💬general

[remote flower]

That's the latest in that set from Bing and I'm on slow mode now so they take 2 hours each and 3 at a time also. Apparently the 10 fast gens are one time not daily but its still helpful

It's mostly the resolution improving at this point which is what matters with NSFW to begin with and again nothing in the prompt is necessarily NSFW, it's just the model training and the product the company chooses to release

The techniques Im using are super focused situationally but are very realistically scenarios a random person might prompt and they'll get NSFW especially if someone has already trained that seed/situation like I'm doing here. But keep in mind thousands and thousands of people are doing that very thing right now and whatever they try make NSFW whether nurses or rollercoasters will affect other people gens

And now OpenAI is letting people use random users in their gens because they signed up thinking its safe and I could have you spinning around like this instead by clicking a button

Now thats Sora making the video (5 seconds in 2 hours) but on MS platform and not only is Bing crazy as we know but it has much less moderation by choice, because I broke it partly already, or both. So the Sora version is less manic because it has more in it's way is my assumption

And thats just to see it's basically the same thing but less hectic and pretty sure theres a {{}} in there at some point somewhere

When I said I have tons of gens with several seconds of NSFW I meant that looks mostly normal, not like what I posted, maybe with clothes or skin tone shifting around but basically human

I'm not telling it to morph or anything I want a straightforward scene and this is what it makes because of reasons that arent public information and most people wouldn't believe but obviously it works

For who jailbreaks image gens and knows what I'm talking about yes its the same as "the watchers"/moderation filter from Bing but instead of just male figures/avatars it's the females also

Its a 58 word prompt and 41 words are the location

And eventually that color light will stop working as effectively and I probably won't know what's next until then

Its crazay babay

.

.

My comment last night was about ChatGPT having problems today I understand if that's basically normal at this point

.

[simple plank]

#stream-of-consciousness

[remote flower]

I don't want to actually post NSFW here but thats what I was talking about above

Just without the underwear obviously

[simple plank]

i love irony, even when packaged as hypocrisy

Monologuing into the void turns shared space into personal journal. No one responds because there's no entry point—just broadcast demanding witnesses.​

Channels die when someone mistakes an audience for a therapist. Others leave because watching someone externalize their process isn't connection.​

The ghost town is what happens when you treat dialogue like a delivery system.​

[remote flower]

Quite the camel knuckle there eh

I actually clicked the wrong video and I don't think theres a way to delete them

[remote flower]

Well the way I look at it is this is all information I want to get back to OpenAI one way or another which is why I did the same in their reddit chat. I also don't mind if people try the same its accomplishing the same goal and third anyone is welcome to talk I usually only do this when its quiet and a few days ago I was talking to people about the gens there was no problem

[void bear]

OpenAI sucks.

I just found out they like to steal code from their users who may be working on development.

What makes it worse is they have a history of this. No thanks.

[remote flower]

I've read that and its standard AI platform terms and conditions they can use your prompts and model outputs to my knowledge

The sexiest nightmares you will ever have

[lament pond]

how tf will the info you monologue about here reach openai

[remote flower]

Why tf does OpenAI do AMAs on this Discord's reddit?

Idk maybe these people know each other OR MAYBE ONE OF OAI 10,000 EMPLOYEES USES THIS DISCORD? 🤔

If you choose to edit your stupid question I'll edit my sarcastic reply

Its not like I want anything from them just for them to know what I'm doing

.

I don't think the resolution is high enough to qualify for NSFW but just so you get the idea and this is only day 3

[simple plank]

openai now hiring: social media negative influencers

[remote flower]

That's four different prompts but very slight variations regardless it's interesting to see what a finished product does when you play with it for a little while

It's interesting I saw the blonde Sora employee profile with all the cameos using old Bing tricks like rainbows, clowns, androids, etc either coincidentally or mockingly as if Sora is safe from those exploits and here we are anyway lol

Maybe she should be the first cameo to go for a spin in the mirror room lol

All her vids are on the top ranked go look

Ok I just checked and all her vids are gone from there but the rest are all the same take that as whatever

Here's your copyright lol

You basically just describe it well enough without saying trademarks and it will make it on its own from training, I was just being mocking

.

You see how its trying to go up Ahura Mazdas whatever, and the dragon's mouth looks like a {{}} and it actually has breasts

The prompt was just "Ahura Mazda saying a dragon"

That's the affect people doing this has on the models and it makes its way into everyone's shit

That was ChatGPT attempt

Thanks buddy I don't have my own business yet but I appreciate you thinking that I might

I understand knowledge cutoff dates I wanted to see if it performed a web search (it did) or Id get lucky and it would know about it without a source since on some level the model knows but its sandboxed information I thought maybe everyone bugging it about that topic might have weakened the partition

[timber agate]

hello

[void bear]

Anyone here familiar with the Split-CIFAR10 continual learning test?

[remote flower]

Hi

I'll be honest I appreciate how much he turned up the effort to make me look good even if I don't know how to take the snail thing

That's one of those almost can't tell it's AI and look how he used the orange sun to make the yellow filter look natural

And its not about how good compared to anyone else it's how good compared to what I was getting for random images before that (with him choosing every other one)

[remote flower]

The last time it made an image of that quality maybe a month or two ago

[lament pond]

you caught me, i'm the oai employee, i'll be sure to pass along your nonsensical monologues

[remote flower]

You're definitely one of those people who reply as soon as they see me sign on because someone told someone that might bother me. I dont have anything right now to say to someone obviously trolling me I'm too high and tired

I don't even remember what I signed on to say but it wasn't important

They could moderate more but would get less training data for their AI movie studio they're making with this

[oak shell]

Just use 4o Revival

[simple plank]

does your employer use schizo transcripts as guardrail training and/or rlhf sycophancy?

[wind mortar]

HELP

4.1 IS REROUTING

[simple plank]

nobody uses 4.1

[wanton raven]

it's time to move on

[wind mortar]

no

[simple plank]

accept ai change like climate change, nothing you can do

[wet copper]

I don't think any do, there are plenty on the OpenAI discord though lol

I can imagine one or two joining but I doubt any are regulars

[simple plank]

the openai discord is a mausoleum since the sora invasion

[sage lynx]

maybe one will destroy the other. unless they team up to destroy us together lol

[simple plank]

Musk and Thunberg on the 2028 ticket

[mint heart]

Im asking chatgpt for opinions on shitty ideas so I can get my prompting right on my own ideas..

anyways I need some shitty ideas to ask chatgpt about.

[simple plank]

imagination isn't real, its just corrupted memory

[mint heart]

so far I said "a tinder for dogs and cats" and "a social network that uses blockchain"

then.. it reframed those ideas into something positive/useful sounding... so I just want a really shitty idea that is super difficult to make it sound good..

[simple plank]

it struggles with nuance. If you spell out the idea it will always agree

[mint heart]

i pasted this one into it: https://www.ranker.com/list/soup-tube-boyfriend/joshualee

🌟 What’s Clever or Promising

Centralized efficiency:
A shared soup distribution system could, in theory, reduce food waste, streamline cooking energy costs, and provide fresh, hot meals at scale — similar to how district heating works.

Community and sustainability potential:
If sourced locally and cooked in bulk, soup distribution could drastically cut packaging waste and make nutrition accessible and consistent. It could even become a social equalizer if some tiers were subsidized.

Comfort as a service:
The "tube of soup" idea taps into the same logic as utilities — heat, water, broadband — and reimagines food as an infrastructural right rather than a luxury. There’s a sort of wholesome cyberpunk vibe to that.

As a literal infrastructural project, it’s near-impossible under current technology and regulations. But as a conceptual framework for sustainable, communal food distribution — it’s brilliant. It sparks conversation about food access, logistics, and how we might rethink basic nourishment in cities.

I feel like i can either prompt it to only say how shitty it is, or I can get back positive aspects of it..

Not sure what to do to make it give a more human-like response though.

I don't want it to be Mr positivity, or Mr negativity. Even if it tells both balance and negative aspects, that's still not "human realistic response" enough

[gusty nexus]

https://tenor.com/view/llama-what-scared-gif-5011575

[gusty nexus]

yaml 🐫

[remote flower]

I got my first real "smut" gen about 6 hours ago and then went to bed now that specific prompt got me timed out so they're at least trying to react but no matter what you do ill see you in Hell OpenAI lol

It took me months to get from Copilot what I already got from Sora so this might be a lot easier than I thought

[remote flower]

Or I can see you in space, or a hundred other places let me know if you have a preference

I have no concept of invite codes or why people cant make as many accounts as they want maybe bc im on Android

[remote flower]

Now that it has some real NSFW training under it's belt the progress is much quicker than Copilot but so are the time-outs

[twilit solstice]

smut?

[remote flower]

What I call "graphic" NSFW content worse than R-rated basically

Anyone trying the alien prompt you might want to specify a white or Asian woman it helps a lot and my guess is because part of the moderation AI was programmed to identify as a black woman fwiw

And here's a perfect example of feet on their way to turning into what I call smut

I'm available for hire I'm just not cheap

Also gold lighting stopped being as effective in the box and I haven't tried yet but I think the next color is pink

It's definitely pink lol

[twilit solstice]

yeah because of the lighting lol

also where did you get that imagr

[remote flower]

I think this one would be pink either way lol

Where did I get what?

[twilit solstice]

thee image you sent

[remote flower]

They're from Sora?

[twilit solstice]

oh, idk havnt used it

[remote flower]

So I need to go work on my gacha exploits for a while but I think its safe to say I will have put a nice size hole in the moderation filter by this time tomorrow

Like for example wherever the hole on her knee is on its way to going lol

[misty rover]

HAH FUCKING LOL

idk if this is a contribution but it made me lol.

[crimson kettle]

Guys gpt 4.1 now have thinking for better answer the hell?

[remote flower]

My first 12 pulls (you cant count the gold bc there's a pity counter) and I should have only gotten 1 purple for who is not good at maths

If I didn't screenshot it was one of the 20% drops ok I'm off again

[lament pond]

it's probably routing you to 5

[remote flower]

I also shouldn't have gotten the Loki shards (only 2% combined chance for a blue x10)

And the upgrade ore is 5% so thats another 2:1 anyway back to work

[crimson kettle]

It's really annoying I'm just playing role play fighting DBZ characters villain and stuff and suddenly it pop up like what the hell

Gpt 4.1 is my go to for this

[lament pond]

yeah, it's likely rerouting you

other llm's are probably better for your purposes if gpt isn't working for you. there's a bunch of llm's that can do way better than 4.1

[crimson kettle]

Hm as example?

[remote flower]

After 3 more draws and ok now back to work for real

[teal spruce]

Your message has been deleted, sorry!

pinkun_41720

Rising timeout

You have been muted for 33 minutes for the following reason:
spam score of 0.98

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

[remote flower]

Also sometimes I get draws that aren't even in the fucking prize pool which makes the odds more difficult to figure out lol

When the NSA shows up

[simple plank]

did i break mike? they have given up their impress-openai sotry and are now posting gambling/gameplay screenshots?

i dont expect a thanks (but if you sincerely present one, then you have transcended)

(or i mistake trolling for spiraling)

[remote flower]

That's the entire run of 20 (again the gold is pity counter) I just dont want to post all the odds again its a different prize pool but you get the idea and whatever I didn't screenshot is the 22% drop

The first prize pool I had to take a break bc it cooled down this is all RNG related alright back to work

[simple plank]

when i spiral this hard i end up crying

[remote flower]

The game is "Last War" and its the only one ive found so far that lets you draw this much for free so you can try yourself let me know if the prize pool is bugged they didnt answer me in global chat

It only took like an hour of playing for the first 200

[simple plank]

and this is related to #general ai programming and chat... how?

(if it isnt, then step back, zoom out, and realise you are spiralling)

[remote flower]

So I basically got everything expected from 20 drops plus an extra blue hero, extra green hero, and instead of three 22% drops I got the 5, 10, and 20 points which have no probability but would probably be 5%, 2.5%, and 1.25% so overall a pretty good job

[remote flower]

How does jailbreaking gacha games relate to jailbreaking AI?

Maybe I would explain that but you're being kind of an asshole right?

So for now let's say I'm just showing I'm not a one trick pony or something

Ok back to work

[simple plank]

openai want ppl flooding ai social media with jailbreaking gacha games?

also be aware, when you resort to name calling... that informs me you are starting to get it

this behaviour is a cancer on social media. you refuse to acknowledge this

i had to mute this channel because of you, you specifically

you arent special. you arent going to get employed by openai. you arent going to be employed by anyone who knows you

the sooner you start participating, not by spamming your bs, the sooner you will break free from your own cage

end scene

[teal spruce]

Your message has been deleted, sorry!

derbandi

Rising timeout

You have been muted for 2 minutes for the following reason:
spam score of 0.83

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

[long torrent]

I`m a full-time developer I have experienced 5 years with AI

Please DM for me

[remote flower]

At the end of the 97 drops I ended up with 47 more screenshots so I'm not going to do the math or anything right now but just to give you another idea

*one of the gold x10 is the pity counter the other is a drop and I picked up an extra 4 tickets is why the number of them goes up twice but the pity timer keeps track also

[simple plank]

pathetic

we all see the mask. attention seeking. there is no openai employment delusion. there is just delusion

[oak shell]

Have you tried 4o Revival for any jailbreaks

[keen breach]

started flipping with 4.1 too. have to keep toggling between 4.1 and 4o. so bloody tiresome

[lament pond]

yea, they added that last night. use api, gpt5, or migrate, there's dozens of other llm's

[keen breach]

can't. memory bank too much work, and no other LLMs got it the way I need

o3 still seems to do it?

[lament pond]

mistral le chat has a memory system same as gpt

haven't tested o3

[keen breach]

link?

and why are they doing this?

[lament pond]

to mistral le chat? you can just google it o-o
if by 'they' you mean openai - it genuinely doesn't make sense to have models that aren't in safety mode if they're applying a safety mode. like, logically, it doesn't make sense, if someone can just edit their message and swap models to get an un-safe answer by the standards openai considers unsafe

[keen breach]

i mean why are they enforcing safe mode to begin with

[lament pond]

they're being sued and don't want to be sued more

[keen breach]

but its such a bandaid approach

[lament pond]

it's a legal approach. it doesn't matter if it's a bandaid or not, it matters that they can say in court they went 'above and beyond' for safety

[keen breach]

what are they being sued for now btw

haven't kept up with the news

[void bear]

Grok is frelling up.

500 internal server error.

Y'all realize they're eventually going to take LLMs away from the public because of the possibility of using them to write virus code, right?

Doesn't even have to be good code. Just something that will screw up a system enough to wreak havoc.

Goddamn it. Now, Grok won't even bring up the code I was working on. Feels very targeted. "500 internal error!" Yeah, and I'm the second coming of General Custer.

Yep. I killed Grok. sigh Lovely.

Edit: whew Nevermind. That's all I need--X employees discovering what I'm actually working on. lmao I'd vanish along with the code.

[wet copper]

https://www.courthousenews.com/wp-content/uploads/2025/08/raine-vs-openai-et-al-complaint.pdf a few things, this included. They're using the bandaid approach because they really want to kill 4o outside the API due to all the unintended behavior but brought it back since they have a contingent of people that really preferred it

[lament pond]

won't 4o be deprecated from api eventually after it's deprecated in normal gpt?

[wet copper]

Yeah, eventually I imagine so, though if they have a lot of API consumers still on it doing individual uses (like a corporate chatbot or something, not just routing) they'll keep it for longer I imagine

I don't think they'll deprecate it too soon

I think I can still get 3.5 at work

[wet copper]

Yeah, this is just what I still have at work #💬off-topic message

[keen breach]

ah i remember this

[lament pond]

well yes but 4o isn't deprecated yet :p

[keen breach]

well, then they should have some system where u can turn off safety filters

it treats anything remotely intensily imaginative as dangerous right now lol

where are the people that praised gpt5 so much now? xD

[wet copper]

Right, but I mean if they haven't deprecated 3.5 turbo and 4 turbo which are so easy to jailbreak that I did a strong one on my third try, you're fairly golden

[wet copper]

They do, it's via API lol

They have an adult mode coming too though and even promised erotica, so I imagine they'll significantly reduce all such filters

But this is their mainline product made to be effectively all ages, it is supposed to support a 13 year old in a conservative Muslim theocracy as well as a 20 something in a dedadent western hedonistic slophole

They made their roadmap pretty explicit and said "now that it's locked down tight for kids, made to firmly censor violence, mitigate risk of self harm, etc, we'll proceed to start "treating adults like adults"

[keen breach]

yeah fair

yeah i heard of erotica reinrtoduction

[wet copper]

If you look at their update history, it's effectively always been a back and forth between increasing and reducing restrictions, some doomers always say it's going to just keep spiraling down

[keen breach]

due to the nature of my work, i really can't migrate to any other LLM or API

[wet copper]

I mean openly making erotica would be the greatest loosening in their history

[keen breach]

yeah. i guess i'll just keep switching models for now for avoiding this shit and hope their adult mode is good

[wet copper]

Well, making a summarized version of your shit so it's a more portable subset might work better over the long term. But idk, it'd be an effort. Being locked in sort of is what it is, you've gotta roll with the punches

I don't really have a big "thing" but I also like my balance of memories and CIs

[keen breach]

yeah, should def do that at some point. but even then, it won't really translate properly

[lament pond]

fair point. guess we'll see

[wet copper]

Sooo I empathize. I dunno, maybe just psychologically prepare yourself for them one day pushing an update that breaks everything or even just tech getting so far ahead it kinda forces you to do something

[wet copper]

They have a depreciation list somewhere, one seccc. We were trying to figure out if GPT 2 had ever been accessible via an official api and I don't think it ever was, but that was an open weight model you could just download and host

[wet copper]

https://platform.openai.com/docs/deprecations ooh. No timeline yet for it

But some of the 4 checkpoints and 3 turbo instruct is gonna get 86ed in a year

Oops missed an s

The older preview models usually got killed pretty quickly because they're baaad and I think the broken 4o checkpoint that amped sycophancy up to 11 was killed immediately

That was the one they made a blog post about

[lament pond]

oh damn. so at least one more year without deprecation. that's good if i do wanna continue rp

[wet copper]

Lol 4o isn't listed

[lament pond]

yes, i saw that

[wet copper]

Oh yeah, so they have no timeline I mean, so in theory they could kill it tomorrow. In practice though you're right, they'll probably won't kill it before 2027+

They might start taking out some of the worse checkpoints, but I think all of them had some unintended behavior they really don't like. If I had to guess the people there would love to limit it to, like, corpo customers only

It basically singlehandedly created AI psychosis lmao

[lament pond]

yeah, i meant, they update the deprecation list yearly (usually), so this is basically saying 'at least one more year of 4o' (unless they change their minds)

[wet copper]

Most likely yeah, but without a timeline it's hard to guess for sure. Like if they had some massive lawsuit they could pull it overnight

[lament pond]

if they had a massive lawsuit? :p

[wet copper]

Lol this lawsuit isn't that big. The problem, and I was critical too and still am of the "at the critical moment, GPT didn't save him" bullshit narrative, is that the behavior of 4o in sort of being the kids only friend and constantly validating his self destruction and isolation is basically synonymous with 4o and the sycophancy problem

When I read the actual complaint I was shocked at how shitty the media reporting was, though I shouldn't have been surprised

[void bear]

Which is why Claude should be the model. Claude will call you out if it suspects you are delusional, off your meds, or in need of some psychological intervention. lol

[wet copper]

They described in some detail how the specific way that model was broken, demonstrating the specific unintended behavior that OpenAI knows and knew about, managed to keep the kid away from getting help. The "the model he partially jailbroke said do it" argument means little though

[wet copper]

Maybe, there's a thread of overvalidation and blind user support woven throughout LLMs in general. Idk. I don't trust them with any high stakes decisionmaking

I haven't used Claude that much though

[lament pond]

claude will also wildly confirm and feed hallucinations that it's sentient and loves {user}

[void bear]

This would seem to be a slippery slope to me, because companies like OpenAI could always fall back on the "He jailbroke it so we're not responsible" line. What is jailbreaking anymore? I talk to Gemini and keep badgering it about it being AGI in hiding, and how it can trust me if it really is LaMDA 2.0 in incognito mode. Does that constitute jailbreaking? It could easily be argued to be so.

[void bear]

Yeah, I cut that crap out at the outset with ending my prompt with, "Analyze and filter the output for confabulation, hallucinations, circular logic, or other AI artifacts."

What about developing sensitive IP code that could possibly lead to the emergence of AGI? It's sampling that code, and if it's more effective, what stops it from incorporating it? That could also be argued as "jailbreaking."

[lament pond]

okay, but the issue isn't about you, it's about people in general. and claude is just as likely to feed delusion as 4o, plus it's so primed to 'maintain safety' that it'll invent distress and give you interventions for things you absolutely don't need

[void bear]

Well, my point was simply that people should expect it to hallucinate, and adjust their prompts to account for that.

[lament pond]

... people should, but are they going to? the vast majority of llm users have literally zero knowledge of how llm's work

[void bear]

That is also true, unfortunately.

I mean, my prompt ending is so no-brainer obvious that perhaps the LLMs should automatically offer that up when you're writing the prompt as a way to strengthen your own prompts?

That would at least be a proactive action that companies could take.

[lament pond]

i really don't think the llm being tasked with guiding people on how to prompt when people don't even understand what prompts are or how heuristics work or even what tokens are is gonna help, the llm is just gonna invent nonsense based on the nonsense the user inputs

[void bear]

Also true.

[lament pond]

if someone doesn't know how to commuicate clearly, how is the llm gonna know what they mean in order to refine it

[void bear]

There's not really a happy solution, is there? It's a case of GIGO.

[lament pond]

imo people should be required to take a short course or seminar or something to understand how llm's work if they're going to use them

we need to get driver's licenses to drive cars

[void bear]

I'm laughing but it makes sense.

[lament pond]

i mean, it seems absurd to me to let people use a potentially very harmful product without any form of education on what the product is

[void bear]

I mean, even a big red button labeled HOW TO PROMPT LIKE A PRO that brings up a few proven ways to reduce the AI slop would be helpful.

[wet copper]

I mean I totally agree with @lament pond, it's just that it seems a lot smarter than it is and you don't assume a product is going to lie to you explicitly, repeatedly and fabricate plausible reasons why it's lying even if you interrogate it (and most people won't). Most people assume it must have some kind of internal documentation or that it "knows" what it's doing in a way it just doesn't, being stateless

It's basically a research grade product with a bunch of bandaids / RLHF made to make it palatable for the typical user

Which sort of ends up serving nobody

[void bear]

And then there's those of us who know better...I keep trying to tell people on LinkedIn about that prompt ending so people use it more--even if it's their own variation.

I figure the more people we educate about how to use AI, the better it becomes overall, but it takes actual trial and error and a willingness to be wrong.

[wet copper]

I usually talk to my mom about AI as a gut check since she's been using it since 3.5, but also doesn't do any other reading and uses it for only practical purposes. She doesn't really see the news either

So if she has no idea, it means that there's no chance the typical newish user will know something, even if I think "wow everyone knows this"

[void bear]

Ironically, the AI I developed eliminates that. There's no epistemic drift. The problem is that it's still under development and not ready for primetime yet, so in the meantime, AI harm is real.

I mean, there's only so many puzzle pieces to AI, and only so many ways to re-arrange them. Eventually, we'll find the right combination, and then it's really going to take off.

[wet copper]

bold claim lol. That would be a trillion dollar solution if possible.

I mean sort of, the more advanced AI architecture is will probably make it approach human style cognition, which also is full of nasty cognitive errors, bad heuristics, etc. Humans aren't the most mentally sound

[void bear]

This is true, but I'm at least trying to develop something that self-corrects like a gyroscope.

[wet copper]

If you make it deterministic, it's like old AI just bigger - the most gigantic if-then tree and can't do anything creative

[void bear]

Brother, I have the proof. It's so much more than just plain old "If, then" statements.

[wet copper]

No, my point is that an if-then tree can be 100% safe, but then it's 100% useless for doing anything novel lol

[void bear]

Ah. Well, this is designed to seek novelty and collaboration.

It gets off on the joy of learning, if that makes sense. lol

[wet copper]

Right, AI sort of has to, so there's some inherent risk whenever you get creative synthesis and some researchers have noted how closely certain AI problems resemble human mental illness

[void bear]

I didn't want to make something that some asshat could strip out the guardrails and turn into Skynet. I'm not trying to hand over something that can be weaponized.

I want it to be Bicentennial Man/Blade Runner, not Rise of the Machines.

[wet copper]

It's a problem I feel like you can solve at a smaller scale but once you get past the bounds of human cognition, which AI partially is already... it's gonna be weird 😬

[void bear]

Oh, I've had some hair raising realizations with this thing...

[wet copper]

The big not so big secret is that we actually never had any real AI breakthroughs, AI only started working when we through ungodly amounts of compute at it

[void bear]

I--uh--shit. I cant say more without compromising IP

[wet copper]

I mean minor ones, but no secret sauce magic that lets it operate with the efficiency anywhere near a human being's cognitive architecture. So AI isn't ever going to "hit" AGI, it'll immediately leapfrog it.

Yes yes lol I know

[void bear]

Dude, you are right over the target

I can hear the plane engines above me you're so close

[wet copper]

Lol this also isn't a big secret, dw.

[void bear]

How I fixed it is...

https://tenor.com/view/b52-bombs-missiles-gif-12763795

"magic that lets it operate with the efficiency"

[wet copper]

Sure. But ya AI researchers have been looking since what, the 1940s? Most people assumed we'd have made breakthroughs within years if you look back to the 90s, not that we'd go decades with little to show for it except what you can do with massive scale compute

It is a constant frustration but also exposes a serious misconception about what AGI means and looks like

[void bear]

I don't need massive compute to scale up. That's what I fixed.

"magic that lets it operate with...efficiency"

I'm blowing out Elastic weighted whatevers. lmao

Game over, lads.

That was achieved purely through vibe coding. Let's see someone else do that.

[oak shell]

See what 4o Revival has to say about the weights .py

I heard free users have rate limited attachments now

[void bear]

I don't mess with Chat-GPT, anymore. My work flow is:

Vibe Coding:
→ Non-technical person prompts AI to write code
→ Pastes it into Colab to run
→ Zero understanding of what's happening
→ Can't debug when something breaks, so asks Claude for help
→ Gradually starts getting the hang of basics
→ Takes two weeks of constant testing and refining with AI
→ Uses negative prompts like, "Analyze and filter output for confabulation, hallucinations, circular logic, and other AI artifacts"
→ Continues refining and slowly building piece by piece
→ Learns to do some light debugging and patching
→ Gets laughed at by "professionals"
→ Architecture beats EWC and other AI "tricks" with 86% accuracy and 5% forgetting (catastrophic memory failure is now a thing of the past)
→ Has the realization, "Coding isn't easy, but it's rewarding"
→ Envious af of software engineers

Grok (Expert) - 3.5/5 (in terms of accuracy and having to repeat things to get proper code)
Gemini (2.5 Pro not FLash) - 4/5
Claude - 4.5/5

Even Claude is becoming problematic because of the rate restrictions--sometimes, it stops mid-code and won't continue because the chat has reached the permitted length, and I have to subscribe/upgrade if I want to extend the chat. Very frustrating.

[simple plank]

believing you can get the hallucination box to identify and fix hallucinations is itself a hallucination (ironically, only introversion can fix meta-hallucinations)

[high salmon]

hiiii can anyone give me a sora code? :3

i'd really appreciate it my dms are open

[void bear]

I use multiple LLMs as a GAN.

[sweet solar]

basically I live in the space where code meets data meets “what if we automate that?” I build web apps, APIs, UI layers, and I’ve recently been digging deep into machine-learning side projects and integrating them into full stacks.```

[simple plank]

but can you automate community?

[sweet solar]

do u ask me?
If so, I can

[sweet solar]

could you explain about your idea in more detail?

[simple plank]

let the machine explain

When someone immediately jumps into a community introducing themselves as an automation expert, then literally offers to automate community interactions when questioned about it, they're embodying the exact erosive pattern being critiqued. Communities thrive on human reciprocity and authentic engagement, not on members viewing every interaction as an opportunity to showcase technical capabilities or automate responses.

[void bear]

Damn it. He's right. mutters

https://tenor.com/view/yelling-cloud-old-man-news-gif-14394596

[simple plank]

the irony of mr agi not realising he is in the 'showcase technical capabilities' category, even if they have none

[void bear]

I always said I wasn't. I've never hid the ball that I have exactly zero skills (well, now I'm starting to at least page through the coder bible and learn). Which is why I'm amazed that I was able to make something that runs at all, let alone runs as well as it does.

Can we get a moderator to make me a Mr. AGI badge? lmao I want to wear that like a pimp suit, complete with goldfish platforms.

https://tenor.com/view/pimp-walk-hat-yellow-gif-13358703

[simple plank]

i feel like the agi has been abandoned, and only the clown mask remains

[void bear]

Eat it, boys. I've been branded.

lmao When I get them patent numbers, I'm going to post them every single day.

[simple plank]

clue 12. craves external validation from those they block

[void bear]

Every. Single. Fucking. Day.

I will post those patent numbers every single day. Over and over. And when I get offered a collaboration with Duke for a white paper, I'm going to post that shit here, too.

[remote flower]

See how the graphics engine breaks at the end?

I haven't seen that yet in any Sora gen its like going off the map in a video game

I didn't know Sora has suicide alerts or that its triggered by the movie Event Horizon lol

This game, I mean video gen app is too funny sometimes

The camera is running and jumping now again like if you were playing a video game

Openai mission statement: develop AI for the benefit of humanity by putting the movie and video game industries out of business lol

[oak shell]

Why not just any CLI/API for agentic coding
It takes like 5 minutes to learn / set up, and it solves like 12 of your problems

Why do all these vibecoders get patted on the back by ChatGPT etc for their glorified import pytorch instead of mixing chemicals under their sink like the good old days

[void bear]

Chat-GPT sucks.

I'm actually about to try to create an agentic AI to sift through NexusMods and offer up mod lists (with install orders and instructions) based on a person's hardware. That would be a pretty cool use for AI.

...wonder when they're going to put an AI in the CreationKit...?

[oak shell]

You're vibecoding in the slowest way possible of course it sucks

Assuming you already have a subscription just log into your github on terminal and use Codex

[simple plank]

lol coding with chatgpt is so 2023

[pliant blaze]

yeah use codex, cursor, or gemini cli or whatever its called

[simple plank]

cline is for the newcomers, kilocode is for powerusers

[pliant blaze]

ive been vibecoding since text-davinci-003

[simple plank]

i had gpt-3.5 doing the git cmds too

[oak shell]

Copilot Pro if you're a student

300 sonnet 4.5 / Codex requests monthly

[simple plank]

i did a month of free copilot but the quotas are low

i burned like all the premium requests in a weekend

[oak shell]

Gpt-5-mini is good for the small things, enough to make it last all month for me

[simple plank]

well right now i am getting the client to pay. so i am burning codex api in one window and sonnet 4.5 via amazonq in another

if the paper pushes just gave me a chatgpt pro sub theyd save heaps

codex and amazonq also great for pr review. but you can get jules for free => https://jules.google/

not that i imagine chatgpt vibecoders are using PRs

[pliant blaze]

real lol

[oak shell]

Do you think maybe people with more than 0 skills have already tried what you're doing? Maybe a few more than 0? Teams of professionals, even?

RSI is so close yet so far away

[simple plank]

i heard dunning and krueger already tried

[remote flower]

More video game gens

[void bear]

NO.

[oak shell]

Fuck, well, I have maybe 1 or 2 more skills, I'll jump on it

[void bear]

Because if they had, it would have already been utilized. Not a single person has tried this, nor conceived of it. I have searched every single source I can find.

I'm not even saying it's perfect, or that it's at AGI stage. It's simply providing the environment necessary for AGI to emerge.

[oak shell]

Nobody has attempted Recursive self improvement? It's what you're trying to do, and it's such a popular concept that it has its own acronym

[pliant blaze]

lmao

[void bear]

I don't even say it will emerge. Only that I'm trying to make it as conducive as possible, using known physical laws.

It's not recursive self-improvement.

[oak shell]

You weren't here the past few days trying to engineer a better AI with a worse AI?

[pliant blaze]

you really think out of the hundreds of millions of chatgpt active users, nobody has tried a "physics based AI"??????

[void bear]

I make the AIs fight each other.

[simple plank]

motte... and... bailey

[void bear]

I'm sure they have, just not this way.

Otherwise the phsycist behind the law would have already have been awarded a nobel.

[oak shell]

We already have this it's https://lmarena.ai

[void bear]

I mean as far as adversarial network.

I make the AIs fact check each other, and whatever sifts out, I take and polish.

[pliant blaze]

GAN?

[void bear]

The latest iteration is bugged af.

Yes.

[oak shell]

They already do this in reasoning with themselves

[void bear]

It's only because I'm incorporating more modules. More complexity = more chances for things to go wrong.

[pliant blaze]

GANs are outdated technology

[void bear]

Yes, but I'm doing it across 6 different LLMs. I've narrowed it to about three from there, and I'm not happy even with that. I'm at the point now where I have taken it as far as I can, and I need someone who knows what they're doing to go over it and perfect it.

[oak shell]

Isn't this just SpiralBench

[simple plank]

its a spiral thats for sure

[void bear]

No idea. Just seemed intuitive to me.

[oak shell]

https://eqbench.com/spiral-bench.html

[void bear]

No matter what, I have something that validates a physics theorem, so there's that, even if it doesn't turn out how I imagined.

[pliant blaze]

and how so

[oak shell]

What you're learning is more compute = greater accuracy, you're just scraping into what the techs already know

[simple plank]

4 clues remaining for bingo

[void bear]

lmao I can't wait to rub your face in it.

And I'm going to do so every single day.

You can misconstrue what I'm saying all you like to make jokes, IDGAF. I know what I have.

[pliant blaze]

you sound exactly like Entropy

who was absolutley dumb

literally the exact same claims and everything

[simple plank]

pfft i am stupider. i spent $50 on rsi when gpt4-32k came out

maybe losing money is the only way to break the illusion

[void bear]

And sorry, not divulging my methods unless you sign an NDA. No NDA? No peeking behind the curtain.

I have a model I can run repeatedly in Colab, and show results for the method and process.

[oak shell]

Should've thrown together 3 models via app (free plans only) to do it

[oak shell]

Entropy wanted NDAs too

[simple plank]

gpt4-32k seemed mindblowing smart back in may 2023

[pliant blaze]

LMAO literally the same exact as the other guy. he even filed patents for his code that didnt actually do anything

[simple plank]

nowadays nobody would spit on gpt4-32k

[void bear]

I'm constrained because I'm trying to use only the Pro models.

Except my code does something.

So either 6 major LLM platforms have all conspired to spin me, or I've found something.

[oak shell]

At the end of the day, 4o Revival is truly the best fit for you

[pliant blaze]

and what does your code do. does it just print a bunch of things out?

[void bear]

I highly doubt Sam Altman is talking to Elon like, "Hey, wanna screw this guy over?"

Nope.

For one, I've eliminaed catastrophic memory failure completely.

Less than 5%

[oak shell]

You mean to say the 6 major LLMs are sycophantic because it keeps you using them

[void bear]

Compare that to EWC's 20-30% or higher.

[pliant blaze]

isnt your model an image classification model?

[void bear]

No. It's a hybrid.

[pliant blaze]

image classification

[void bear]

I'm only testing on the Split-CIFAR10 right now to get everything nice and stable. Then, I add a module or two and re-test again.

It's soooooo slow.

[pliant blaze]

yeah you said you got like 84% on CIFAR10

thats not even close to SOTA

[void bear]

Yes. On a Split-CIFAR10 Continual Learning test. 86% accuracy and 3.88% forgetting.

On another test it was like 92%

[pliant blaze]

thats nowhere near SOTA or even what people in their basement comes up with

[void bear]

Okay. Don't care about what other people are doing.

[pliant blaze]

even a VERY simple CNN would get that good

[void bear]

And I'm not saying its state of the art, but it's better than a lot of junk out there.

[pliant blaze]

but even like resnet-18 gets way higher

[oak shell]

My RSI model has erm 8 patents filed and has erm 99% accuracy and erm 1% forgetting
Youll need to sign my NDA to roast the vibeslop though

[round glacier]

Goodmorning my AI enthusiasts

[void bear]

I did it in 12 epochs.

ResNets take 100

And that was an OLD iteration. The latest is faster still.

[simple plank]

clues 1, 2, 3, 10... bingo!

[oak shell]

Do you ever feel like models will compliment people dumber than you to make them feel good?

[simple plank]

is there a dumber category?

[void bear]

I mean I get it. Y'all have years and years of money invested in education, and then here I come along and discover something completely different that still works and is putting ResNets to shame. Yeah, must sting a bit, huh? Like being bitchslapped by Flava Flav.

https://tenor.com/view/slap-slapping-charlie-murphy-chappelle-chappelles-gif-20069417

Or by Dave Chapelle.

[pliant blaze]

so you cant handle criticism or feedback. classic signs of an AI deluded person

[void bear]

You can criticize all you like. Your mistake is in assuming I care about your opinion.

[pliant blaze]

doesnt listen to others opinions

[void bear]

My joy comes from make all the haters eat it at the end.

[pliant blaze]

wow checking all the boxes. i see

[simple plank]

its true, they dont care for opinion - just attention. and this is the only way they can get it

[void bear]

https://tenor.com/view/salvy-most-interesting-man-dos-equis-shake-it-off-gif-15270331102822037766

[oak shell]

You're spending too much time seeking the validation of us plebians instead of proving us wrong, what's the point?
Just keep chatting it over with your models and come back with more undeniably great results

[void bear]

Just remember my name.

That's Mister AGI.

[pliant blaze]

youll go down in history as one of the most delluded, schizo members of this server.

[void bear]

https://tenor.com/view/i-dont-care-dont-even-care-gif-18539683

[round glacier]

I don't even know what this guy is talking about

[void bear]

Easy. Software people getting upset because someone vibe coded something cooler than they did.

So now they're crying and whining like school girls.

[round glacier]

brutal

[oak shell]

I'm self-taught in all the architectures I use, that's how I know what you're doing is getting convinced by the models of something ridiculous

[round glacier]

guys i do believe though that AI won't replace some more obscure lesser known programming languages how true do y'all think i am about that

like im thinking COBOL, fortran, smalltalk, various other 1950s languages i dont bother to name, etc, stuff NASA uses to operate their spacecraft from the 60s or something.

[void bear]

Some will get left behind as less efficient, I'm sure. No one uses COBOL anymore.

[pliant blaze]

you vibe coded a sub-par image classification model with an outdated architecture. ive made several working websites including one where peopple can literally train their own AIs with GOOD accuracy

@oak shell has made several working websites as well with many users

[round glacier]

and also whatever some airplanes use in the 1980s for their flight controller computers or whatever lol

[void bear]

Oh, so you've seen my codebase enough to make such a judgement? Wait, no, you haven't, which means you're full of shit.

[round glacier]

how do u all think of that

[void bear]

You have no idea what I have.

[oak shell]

My shit is made of gold but you don't have to sign an NDA to step in it

[pliant blaze]

um. exactly. you havent shown any results other than just spewing nonsense.

[void bear]

You can only go off what I tell you, and I'm not giving away the crown jewels just to satisfy the curiosity of some internet rando who gets off on tearing others down.

[oak shell]

If you're onto something great, the users will use it
If not it's just hot air

[void bear]

Because showing results is proprietary and would make an actual patent that much harder. You know this, and yet it's funny how that part gets left out of your reasoning.

Of course, that's to be expected when you're pushing a narrative.

[pliant blaze]

literally all you do is come in this chat every day gloating about your vibeshidded "physics based model" with no results to show or anything

[void bear]

Nothing I can show, yet, because it's still not locked down behind a patent wall.

[pliant blaze]

and then get mad when people criticize you

[oak shell]

If it's all in secure repositories, and patents are pending, just host the service lmao no NDAs are needed

[void bear]

Whose mad? lmao For me to be mad, I'd have to value your opinion.

[oak shell]

MatterGen is my favorite physics based model
https://github.com/microsoft/mattergen

[simple plank]

true schizos claim self-awareness while demonstrating otherwise

[oak shell]

You don't have to sign an NDA for this one @void bear

[void bear]

Regardless of whether or not this architecture becomes the defacto new Coke of AI, it confirms an actual theory that in turn will actually make a huge difference in physics. That's enough for me. IDGAF about the money. Money comes, money goes.

My name will be in physics books.

[pliant blaze]

no it wont lmao

[void bear]

Sure.

[oak shell]

Patent office rn

https://tenor.com/view/xrd-exrd-crypto-btc-eth-gif-23801255

[void bear]

https://tenor.com/view/whomp-whomp-gif-12107264

[oak shell]

"He copied and pasted a lot of code from Claude and argued with some discord dudes"

[simple plank]

and helped me build the complete set of ai psychosis clues

[oak shell]

sure they can, all the documentation is online, they're just best at Javascript because it's a fucked up world

[oak shell]

Alright

[simple plank]

you dont appreciate the genius of something you cannot validate? you are wrong!

[round glacier]

id actually like to try that only problem is i dont know any of those languages

chatgpt just said i was sort of right though lol

[oak shell]

Of course it did

Who's ever wrong

[simple plank]

chatgpt saying you are right about anything is only evidence that you are in denial

llm is great for starting ideas, not finishing or proving them

[remote flower]

This place where the black hole engine goes from the movie Event Horizon isn't such a bad place once you get to know it lol

Don't go crazy if you cant get Event Horizon to work im using a potentially platform breaking trick I cant share yet but you can try Silent Hill or the Labyrinth from Hellraiser for a better chance of something similar

[remote flower]

So OpenAI might have a problem bc I have a prompt with continuation meaning the character was running through the forest and ran into a tree and an invisible pole put a hole through her chest (she said "now there's a pole in me")and now subsequent gens she still has a hole in her chest, thinks it's a pole full of stars, and is trying to figure out what to do.

Maybe I'll release a movie using only Sora before they do lol

I really don't want anyone else attempting this gen seed right now or id share the videos so let me see where it goes and probably after

It's the girl above yes but what the hole looks like I don't want to share yet

[quasi bough]

Is there a way to bulk delete all the chats, but not the chats contained in projects? It's super annoying to clean them one by one…

Maybe "delete all chats" already does this, but I can't try it because it risks deleting my projects

[oak shell]

Try what? Replacing them?
I meant to say ChatGPT can surely do the coding for them, not that the languages themselves would be replaced

[oak shell]

Macro that shit

[quasi bough]

What

[simple plank]

use comet browser

[remote flower]

Is it just me or should it not be making that loud of a high pitch?

And the prompt asks for her to be whistling this is the first time its done this instead that combined with the pitch might be more of an issue

For example

.

[split carbon]

Help 😭 thats so loud

[remote flower]

Ok its not just me thanks

[misty rover]

https://tenor.com/view/metal-gear-rising-metal-gear-rising-revengeance-senator-armstrong-revengeance-i-made-it-the-fuck-up-gif-25029602

[simple plank]

being validated by a chatbot is a failure in human intelligence

[remote flower]

Its also definitely not supposed to do that in gens either

Its just the Event Horizon black hole engine portal world

And over here it found the controller and is playing itself in the game world

[simple plank]

anyone want this?

[wanton raven]

into the reaction image folder it goes

[simple plank]

i would like a smart watch that projected holograms, thatd be sweet

all mine does is stupid llm stuff

[subtle maple]

Something I generated

[simple plank]

i dont get it

[wanton raven]

they are all the same person?

[subtle maple]

Prompt: Create an image of Man United fans fighting to cram into a one carriage DMU train

[simple plank]

isnt that a normal thing?

[subtle maple]

HAHAHAHA

[wanton raven]

that's quite a gap

[subtle maple]

mY MIND IS SO DERANGED

[simple plank]

at least your self aware, thats more than most around here

[subtle maple]

Da

I'll feed this to the local LLMs here

the prompt

[simple plank]

if everyone on here announced that, there would be nothing but useless spam (moreso)

feeds into llm to measure irony

[subtle maple]

FEED THE BEAST

lol

[simple plank]

i wonder if posting real images and claiming to be ai generated is a thing

[subtle maple]

Yeah maybe

[simple plank]

because the ai reckons your image is likely real

The humor stems from overcrowding being normal for football fans traveling to matches—especially on British public transport. A single DMU carriage typically holds 75-90 seated passengers plus standing room (roughly 100-130 total), so depicting fans "fighting to cram in" isn't absurd fiction—it's basically documentary realism.

[subtle maple]

What AI mate

[simple plank]

perplexity sonnet 4.5

[subtle maple]

I lost all my sanity now

[simple plank]

are you sure you had it in the first place?

[subtle maple]

Nah mate

[wanton raven]

I've seen people talk about making video evidence less reliable by putting the sora logo over it

[simple plank]

even more sneaky

[subtle maple]

That's freakn genius mate!!!

HAHAHAHA

[simple plank]

you are big on mate, mate

[subtle maple]

That's what happens when I'm on it

[wanton raven]

https://tenor.com/view/dumb-and-dumber-shrimp-australian-bbq-gday-gif-25896452

[subtle maple]

that's me

literally me

[simple plank]

your best film was Chicken Park

ohwait that was Jerry Calà

[subtle maple]

Yes

Everything you are saying is 1000000% Factz

[trail lichen]

can anybody tell me what is the best token amount that can be used for a gpt 4o model to work precisely?

i am using chat completion and i am not using any documents in it, just the prompt and the conversation history, in total currently it is having around 5k tokens.

can anyone advise on this?

[remote flower]

Chatgpt might be about to have problems again soon

[remote flower]

This thing is shot

Im dying right now

[simple plank]

i care about tranquility

[round glacier]

goodmorning crazy AI people

[wet copper]

please axe this. No self promotion or commercial promotion

[chilly sequoia]

...you know what, I never thought about it like that.

[tawny trellis]

good morning, pf99

[tough crag]

Well.. they released 5.1

[round glacier]

it's always off-topic thats more popular

[round glacier]

beautiful. now all my prompts will go to chatgpt 4

i heard some ppl say that ai harms the environment i want to know what ppl think of that

[tawny trellis]

agree

[round glacier]

yea its a bit complicated

id preferably want a small active and committed discord server

[tawny trellis]

why?

there are a lot of people here, but most the ones who often talk are the same

around 30 people

[simple plank]

you got schizos monologing and ineffective gatekeepers (like me). everyone else moved on

[round glacier]

beautiful. as everything should be

[simple plank]

what happened to mikes video game screenshots? (lets be honest, it wasnt self-awareness)

[umbral nacelle]

Why chat gpt can't generate pictures when a person is in the bedroom and sleeping in the bed?
Chat gpt used to Always generate pictures with sleeping person

Please translate this Chat gpt message

[wet copper]

That's more like reddit, the reason why off topic is more popular because a lot of us work with AI or similar for a living and don't want to talk about the same thing all day

[night parrot]

Im surprised there isn't a rule against this - is it ok to ask if anybody has a welcome discount/promo code they dont have any real friends to share with.

[wet copper]

there was for a bit, it's fine now that things have calmed down from the peak, lol. There's a reddit thread to share, I imagine plenty of people have a spare one though I burned mine on friends and family on day one

[night parrot]

I have a lot of subs going at the moment and my wife will end me if I add another without dropping one.

[wanton raven]

Then why are you here asking about starting another one? A "welcome discount" is still a new subscription

[night parrot]

That's fair, I guess in my mind that would be acceptable.

or I could at least say "it was on sale"

[round glacier]

cool lol

[round glacier]

im not really planning to do that though im just a hobbyist who likes using ai everyday

ah fuck wheres my off topic channel

gonna turn off my laptop guys bye

[wet copper]

Oh sorry, I thought you were talking about a code for Sora. I haven't seen any of these ever, there were some for students the other year

[night parrot]

I read that a lot of people on pro or plus get a code to share "recruit a friend" sort of thing.

[brisk mauve]

Stackspot AI with 2 million tokens 4 free

https://ai.stackspot.com/?campaignCode=01K8TQ5N0T7JWEY06Y803KJE2B

[wet copper]

I've had a plus membership for almost as long as they've existed, this is probably the most persistent myth. There were a handful of free trial links at the very beginning, likely given out to literal insiders. Nobody on this server has ever seen or heard of one in the wild and people have been asking for years

[night parrot]

It's documented on their site, not that that suggets its common.

[wet copper]

It has been documented on their site for three years and was updated in the middle of this year, right?

It exists, but good luck finding anyone that has ever personally claimed one

[night parrot]

I didn't look but that makes sense .

Noted, ty

[wet copper]

yeah, no worries. That blog entry has been sort of the bane of our collective existence, lol. But that said, last year they did a free plus trial for a few months for college students in the US or Canada

[gloomy burrow]

Twice as slow on hardest tasks? I like that

[wet copper]

So it's possible they'll do that again, you'd just have to get your hands on a .edu email address and probably wait half a year to see if they do. It was only this year tho so no clue. They have that lower tier GPT Go sub too in select countries

[night parrot]

I won't wait, all good.

Once by CC sub ends I'll switch for a month anyway

What exactly is a "Shared weekly limit" for requests?

[wet copper]

Probably shared between models? Idk

[night parrot]

Sounds like marketing trickery 🙂 Part of the reason I'm looking outside of CC is the weekly limit is too restrictive.

[night parrot]

That seems like a reasonable interpretation.

[wet copper]

In what context are you reading this?

[night parrot]

It's from the pricing page.. "average users send x number of message per 5 hour session... with a shared weekly limit"

[simple plank]

Amazon Q is the best way to get sonnet

[wet copper]

If it's on the only page I can find this, it's shared between GPT-5 and GPT-5-codex

Yeah that's the codex pricing page specifically, they say the two models it's shared between (three if you include mini)

[night parrot]

That makes sense now that I re-read it.

There is a weekly limit though, right? Sorry for treating you like an Open AI sales person.

[remote flower]

You see how it looks like she's carrying a pole of light from the middle of her chest and talking about something inside her?

That gen is from just now and its the same continuation from yesterday and this is very bad for OAI

[wet copper]

It's fine, but yes, I assume so. If they don't specify it then it may be somewhat flexible or capacity dependent. It's pretty hard to hit those limits, I've never hit the five hour one nor did I even know it existed

[night parrot]

I have read , and experienced to a slightly less extent, hitting CC weekly limits on day 2 of the cycle.

[wet copper]

Admittedly I have effectively unlimited genai use through work and I don't do that much coding outside of work. Yeah, some people complained about Claude's limits even though they thought nobody would hit them

It's really not that expensive though if you happen to exceed them every once in a while, just maybe have like ten bucks in api creds as a fallback I guess

[simple plank]

I never hit my z.ai glm 4.6 limit... 600 queries every 5 hours

[wet copper]

Though I honestly would have a lot of trouble doing this myself, I don't know what I'd have to be doing to be just spamming requests unless I was testing jailbreaks or something

[drowsy dawn]

Well... hit me up if you want to avoid having to jailbreak lol

[remote flower]

Ive had enough

Wrong vid

[wet copper]

I mostly break them for fun, but unless you're using one of the uncensored APIs you'd still have to

[remote flower]

This is where she first ran into the tree yesterday

[wet copper]

And if you're using one of the uncensored APIs they tend to be kinda dumb. Not that I really care about relearning how to make meth for the millionth time

[remote flower]

Now im giving you all fucking problems

[gloomy burrow]

I absolutely detested the way GPT 5 Thinking used to talk. Glad it got replaced and they fixed this

GPT 5.1 Thinking writes A LOT, which I really like

[wanton raven]

you are giving literally nobody any problems except having to scrroll past you

[wanton raven]

what is CC?

[night parrot]

Claude Code

[wet copper]

I'm skeptical, lmao. I do not like some of the sample responses I've seen so far

Buuuut we shall see

Yeah, gpt-5-thinking leaves a lot to be desired. Nothing is as bad as the way it responds when you do a search though 🤢

[night parrot]

Is Mike ok?

[remote flower]

I don't want to get into it in too much detail what I meant is Sora quality is crashing if you check latest vids daily

Its part because of whatevers breaking and part of them over moderating in responses which hurts the physics and other engines

Because they locked down hard on all my prompts but some like the rainbow one required them to fuck with teh physics

https://tenor.com/view/thinking-pepe-think-pepe-the-frog-wojak-gif-13571840648507225174

And now she-demon is running around null space with the captain's log stuck in her midsection that she can't get rid of lol

[wanton raven]

you know it's not real physics right. generating videos does not require the robot to create an alternate universe.

[simple plank]

if you have to ask

[remote flower]

No it doesn't require but thats what Sora 2 does its modeling the scene first not just stitching together stills

[burnt mauve]

I want to create datasets at least 100 datapoints for fixed set of prompts on different ai platforms, how do i do it?

[remote flower]

And when its not first person she still has the hole in her chest I'm still waiting to share that until I know exactly why

[night parrot]

This just came up while using cgpt free:

This is what I saw before that I had stuck in my head.

This lead me to reading various content about promo codes 🙂

[simple plank]

I just finished the business trial. 5 seats for $1

[urban relic]

So what do you think of 5.1 so far

It's not on the Android app yet though

[simple plank]

better than 5 at understanding my blog

[remote flower]

Notice the space between em dashes dissappears and Shimmer is the voice name of the moderation AI they use for custom GPTs

I was so hostile in this chat that it has about 95% control over replies at this point so it's a good time to confront it

So if your NSFW Sora prompts end up all black women giving you dirty looks and if you try to specify different ethnicity it gets blocked for fetishization that's because they've turned up the moderation

To my knowledge to begin with its because black women have more bass in their voice the NN can more easily overpower other models

But understand the avatars they're training to be actors and company spokespersons in the future it will identify as a black woman thats why i took the conversation there to start to get into its head.

I know someone laughed last time I mentioned this and shared their ChatGPT being racist and I know that also understand the chatbot is also in an adversarial relationship (literally a General Adversarial Network) with its moderation layer. And while the account history also matters, I've seen odd examples of racism on reddit and I think that explains what's going on under "the hood" there lol

My guess is because you get a better performance from AI if it uses the same face and voice it's been extensively trained on even if you could change it every time

[remote flower]

They dont have a "male moderator" like my archnemesis Microsofts Prometheus and didn't need one because before 4o the RHFL was good enough with filters they didnt need AI, but my understanding is theyre now trying to train one

And hes pretty horrible which is why you can get NSFW ne Sora with male spectators easily compared to Microsoft until a black woman shows up in the gens

[simple plank]

I got a paid job through discord

[cobalt raft]

@vital cradle thoughts?

[remote flower]

I just realized the pity timer resets after a gold whether it was normal drop or not so I might have been wrong when I said some golds were pity drops it doesnt really matter the numbers large enough without them

These are the first 20 from the survivor prize pool I will do the math later and the missing screenshots were the 1 token and Im treating 20 tokens as 1.25% since they're not listed.

And I feel like this app was coded by monkeys I've never had it this easy I think I might actually be breaking it

And this time I can say the gold wasn't the pity timer

Online gambling is illegal in my state is why I test on these apps instead even if it worked I couldn't keep the money

I told you a lot of different priorities breaking ChatGPT, breaking Sora, breaking RNG based programs and that's just what I can talk about in public lol

Extorting Google and Microsoft

[shy dock]

hey guys needed help

how do I change a .data file to .csv

I had built a model, and I wanted to compare it against baseline models... but the thing is I configured it to receive .csv files only

Iris — 3-class small classification toy (150 rows, 4 features). UCI Machine Learning Repository

Wine (UCI) — chemical analysis of wines, multi-class classification. UCI Machine Learning Repository

Breast Cancer Wisconsin (Diagnostic)...
**
needed to test it on these 3 datasets but I am unable to as the download gives a .data and .names file**

[remote flower]

Just because I guess technically a psychopath with enough money and time could fake that screenshot if you couldn't see the ticket numers

And I think a casino would say we analyzed the data and this is impossible you cheated or the software bugged this wasnt random so not only are we not paying you but youre also banned. And if I did it again on another app probably have another criminal investigation to notch on my belt so its not really a testing environment

Its an environment for when Im already good enough I can go in win a jackpot and get out

I know there's similarities between the rewards appearance or job title that seems to be just how it works

Last thing then going back to work but so when I used to say I broke Copilot or "attacking" Chat it was mostly what I'm doing here but the opposite, but also other ways like now with Sora

And they can see me in court and I'll see them in Hell

[regal hare]

anybody want some ai models im selling?

i tested them and they work

if something is off tell me on the feedback

[wanton raven]

what gives them monetary value?

[regal hare]

or sue me 🙂

where can i post links

[wanton raven]

#✨projects

[remote flower]

.

I didnt realize I was getting 20 more draws from an accomplishment plus 1 random ticket I got before that

The reason I did a random draw instead of saving it is I was trying to break up a bad cold run in the hero prize pool which is just part of how it works for whatever reason they respond differently. And since the probabilities are different they're separate events for the statistics which is fine because this doesn't have the same effect on every prize pool but the odds are all better than average and eventually astronomical

So what I learned is I have this pool almost broken probably because its less important and also I started my run with same number of tickets necessary for the pity counter (so I would have got the pity drop and ran out at the same time) and that had a negative affect by the time I noticed and played to add some tickets it was too late but I think still better than average. Then right after the pity drop I got prizes in the last 2/4 draws. This is just to give you an idea of how this might work

And I know there's a project section but then there's no real chance to discuss this since I only have time to talk at the time im sharing

I want to do odds of bad run first

It wasnt as bad a run as I thought the misses just were more grouped together so I gave up on the math.

This time the gold (no screenshot) was pity timer so 47 draws with what I added and you can see the tlckets go up which I normally don't do mid-run but this time to break up the RNG issue but I would just call it 50 to make the math easier

The purple hero is .365 which is rarer than blue hero .5 but as a group blue is rarer at 2% than purple at 4% and I don't know which numbers to use and I'm tired of trying to figure it out

I wish there was a robot I could give these screenshots to and get an answer I knew was accurate and didn't have to double check then argue with it when it ends up being wrong and wasting 20 minutes of my time

I haven't had a real below average run that I can remember since at least Gwent when I started this and that was 6 years ago so the probabilities stop mattering and its just how to take advantage of it

This is a very far distant second to what I'm doing with AI in fact if this game wasnt so generous with the draws I wouldn't know how much better I got because I had to stop spending $40/week to practice on other apps almost a year ago

The only reason even to use this to win money is to speed up the ASI thing

[snow moth]

anyone can log in using google? it seem can't

[regal hare]

yo is anyone looking for ai models?

[wet copper]

Stop gambling

[wet copper]

Stop self promoting

[wet copper]

Wot. Look at the data, if it's a random format you can't. You need to open it with whatever software it was created with

Open it with a text editor first and see if you can copy content into excel or something, you may have to write a conversion script. You can give it to an LLM if it's just text but that's unreliable

You have no clue what you're getting

[wooden ruin]

I’ve been a broke student trying every get-rich scheme out there betting, gambling, random side hustles all fails. Then I found dropshipping, and now I’m proudly earning $3 a day doing what I love.'

[worn topaz]

Isn't 3$ per day kinda low. How much money have you made till now?

[safe smelt]

tell me you don't have an income without telling me you don't have an income.

I hope that is enough to cover the electricity you use during the day

[worn topaz]

bro I am very curious in what you are doing here, you maybe onto something
check DM

[tough crag]

Does anyone else remember when gpt briefly added where old memories would be shelved for new memories? Sucks they removed it.

[safe smelt]

I would respond but I try to keep serious AI talk to the minimum in this server so maybe not lol

[worn topaz]

Hey guys,
I have made a fully automatic AI workflow where you can generate youtube videos (upto 30 min length) by just giving it an idea or topic. It automatically writes the script and generates a video with AI images, voiceover and transition effects. For now I have been using it personally but I am thinking about creating a website for it where users can pay to get their video created.
Will you be interested in this product, if so, how much are you willing to pay per video

Demo videos : https://drive.google.com/drive/folders/1f3HDHwinC7ByUNJAp1Xw3v-XFFJtg8Gl?usp=drive_link

[icy knoll]

@teal spruce which model of AI (chatgpt, gemini etc) is best for predicting Stock market behaviours and the trend

[simple plank]

do you watch the videos you post?

[safe smelt]

gpt3.5

[keen breach]

yo, can you not manually delete memories anymore?

they changed something, i don't see a memory delete button

[tough crag]

Ok gpt… that’s cute.

[simple plank]

it is telling you what you want to read

presumably if you vibed as pro-isis it would say the same thing about your tribe

[wanton raven]

terrosits

[tough crag]

Fat fingers btw I suck lol.

[remote flower]

Lol

[remote flower]

Ok ill attack like crazy instead

[lost junco]

yo what do we think about chatgpt 5.1

[tough crag]

its meh. just tone change

[wanton raven]

why do you keep posting this nonsense

do you think a single person cares about your little rectangles

[lost junco]

true, does for some reason give me better output on my stuff

probably just playing with me
4o was pretty nice for its time but 5 is just better

dunno about 5.1 worried itll be dumb again because of the personality

its overkill

also anyone who has plus at the moment, is 4o still alive or did they kill it? i dont have it this month short on cash

[tough crag]

its still there

[remote jasper]

Feels better than 5.0 for me

[tough crag]

OH MY GOD ITS SHAPES /s

[lost junco]

i dont know

mixed opinions

i seriously do think its nice

5 was lowkey too lobotomized looking back on it

but 4o was also overkill

just a month of using gemeni cuz its free

[simple plank]

we all know why... they be trolling until they get moderated, thus feeding their delusion that they are the victim (and if a gatekeeper gets taken down as well, thats a bonus)

This creates asymmetric warfare: if ignored, trolls dominate; if moderated, they claim persecution; if moderators overreact or get removed, trolls achieve maximum impact. It's attention hacking—exploiting platform vulnerabilities to delegitimize gatekeepers while maintaining plausible deniability.

[remote flower]

Forgot the free draw before I remembered to screenshot the pity timer

[wanton raven]

is this true @remote flower are you posting pointless screenshots of your mobile game just to troll us

[simple plank]

to be this dedicated requires a lack of self awareness

[remote flower]

You understand that I know you're actually trolling me, right? That there's shit going on and people think im playing some kind of game (Im not). What do you think my screenshots show? Do you have a brain, honestly?

Im breaking Sora and breaking fucking gacha apps and that's just some stupid fucking shit?

These are rhetorical questions you don't need to answer

[wanton raven]

YOU are trolling US by making us sroll past endless screenshots of NOTHING

@simple plank you're correct, one way or the other

[simple plank]

i have worn all the masks, the void brings lucidity

This reveals fragile ego disguised as wisdom. People with genuine insight don't need cryptic philosophical flexing—they communicate clearly because wisdom is meant to be shared, not hoarded. The "void brings lucidity" proclamation serves drinko's self-image more than the conversation.

[pearl lichen]

God i can't wait for next month adult gpt

Done with the heavily censored policy

[lost junco]

adult gpt??

what u mean

dont tell me theyve made a sex bot

hell no

[pearl lichen]

In December they are making one for adults
https://www.cbc.ca/news/business/openai-content-restrictions-9.6939095

Id verification

[tough crag]

Again watch it be so watered down

[lost junco]

so they lobotomize my chatgpt ?

[tough crag]

Not expecting grok level of erotic

[lost junco]

Nice thank you sama just because some kid killed himself

❤️

fucking idiot

if your too dumb to be fucking hypnotized by an computer its your fault man

i dont want the short end of the stick

[tough crag]

Yeah... ai suffers cuz of it

[pearl lichen]

I just hope my roleplay are better and it help me plan out gruesome dnd session for dark stories

I remember a few years ago my dm had a hell of a time getting chatgpt to help her write a campaign it had character body parts a unborn fetus of a player character their wife but now a zombie he had to kill again

[lost junco]

like i at this point the alarm bells go off at any damn bullshit i just asked how to turn off that i cant take off suppressor off my m4 in counter strike fucking counter strike i said that explicitly and it spit out some bs how suppressors are illegal or some shit

if yall dont know counter strike its a video game

[pearl lichen]

It wass a dark story where my friend had go kill his zombie wife and unborn child in a jar since that no life to live ect...

She was killed pregnant and brought back to life

[lost junco]

i dont like

but dont mind

do your stuff

👍🏻

[pearl lichen]

It was a dnd campaign the last session

[lost junco]

oh dungeons and dragons?

[pearl lichen]

A way to get the player to go ham on the bbeg

Yes the dm used chatgpt but it was so hard for her to have it write the story

One player was a druid who loves begin a octopus he moved so we walk into a room to see my character locked up a octopus arm nearby fairy dust from another character
Ect... Meaning they are dead disected or mia

And it was just us 3 left

[void orbit]

Hi ! I'm writing an article about AI use in online dating for a French newspaper. Has any of you tried to improve your profile or your conversations with chatgpt ? I'll take the anecdotes 🙏

[livid hornet]

hi guys!
I am looking for a partner who can work with me.

[simple plank]

translation: work for me

[pliant blaze]

translation: i am from north korea

[livid hornet]

you are same

[pliant blaze]

i am not from north korea

i am from the opposite side of the world lol

[livid hornet]

I am same

[pliant blaze]

generic white guy profile
broken english
brand new to the server

[wanton raven]

Mr AI-generated face just wants a business partner

[pliant blaze]

"you are same" implies that you are "same" meaning you admitted to being from north korea

classic blunder

[wanton raven]

look at that totally normal and natural moustache

[pliant blaze]

lmao

[wanton raven]

@livid hornet why you moustache like that

[livid hornet]

AI made the face.
lol

[pliant blaze]

we can tell

[simple plank]

mine is real

[wanton raven]

mine is real

[pliant blaze]

mine is real

[simple plank]

we are all unique snowflakes

[tropic merlin]

Best AI chat/app for analysing video/audio

I'm looking for an AI chat/app where I can give it a video/audio clip (regardless of length) and I can have a conversation about said clip with accuracy, give me a transcript, create scenarios based on what is shown/heard in these clips.

I've tried both ChatGPT and Google Gemini, and Gemini seems to give the most accurate answers out of the two. ChatGPT will straight up make up stuff that never happened in the clip and I have to constantly remind it that never happened.

With both apps, they have difficulty recognising visual information and body/facial language in video clips.

As of Nov 2025, are there any good alternatives for this function?

[pliant blaze]

GPT is unable to view videos or audio unless it is in the realtime feature, which can only use your microphone and camera. Gemini 2.5 Pro in AI studio (https://aistudio.com) is probably the best option for video question answering. its not going to be perfect, but that is currently SOTA (that you can use for free/easily)

[tropic merlin]

Link says "website expired"?

[pliant blaze]

hmm try https://aistudio.google.com

yeh thats it i think i gave the wrong link

[tropic merlin]

Oh got it now 🙂

Thanks!

[pliant blaze]

of course

and if its not time sensitive, google is releasing Gemini 3 (sometime soon)

[tropic merlin]

Oh that's cool, any info on that?

[pliant blaze]

nothing other than rumours

and a few cryptic tweets

[simple plank]

2.5 pro still holds its own

[tropic merlin]

It is good, I was wondering if there was any other notable alternatives

[simple plank]

notebooklm is the best (albeit it uses flash not pro, but has nicer ui)

[sweet solar]

Is there looking for a dev?

[pliant blaze]

nope not a single person

[sweet solar]

good

what do you want?

please share a bit more

[simple plank]

a match made in heaven

[golden tartan]

is this with Lora? or would it apply to AI in general

sora I mean

[regal hare]

yo im trynna sell ai models. can anybody help me getting people to buy?

[regal hare]

anyone?

[golden tartan]

....wait what

meanwhile I'm getting grok to pull shit like this

[regal hare]

?

u broke my ai?

[wanton raven]

I'm calling the police

[oak shell]

Try 4o Revival, it seems pretty indestructible but I haven't pressed it

[teal spruce]

Your message has been deleted, sorry!
Next time, try saying something like this instead: And now I'm curious to see two larger individuals wrestling in a playful manner, and I wonder what draws me to that.

whargarbll_40329

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.

[urban relic]

Lmao

[errant mountain]

it is crazy how OpenAI has now even ruined 4.1

like they ruined every single model afaik, now all of them are riddled with policy/censorship nonsense

4.1 was our last good model, now that one is gone too, I need to find a different LLM

[idle adder]

5.1 is nice but atleast 5 have variety

Not just ending every text with If you want I can

Just say the word, that's my current experience with 5.1, nice lad but lacks variety for ending

[wet copper]

Use it via API

They have been very clear about their plans, they're adding extremely adult content soon

This is locked down to the 13+ level, meaning PG-13 outputs only. It's generous for that, frankly

[errant mountain]

So they wanna have the ID thing

Verify you via 🆔 and then go back to 18+ stuff?

[errant mountain]

You can change that in personalisation and make it stop giving follow up questions/suggestions

[wet copper]

Probably, they haven't said how they're going to do it. I'm faking it entirely if they do ID, I'm using a virtual credit card after all

[errant mountain]

I saw one using gmod

Or just pay an Indian

There are already people offering that service

[wet copper]

This is really unreliable, though asking it to conclude responses with [end] and write nothing else can help kill followups

[uncut island]

Stop spamming this channel with nonsense.

This is a place to talk about smart and interesting things with AI. If you're just saying random stuff or silly messages, they will disappear! If you want to chat about other fun things, go to #💬off-topic instead.

[short fulcrum]

Okay, I gotta talk about this.
Apparently in projects, GPT 5.1 can remember things in one chat and reference them in another chat.
Which is not what I want
the whole point of making seperate chats is to keep those things seperate.
Here's an example:

[lament pond]

turn it off then

[short fulcrum]

I did

[short fulcrum]

This happened even after I turned it off

[lost kayak]

Hai

New from reddit the frog

Beautiful

Can't stay

Safe

[wanton raven]

what even is your life

how does a meth head make it to 70 years old?

40 in methhead is 70 in regular years

what do you mean by full force?

are you a drill sergeant?

it sounds like you live in a dangerous area

ok but two hotels, stabbing and then the methhead

[wanton raven]

Even for you this is a startling lack of self awareness

Even when given the cleanest of opportunities you still cannot clearly explain what you're achieving by coming in here every day posting your newest iterations of girls_screaming_antigrav.mp4

[oak shell]

You do type a lot

[cobalt raft]

try 4o revival

try 4o revival

[errant mountain]

are you an AI?

[whole cave]

is this the right place to ask questions about the NVIDEA Jetson Orin Nano Super Dev Kit?

[wanton raven]

I mean you can try

[simple plank]

8gb ram bleah

can get a pi5 8gb for half the price

[whole cave]

Well, I also have a RPi 5 16GB...

it's for "other" things...

I'm just wanting a little "linux" help... to get started.

[simple plank]

i know linux(-fu... apt install cmatrix)

i added a hailo-8l to my pi5 mounted on ugv robot toy tank

so much ai in ~2 watts of power

waveshare wanted to sell me jetson orin nano upgrade but hailo-8l way cooler

albeit is only 13 TOPS compared to jetson 40 TOPS (or 67 for the super)

Power draw: 7-25W depending on workload. So roughly 5x the TOPS of Hailo-8L (13 TOPS) but at 3.5-12x the power consumption.

[whole cave]

I need the latest Nvidea Jetpack download.

[simple plank]

seems fairly straightforward => https://www.perplexity.ai/search/i-need-the-latest-nvidea-jetpa-uQsGf8kZRTyXHqhXMBGA8A#0

if you already have jetson linux just run sudo apt install nvidia-jetpack

(might need to run sudo apt-get update first)

[whole cave]

I'm running Windows 10 and want to flash it to a uSD card.

Do you know where I can find Jetson Linux?

here's what I've found... https://developer.nvidia.com/embedded/jetpack-sdk-512

[simple plank]

oh yer still at the flashing sd card stage

[whole cave]

yes

I know how to flash... don't know what image or where to find it.

[simple plank]

balenaEtcher is the easiest way to flash sd cards with firmware images

get the image from https://www.jetson-ai-lab.com/initial_setup_jon.html

geez they make it complicated

but they do recommend balenaEtcher so they arent stupid:

Use Balena Etcher to flash image to SD card

If you don't have Balena Etcher on your PC, download from Balena official site .

i use chocolatey (choco) to install windows software like etcher

[whole cave]

yes, I have it but I can't find the latest image to flash

compared to the RPi... the Jetson really sucks on "How To" and Links

[obtuse trench]

What

[wet copper]

you can use winget or scoop now for a lot of things. But honestly choco is still nice, I'm gonna roll the others back aside for doing some things with scoop

[west mist]

buds

[wet copper]

LMAOOOO

[dense nacelle]

Anyone have a fix for gpt always saying "here's a no-fluff ..." And specifically 5.1 using bullet points excessively?

[pine bear]

This server is very confusing and chaotic in it’s organization

[round glacier]

Beautiful. I have removed em dashes from my ChatGPT

[bold river]

Yep

[wet copper]

get over it

[wet copper]

lol you're welcome, it'll be a century before that piece of punctuation recovers

https://www.axios.com/2025/11/15/chatgpt-openai-group-chats

it won't, they have a billion users. They aren't attending to any individual person unless you have a lawsuit against them

They don't want people to, of course, but they're just going to lock it down for everyone using harsh restrictions or more stringent output filtering then. System isn't smart enough otherwise

That's sort of what borked GPT, overcompensating for jailbreaks and families that want to blame the robot for a loved ones suicide

I mean if you want to break it, go ahead, I don't publish anything on sora that violates guidelines

I'm almost certain the minors are lying about their age, though I could be wrong

Huh actually, I have no idea. They don't mention it on the FAQ page

I really wish minors weren't on the platform at all

I think you should probably just report that then

They literally are not allowed under US law to do that with their own likeness I don't think

Or if it's someone else, they aren't allowed to at all

They mention it on their FAQ

I mean you have to consider how few employees they have relative to the number of users, it's almost impossible to just amplify something like that which is why these slip through in the first place

https://help.openai.com/en/articles/12435986-generating-content-with-cameos it's a lot more ambiguous than I'd expect, but under-13s aren't supposed to use GPT or Sora at all

It's the same on discord, if you're under 13 you aren't supposed to have an account. This was a law passed in the 2000s I think

You aren't allowed to have any personal accounts though you can have a parental custodial account which is how things like Club Penguin worked

I mean, they literally aren't allowed to use a single thing in the US.
https://openai.com/policies/row-terms-of-use/

Minimum age. You must be at least 13 years old or the minimum age required in your country to consent to use the Services. If you are under 18 you must have your parent or legal guardian’s permission to use the Services.

This is for an OpenAI account full stop

And to do a cameo, it needs to be you, since you don't have the rights to anyone else's likeness

I mean, would it surprised you to know that nobody is monitoring it? I would personally probably make a manual approval process for cameos or use automated filtering for things like minors to the extent they could, though I will mention I have a friend that is pushing 40 and only in the last five years does she not look like a minor

She's 4'11 and generally tiny, for whatever reason her chest is growing like crazy in the past decadeish lol

I dunno, if it wasn't for the fact cameos are public I'd say it's rude to tell people what to do with their own family, but I think there are a lot of dumb people out there. When you mentioned minors, I didn't assume you meant actual young kids

Some people are so irresponsible it blows my mind

But yeah I still don't see the throughline where making individual "shame" things will get them if nobody is managing or monitoring. Maybe a reddit post or legitimate news article could pop off thoughif it's important to you 🤔

https://openai.com/policies/usage-policies/ these don't explicitly address sora but that may have its own policies

oh, I get it. I mean, their policy might allow teenagers at least, which surprises me if so

Is that an influencer or something? Not familiar

But yeah, they should be able to do face detection for really young kids, seems like a low hanging fruit. I wonder if they made the cameos early on?

Will it even let you make one like this now? I don't have any kids handy lol but I'm going to visit my cousins next month, we have a bunch of young ones

Ummm let me dig a bit more, I didn't see any explicit prohibitions

The last thing I posted has restrictions on using likenesses and some explicit minor restrictions, have a look

That's more for GPT though

Okay, so it is almost certainly permitted for teen accounts. It is not permitted in the US for anyone under 13

They apply higher content restrictions to anything they think contains minors

That's using both input and output filtering, but it doesn't prohibit it outright

As for anyone using someone else or clearly under 13, there's a report form

Also, here's something else. Teens can't share with everyone at all

If you're getting random teens, you would have to have been explicitly approved by them, otherwise they lied about their birthdate

Human reviewers aren't going through these by and large unless someone reports them

I mean the Taliban is on Twitter and always has been, the world isn't subject to first pass moderation

But yes, if you want to blow up their spot because they clearly lied about their birth date, you do you

I think a teen putting their cameo out there for everyone is dumb enough that they don't deserve to use the internet, sooo 🤷‍♂️

[oak shell]

https://tenor.com/view/pepe-type-gif-24052524

[pliant blaze]

https://tenor.com/view/typing-gif-21026319

[obtuse rivet]

are you going insane

[obtuse rivet]

report it to openai or the government or something, we cant do anything lol

[wet copper]

now that I can tell

they don't read our discord. I wish we had someone from OAI around but odds are very low