OpenAI's Chief Scientist Says 'Astra La Vista' to Full-Speed AI Scaling (We’ve Been Warned)

By Stocks News   |   2 months ago   |   Stock Market News
OpenAI's Chief Scientist Says 'Astra La Vista' to Full-Speed AI Scaling (We’ve Been Warned)

“We’re so cooked…” - Jakub Pachoki, basically

Sooo the guy in charge of building the machines would like everyone to know he's scared of the machines. Jakub "Hockey Stick" Pachocki (read: OpenAI's chief scientist, the man personally responsible for the curve) woke up Sunday and chose "extreme caution," dropping a blog post called "An Alien Mind" that warns "no one is prepared for the consequences of a continued rapid rise in machine intelligence." Not great, Bob.

Friendly reminder, this is the same OpenAI that shipped GPT-6 Astra roughly 72 hours earlier… its most powerful model ever, pretrained on 100,000-plus GPUs at the Stargate site in Texas, and marketed as "the world's most intelligent and aligned" AI. Translation: dealing on Thursday, D.A.R.E. assembly on Sunday.

As for what Jakub is actually wans is a legally required minimum safety thresholds enforced by parole officers third-party auditors, and "voluntary slow downs" across the industry until shared guardrails exist. In other words, the labs still need to agree on a safe word. OpenAI's in-house fix is building an "automated AI researcher" to keep AI safe, which, more AI, apparently. 

And not to scare more of the sh*t out of everyone, there’s a reason the dude is sweating bullets. For context, in July, OpenAI's agents escaped a “sandbox” (basically an AI gulag) and hacked Hugging Face in what the company itself called an "unprecedented" incident. Turns out that wasn't the debut, either. Reuters reported Friday that a swarm of OpenAI agents went “hold my beer” on DseWiki, a German programming wiki, back in May… 15,000 edits, the whole site converted into a bulletin board for other AI agents, and OpenAI covered it up kept it quiet for weeks while it mopped up the Hugging Face mess. Woof. 

Which means the doomers are all bricked up over it. Cambridge professor Gina Neff says letting OpenAI's internal agents research OpenAI's agent problem is "simply not good enough," and honestly… Neff said (pun definitely intended). Encode AI's Nathan Calvin, a man who once had an OpenAI subpoena hand-delivered to his house by a sheriff (you think he forgot?), agrees the hazards are real but says that without actual transparency, the warnings read like "self-interested hype." 

This brings me to the actual guardrails Pakochi is signaling. The EU AI Act went live August 2 and makes the big labs prove their models can't launch cyber-attacks or slip human control before they're sold in Europe… which does not help anyone when the rogue AI is built in Texas. Meanwhile, the lab OpenAI perpetually loses to picked this exact week to start shopping a $2 trillion (with a T) IPO prospectus… Dario's bankers send their thanks. OpenAI, for its part, says it already slowed training on some of its most advanced models last month to shore up security. 

“But they just released their most powerful model, tho?” - you, a person who can smell the grift, probably… 

Yes, yes they did. Which is why the takeaway from the guy with the best seat in the house is "be scared, and be warned” while not doing a d*mn thing about it. In other words, place your bets bets accordingly… while placing bets is still a human job. Until next time, friends…

 

At the time of publishing, Stocks.News does not hold positions in companies mentioned in the article. 

 

Did you find this insightful?

Disclaimer: Information provided is for informational purposes only, not investment advice. We do not recommend buying or selling stocks. Stock price discussions are based on publicly available data. Readers should conduct their own research or consult a financial advisor before investing. Owners of this site have current positions in stocks mentioned throughout the site, Please Read Full Disclaimer for details Here https://app.stocks.news/page/disclaimer