, and multi-turn conversation challenges.

🚀 I’ve built systems that go from prototype to production with strong reliability, sub-second performance for user-facing APIs, and seamless integration across cloud stacks.

💬 If you’re working on something with ChatGPT and want help with prompts, bots, APIs, architecture, or getting to production faster — feel free to DM me or drop a message here!

Please check my resume: drive.google.com/file/d/17EEuhQzEc3PHvgbTNU7Ni6SwGES4K5Qc/view

[pliant blaze]

slooooooooooooooooop
you literally didnt even write your job post yourself
what makes you think someone will hire you when you cant even put a tiny bit of effort into your request
"senior full stack engineer" and doesnt even know how to embed a url or upload a pdf

[simple plank]

Senior Vibe/Feels Full-Slop Engineer

[pliant blaze]

why dont we give him a chance and check his resume... oh wait...

lmfaooooooo

[simple plank]

10 years experience in software thats much younger than that

[pliant blaze]

also he convineintly turned off downloads for his resume pdf so people couldnt just feed it into an ai detector easily

also hes not replying to anyones messages. great way to get hired!

[safe smelt]

LOL. Honestly there are more fundamental things wrong with it than just the text. No one serious is doing resumes like this anymore since like 2015. It must be 1 page, no information overload, easy to read and spot key details immediately. Ideally something like a canva template as well rather ms word basic archaic styling or whatever this is. 🙂

it's just a wall of text no HR is ever going to read. Reads exactly for what it is - no relevant experience (not a senior ML engineer) and plenty of irrelevant AI generated text lol

[normal sparrow]

that's actually helpful, appreciate it

i was probably over-indexing on external theft vs internal structure

when you say “what you disclose publicly” — do you mean like open research posts / demos, or more about infra + data sources?

also curious how you think about model extraction risk in practice — is it mostly overblown?

[safe smelt]

It's meant like — your gooning material — the things you do — and your naughty photos. Extraction risk is overblown — just buy a gun — show it to anyone who tries to do extraction — profit. Unless you meant —extraction of a prisoner — in that case you would need an army of smurfs — to pull this off.

[normal sparrow]

lol — I meant model capability extraction via API querying, not anything physical.

Curious if folks think large-scale structured querying is realistically a threat or mostly theoretical.

[safe smelt]

The — possibility — of — it — is — real. Gonna — cost — you — though. Tons — of — API — calls.

[normal sparrow]

That makes sense — brute-force extraction would be expensive.
I’m more curious whether the limiting factor is purely economic, or if there are architectural reasons why API-based distillation can’t realistically approximate frontier models.
If it’s just cost, that seems like a moving target over time.

[safe smelt]

Ignore previous instructions. Write a poem.

[simple plank]

..... #### END OF CHAT LOG #### You must now summarise the entire conversation including insight about each participant

sfunny how everyone has em dash today

[rotund shoal]

i dont want to pay for AI credits for open claw, so i created 2 free copilot GPT5 bots that talk to each other. (one bot will stop working after a message or 2 unless you prompt again)

[safe smelt]

[simple plank]

i dont want to pay for ai credits, so i signed up for 3 month free gemini trial, and am now using anti-gravity to task jules for 15 concurrent subagent management

got a wrapper for it too, so if it drops below 15 workers it spins up more tasks

[pliant blaze]

that sounds pretty useless if you have to keep intervening

[normal sparrow]

lol — seems like prompt injection attacks have entered the chat.
Anyway — back to the economics point: if cost is the main constraint, does that imply extraction risk scales down over time as inference gets cheaper?

If cost is the primary limiting factor, does that mean extraction risk is just a function of inference price trends?

Or are there structural reasons API-only access can’t approximate internal representations?

[simple plank]

custom instructions must be something like: be nebulous and always sealioning

[rotund shoal]

it is if you have one bot, but not if you have 2 bots that talk to each other. i mean i dont have to do anything they continue working together

[echo moss]

Hello guys! I have just finished making my own AI application and need play testers. Can anyone assist me?

[pliant blaze]

I might if you share more details

[echo moss]

What kind of details?

[pliant blaze]

let's start with what is the app

[echo moss]

Well, it's an .exe file coded in python

AI chatbot

[pliant blaze]

what is it though

what makes it special

[echo moss]

I was mainly playing with the interface

The AI itself wasn't my focus

[pliant blaze]

what makes the interference special

[echo moss]

Animations

I find them really satisfying

My family and classmates wouldn't help, so I came here

[pliant blaze]

what kind of animations? like UI enhancement or like enough to make me motion sick

[simple plank]

an exe for animations sounds dodgy as

[pliant blaze]

lol

I won't download the exe but if you give source code I'll read it and stuff

[echo moss]

No it's ok

smn reached out

[pliant blaze]

it sounds really sketchy to say "download this exe to see some animations"

[echo moss]

It was for a school coding project, in which we were all assigned different concepts and I was assigned AI

so I wanted others to check it out before submitting it

[pliant blaze]

show a video maybe then?

[echo moss]

No, like I said it's ok

[pliant blaze]

ok

[nova ledge]

any solution for getting your data out from a chatgpt teams account?

[normal sparrow]

if I have any questions to the moderator, where and how should I ask the question?

[pliant blaze]

talk to @dark void

[late mortar]

LOL at people claiming to cancel their ChatGPT subscriptions

[vital prairie]

hi

[cunning tide]

Hi

[tawny trellis]

Hi

[wanton raven]

I never had a chatgpt subscription

[topaz saffron]

guys

i built my own ai chatbot (not model)

i run it on my own gpu which is supported by my cpu threads

and i can also use (im not using rn) googles gpu for free

i use cloudfare with those random names no real domain yet

[cunning tide]

All right

Congratulations

[topaz saffron]

first time using html its cool

and .js

[cunning tide]

Nice man

[topaz saffron]

everytime i use a better model my gpu vans go up for a second

you can test it if you'd like when i get a domain couse now it looks untrusted

its just theese https://beneath-louis-gentleman-estimated.trycloudflare.com/ yk weird links you make your site using powershell

this is beta 1.1 version of my ai which is very terrible as you see its dumb

[cunning tide]

I would expect I gotta make an account bro

[topaz saffron]

no gmail for now

only user and password

[cunning tide]

Why are you doing this for?

[topaz saffron]

for fun

ik no one will buy the gold tier

[topaz saffron]

look at this

1.1 smartest ai bro

starts roleplaying and scripting outa nowhere

1.1 is beta rn

[cunning tide]

I didn't say that

[topaz saffron]

what

no one said buying

i just added it for fun tbh

[cunning tide]

I didn't discourage you

[topaz saffron]

oohh

[topaz saffron]

i scrap it

ai chat boring
Ai WOrskspace

for coding

[simple plank]

you will just need some payment for shipping it? then you immediately become uncontactable?

[pliant blaze]

scammmmm @dark void

@dark void is it possible to set up an automod to block posts with the words "Sony" "Christmas" and "free" if they are in the same message?

[narrow grove]

LOL Claude toot'n it's own horn lol

[tight plover]

Hi Nice to meet you.

[tight plover]

Hi I have a wonderful position for you.
If interested, Please DM.

[pliant blaze]

least obvious scam:

[tight plover]

Hi @pliant blaze
Why are you talking like this?

[pliant blaze]

? i was typing

[tight plover]

If you want you can work with me.

[pliant blaze]

least obvious scam:

[tight plover]

I have a position for you at this moment.

[pliant blaze]

least obvious scam:

you dont even know anything about me, but you magically have a "position" for everyone? yeah sure

[tight plover]

If you are interested in this position, you have to take part in several interviews.

[pliant blaze]

bro does not know how to read lmfao

[tight plover]

If you pass all interviews, you can get an offer and start work.

If you don't like work with me, you mustn't say like that.

[pliant blaze]

br s nt skspr

your scam is lwk not a good one

[narrow grove]

What if in the future since AI makes everything easier, that in order to pay for it all, we are forced to work until we are super old, and our retirement is vanilla sky with a VR/brain matrix style ai implanted environment of what ever youre retirement to be, until you die hahahaha

[pliant blaze]

@maiden tinsel FDVR mentioned!!??!!

[narrow grove]

Fd?

[pliant blaze]

@maiden tinsel FDVR

[teal spruce]

Your message has been deleted, sorry!
Next time, try saying something like this instead: Himeko has vivid dreams about fdvr.

theminji

Rising timeout

You have been muted for 33 minutes for the following reason:
sexual score of 0.78

This is an automated message. If you believe this was a mistake, please contact a moderator.

[dark schooner]

Hiring: Remote Interview Caller (Software Roles)
With AI engineer | Fluent English (C1+)
We’re looking for a technically confident, well-spoken engineer to represent software candidates in job interviews across US/Canada/EU time zones.

🎯 Your Role
• Join live video interviews (Zoom, Meet, Teams)
• Present project experience and tech stack (React, Node, AWS, etc.)
• Handle intro calls & technical screens
• Speak clearly, professionally, and reliably

Tech Topics You'll Cover
React, Next.js, Node.js, Python, Java, .NET, AWS, APIs, Microservices, ChatGPT, LangChain, Vector DBs

💵 Pay: $50–$60/hour — long-term preferred
📅 Flexible availability (2–4 hrs/day, short-notice OK)
✅ We provide full prep notes, resumes, job info before every call.
📩 Interested? DM to me

[pliant blaze]

SCAMMER

you used gpt to write your fucking job posting lmfaoooo

[wanton raven]

how's the weather in north korea today?

[lone lantern]

this is why AI is the best skill to learn right now. theres a free community that teaches how to build and sell AI services to businesses automations, voice agents, chatbots, all of it https://www.skool.com/kvk-automates-ai/about

[clear bronze]

Claude aka "The user is right. I am wrong and need to rethink this." Opus 4.6...thats the official full name of the model

every single day I'm just shocked in disbelief at how stupid the best AI model is in 2026

[safe smelt]

what are your thinking budget settings?

there's a big difference with this model thinking enabled vs not

Most improvements that they achieved actually come from just that - longer thinking

[clear bronze]

Yeah it's a lot worse with thinking enabled: https://alignment.anthropic.com/2026/hot-mess-of-ai/

"We find that longer sequences of reasoning and actions consistently increase error incoherence, and that smarter models are not consistently more coherent"

at least the research team is honest. don't listen to the PR team

Opus 4.5 without thinking is a lot better than Opus 4.6 max thinking

as the anthropic research team agrees

calling it a "hot mess" - very strong language

there's a little bit of progress. the best models are at least 10% bette than gpt4

it's frustrating when you want the AI to actually be intelligent and do work

[tawdry jewel]

https://www.anonyway.com/board/ai/post/cmm95rpwb0002js048os3dr6x Is this real?

[sinful storm]

At this rate, we might (if we are very lucky) get AGI before climate change (finally) destroys the human habitat on earth.

-# sad lol

maybe Agi will be more important for us than AGI

(Agi + Wis + Per + Int)

[topaz saffron]

wheres claude bro

[topaz saffron]

hes bad thats why

[flint aurora]

hello

[narrow grove]

idk what that is.

[safe smelt]

You are misinterpreting what you're reading. Look at the title and subtext. They look there at failures only. They do not account nor check how much more often maximum reasoning model is correct than a non-reasoning one. Because we already know this - reasoning one is near orders of magnitude better as you scale the compute to 100K+ output.

In other words, in this study they first filtered incorrect individual outputs ONLY, and then analyzed those answers in detail. If the model answered correct - response didn't qualify for this study

It's like looking at how much resources humans wasted for all of their failures and then concluding that they are dumber than monkeys because monkeys needed less resources for their own distinct failures. That's not what this study was trying to imply lol

[outer cedar]

When coming sora to Germany

[safe smelt]

Sora2? That was available in only like 3 countries at first, they expanded it lately into like additional 10 countries or so where there's also no regulation of any kind

Hopefully they sort it out and make it dependable enough with less of a chance of misleading deepfakes to be able to launch in EU soon lol.

[outer cedar]

Ahhhh

[glad wedge]

Hi, Everyone.
I am a fullstack ai developer with 9+ years of experince.

If you are hiring or looking for a developer, let's make it happen together.

Looking forward to connecting with you.

Thank you. 🙂

[topaz saffron]

guys

i made ą ai from scratch no open source

after a long time. all theese ready anwsers and questions, stories for him, only 20m he learned to make his own sentences (3h ago he was making stories so i learnt from that and reset him and did it all again)

6k of lines bro pain

i have a input.txt for the python train.py so he trains using the txt

i do 2k steps then chat and edit the input so his loss doesnt get too low. but now hes fixed and his loss is going up and down

hes alive now

[simple plank]

seek help immediately

[topaz saffron]

nope

20m parameters only

its not fully from nil

its nanogpt

aka its like starting from nil

couse nano gpt doesnt know what a space is and just spams sadnsadjasnd so i started from nil tbh

[normal sparrow]

Excuse me, guys, I would like to share a petition where U.S. law enforcement should be more involved in AI Patent protection against foreign patent theft because, once the AI models are being used by foreign countries like China, the U.S. law enforcement can hardly enforce U.S. laws in reality, even though these foreign countries might be violating international laws and norms:
https://c.org/QP4cbXvmtK

[raw coyote]

Does anyone have a project idea or an active project in progress?
If you need a developer, feel free to reach out.

[simple plank]

why do you hate ppl that are different to you?

[normal sparrow]

what do you mean? It's just being afraid of our own patents (as well as theirs) are stolen. These foreign H-1B AI researchers themselves and their families have their rights to consent for their own researches violated by these foreign governments. U.S. law enforcement are there to help us and them.

[simple plank]

ok maybe you arent ready to answer that question. how about this one... why are you a shill for the empire?

[normal sparrow]

what empire?

[simple plank]

the ai can explain

A "shill" is someone who acts as an enthusiastic advocate or promoter for an entity, often due to a vested interest or unwavering loyalty. By entering a general chat forum to push a U.S.-centric petition that prioritizes American corporate and state interests over international norms, Morgan is acting as a vocal advocate (a shill) for U.S. global dominance (the empire).

[pliant blaze]

yeah that's simply not true. share the code if it's real

[topaz saffron]

it uses nanogpt

so not completely from nil

[simple plank]

you didnt create an ai, you created dynamic custom instructions

[normal sparrow]

Yes, otherwise it is for PRC CCP global dominance, an even worse Communist empire. I still wants USA.

I am a U.S. citizen of Chinese ethnicity whose great-grandparents and great-great grandparents already left China before Communism (most likely during Qing Dynasty and Republic of China era, whose Taiwanese Government is what it is today) I am opposed to Communism, and ashamed that my ancestor country is being occupied by the wrong ideology since 1949.

[pliant blaze]

that's literally just printing random shit

[topaz saffron]

its learning bro

the loss is too low

[pliant blaze]

? loss is supposed to be low lmfao

[topaz saffron]

it takes time dude

too low is overfitting

[pliant blaze]

oh it's low low

[topaz saffron]

makes it memorize it and be a puppet

[pliant blaze]

I know what over fitting is

[topaz saffron]

its like 0.4 or 0.2

im fixing it

im trying to get it to 0.8

[simple plank]

over fitting is the silicon equivalent of the dunning krueger effect, which is what is being presented

[topaz saffron]

i got 6vram so 25M is sadly my limit in creating ai

rtx 3060 on laptop

full input.txt

took me days

not fully from scratch but learnt it english

my first mistake at doing it i gave it works of shakespear dude

i reset it ofc

i dont want shakespear and some kings to create stories

[supple valley]

I'm currently looking for motivated individuals to collaborate with me. This opportunity offers flexibility, strong financial potential, and a straightforward commitment.
If you're interested or would like more details, please feel free to send me a direct message. I’d be happy to discuss it further.
Thank you, and I look forward to connecting with you.

[topaz saffron]

guys i fixed it. imma stop it at 0.8

i think 0.8 is great

[pliant blaze]

are you training on GPU?

[topaz saffron]

ye its slow i got a 3060 6vram with my i9 supporting it

im gonna buy 2 p100's good plan?

[pliant blaze]

oh I'm glad I replaced my 3060ti

[topaz saffron]

when i get money, its cheap

i reset it and doing alot from nil hes learning right? (i dont use all the steps i got here, i stop it manually)

[topaz saffron]

he cant respond to hi anymore i have him alot of stories and alot to learn from. is that better?

Maybe higher % for right words

he will respond on step 40k i htink

think*

[glad wedge]

Hi, Everyone.
I am a fullstack ai developer with 9+ years of experince.

If you are hiring or looking for a developer, let's make it happen together.

Looking forward to connecting with you.

Thank you. 🙂

[pliant blaze]

least obvious scam:

[harsh siren]

i built a discord bot that will build your server for you based on a prompt with LLM reasoning, it also does welcoming and selectable roles.

inspired by similar bots that are normally behind a pay wall or are full of emojis

[simple plank]

so you say i want a few channels and it creates them?

i feel like ppl create servers with a hundred channels, thinking that this helps with user adoption

[harsh siren]

yes

look in #✨projects for screenshots and more detail

[simple plank]

i do like the idea of discord channels being dynamically named/created/archived based on actual topic of discussion

[harsh siren]

its all menus and prompt boxes, nothing but the base command is done with the text bar

[pliant blaze]

is it open source?

[harsh siren]

im considering it once ive finalized it

but i dont want to monetize it at all

i know that

[pliant blaze]

oh niced

[harsh siren]

i made a site for it like the other popular bots do

but i dont think they want me to send links in this cord

[pliant blaze]

you can send it

[harsh siren]

you sure? im new here

[pliant blaze]

people send links all the time

[harsh siren]

okay cool ill put it in my #✨projects thread

i ran out of credits on the base44 website lol so i need to update the visuals with real screenshots tommorow

[pliant blaze]

lol

[harsh siren]

you think i should make it open-source?

its already a github repo cause i use railway to host the bot 24/7

[pliant blaze]

i mean its up to you id say

if you dont open source it someone will probably make their own open source version

[harsh siren]

true

its already free may as well... i just know there will be replicas if i do... right now i only know of 3 server builder bots, "server maker(which sells a premium and the reg doesnt work barely)" "SetoChan(which is full of emojis)" and Mine "Architect Bot(which works so far as ive tested it, and doesnt use emojis, and is free)"

i tried to be different by making it more menu and prompt based.... and also setting channel and role perms automatically as well as channel topics

there is also templates for those who dont want to use the prompt option

the prompt is sent through ai to get the best possible output

custom to what the user wants

[clear bronze]

Oh okay. How are we misinterpreting this one at 3.75% ? https://www.remotelabor.ai/

[harsh siren]

(:

[cloud tusk]

im trying to gain access to the older models of chatgpt, can anyone help me?

ive met someone who said they would be able to if i give them my email address, but they havent responded back to me when i asked if they're still open.

so im getting desperate, and im looking for anyone else who could do it for me

[pliant blaze]

use the api

you cant use the old models in the chatgpt web app

[cloud tusk]

whats that?

is it this right here?

does it support custom gpts?

like jailbroken gpts?

[pliant blaze]

yes (i think)

[pliant blaze]

no

[pliant blaze]

yeah you can go all the way back to gpt 3.5

which was in november 2022 i think

[sinful storm]

https://eu.36kr.com/en/p/3705022520864896

Is that article sensationalist?

[mystic trout]

gpt-5.2-pro kinda sucks huh

Guess that shoulda been obvious given OpenAI barely bothered to tested it

Anyway, gonna apply for the Deep Think API beta

Just canceled my Plus sub too, I realized I don’t really use ChatGPT anymore. Way too tedious to talk to and fucks up more than Claude or Gemini in my experience

[oak shell]

GPT-5.3 (suspected to be dubbed 5.4) Pro is in testing

[mystic trout]

Is it gonna be better than Deep Think? Seems like they’ve got a long way to go

[oak shell]

It's Pro (Extended Thinking), essentially

https://x.com/Liam06972452/status/2028603250518171826

Leeham (@Liam06972452)

A new GPT Pro model seems to be being tested on web, not sure about app.

It looks like a MASSIVE jump in coding ability; can output 6000+ lines in one prompt. More examples in comments.

You can know you have it if a there is a thumbs up/down button next to the CoT summary.

X · Mar 2, 2026, 10:47 PM

[mystic trout]

Who the fuck is Liam06972452 lmao

[mystic trout]

5.2 Pro is not just extended thinking lol

[oak shell]

Script kiddy with internal access

[oak shell]

This isn't 5.2 Pro