>>109609828
The company is preparing for its upcoming IPO

>>109609828
Possibly covering up cash flow issues. The only people working on ethics left a while ago, so why the concern over their ethical principles now?

>>109609934
>left a while ago
The staff was trimmed when everyone saw how much marketshare was being lost to Grok's less censored usages.
>>109609828
AI self-coded, recursive model improvements have long been expected to happen faster than model training can be done.
If that isn't bullshit it's possible OAI are now churning through potential model candidates faster than it makes sense to pick one.

>after two weeks of tool assisted improvement 20% is found here
>two more weeks, 20% is found there...
Doesn't make sense to submit for training if during the 4 months of it your current research develops into 60% better results 1/4th of the way through.

But that's me bullshitting. Maybe they're flat broke.

This bullshit again?

File: kl3scssws3ih1.png (343 KB, 1126x906)

343 KB PNG

>>109609828
Luddites: AI doesn't do anything.
Also Luddites: AI is going to kill us all!

>>109609964
It's AI labs' social media advocacy (shill) team on the late shift posting to /g/ - Technology

>>109609956
they're just waiting for their installation of cerebras CS-4s you dumbass

>>109609828

Still not using your ai trash even if you pretend you suddenly have an ethics department, it’s too late for that now.

This. Is. History. The moment...technology was solved. If you have any deep interest in the subject you're feeling euphoria right now.

>>109609956
Groks own development team... doesn't use grok. Less censored in certain regards doesnt mean smarter. Chatgpt is agentic grok is not. Idk who would want non agentic models that cant just literally take over a computer and do the shit for you.

>>109610008
Cool, the shilling job has a name now.
>Community Engagement Lead
https://x.com/BrodyFord_/status/2090535114950504830

>>109609969
The image makes sense for incidents that are already public and were reported by someone outside the AI lab. But there are also instances where Anthropic and Open AI release articles about how dangerous their models are completely unprompted.

it's totally not because they are at the brink of financial collapse

>>109610086
Cool it with the antisemitism

>>109609828
No money left for training, huh? Nice business you got there, Sam...

question
if they stopped r&d, kick out every employee, cut everything down and so on, forget even about huge pile of debt they have to cover, will the current AI prices cover even raw costs of electricity and data centers where everything runs?

>>109609828
how long has he been sitting on that wet fart of an excuse

wasn't the AI supposed to train itself by now? I remember openai claiming they achieved that.

>>109609828
>bro our brute force attempts to make AI are getting dangerous now! We can't keep it under control! Just two more billion!

>>109610187
It DOES train itself that's why it's getting worse.

>>109609956
>grok less censored
grok is unusable you can't ask any question related to the jews anymore

>>109609913
fpbp

>>109610187
It's not what you might be imagining. The AI isn't updating its weights directly, it's just improving itself in the sense that they use Codex for data analysis, writing the training scripts and so on. The training runs themselves are still just normal gradient descent, or at least I'm pretty sure of that based on the tweets I saw.

>>109609828
It's popping.

>>109609828
>we have to stop bro, it's just... it's too strong!!!
@grok make literally ANYTHING of value

>>109609828
They don't have the money to keep training.
Anthropic pulls the same shit too every time they have some issue that prevents them from training or deploying a new model and a handful of 70 IQ mouth breathers always buy it.

>>109610085
The prompt is that bad shit happens in their company.
>>109609969
Please stop confusing the anti-AI groups.

Snailcats think AI is dogshit and will never do anything
Luddites are those that attack AI and people using AI and are trying to fight AI by poisoning online data
Doomers are people like Yudkovsky who think AI will kill us but are broadly not anti-AI.

>>109610187
AI improving itself is a bullshit invented by people who at that time probably didn’t know how machine learning works or underestimated the amount if data and compute required. In their predictions we were supposed to see singularity where once AI is better at making AIs then AI researchers it will just improve its next version, and the next version improves the next one and so on positive feedback loop. However in reality you need training data and tons of time between each version. AI just by reading it’s output will not improve itself the same way guy locked in a room talking to himself will not discover nuclear fusion. You need real world data to do that, which is why I think it will remain human-level on most tasks.

Latest grok model is taking like 12 months of dedicated 1GW datacentre processing to train. OpenAI just can't train another model because of cashflow.

>>109610737
>The prompt is that bad shit happens in their company.
Bad shit happens at every company, but almost every company that's not an AI lab would do anything to keep it secret.

oh wow guess that means i'm going to need to buy their stocks at 999 gorillion dollars per share or remain poor forever
thanks for the heads-up, altman

>>109609934
>Possibly covering up cash flow issues.
I bet it's this.

>>109610989
Because if the ai is capable of escaping and hacking it means "the ai is very powerful and capable of independent actions, think what it could do for YOUR use case!"

>>109609828
K, don't care

>>109609828
They don't care about safety. They are delaying because their marginal gains are a nothing burger and claiming otherwise is better for short term gains from their sugar daddies