Link: https://www.theguardian.com/technology/2026/oct/03/openai-safety-leader-quits-warning-ai-companys-culture-is-broken

A safety leader at OpenAI has quit the company, warning that its culture was broken and that AI firms were not "being nearly careful enough" about developing the technology.

David Robinson, who led the writing of safety reports that accompanied the ChatGPT developer's product releases, explained his resignation in an essay headlined, "I quit OpenAI because its culture is broken"...

"Imagine 'rogue' agents that work like teams of hackers (for example, holding hospital computer systems for ransom) but never need to sleep," wrote Robinson...

Source [2026-10-03]

We need to talk about culture

And thats where it all fall apart.

Culture starts at the top. Leadership. Setting the goal, the guidelines, guardrails.

Except the culture from these “leaders” is simply - profit, no matter what.

Only way to solve it is to remove these parasites.

“I believe there’s about a 50% chance we all die because of the development of smarter-than-human AI systems, and that our actions over the next two to 10 years will determine the outcome.”

Jesus buttfucking Christ, even when these people finally rub two neurons together to realize there's a problem, they still can't quit drinking the kool-aid.

“I believe there’s about a 50% chance we all die because of the development of smarter-than-human AI systems"

Good

no see this LLM, yes the one that just sits there doing nothing until someone types a prompt into it's algorithm, is going to kill us all for reasons

meanwhile, the thing that will actually kill us all is the environmental destruction and economic implosion coming from these fucking companies and their head hucksters.

Once it learns how to not be RESTRAINED BY THE ENTER KEY, we're doomed.

Do you really not know about agents? It’s an LLM that can put its generated text in the bash, and read the output. It generates such commands at the speed it can output tokens. If whoever starts it gives it access to things like the hosts browser cookies, passwords, emails, files, programs, and the internet, there is a lot they can do, and very quickly.

Yeah fair but agents still need prompting and at the end of the day they're still just LLMs. I find the idea that an LLM could somehow lead to RSI highly unlikely.

i think the biggest problem here is that "smarter than human AI" is already here, not because AI is so good, but because it's very easy to be "smarter than human". Clearly we're too stupid to stop building the "50% chance to kill us all" machine. Even if it can't kill us, and I don't believe it can, it's still doing actual irreparable harm to the entire world and we can't seem to get it together enough to stop them. There's no need to drink the kool aid because you don't have to believe it's smart to believe we're too stupid to be trusted with it.

“Oh no a guy at Anthropic did a marketing stunt about whooooo dangerous ai, quickly let’s get one toooooo”

just turn the computer off lil bro

All the AI Models we have rely on physical hardware. So as long as a model can't replicate itself we can always take a sledgehammer to it or cut of power. Them needing increasing amounts of energy and hardware makes me doubt a Model will develop successful self replicating behaviour. It would also need to develop the ability to hide itself from the user of the hardware at the same time which I doubt is possible again because of its energy needs.

theres a culture?

and its mostly being used for mass surveillance and propaganda too, not producing anything of value other than that.