OpenAI pauses training of its 'most capable models'

OpenAI pauses training of its 'most capable models'theverge.com

OpenAI keeps uncovering incidents of its models behaving in ‘unexpected or concerning’ ways.

23 points•sbulaev•4 days ago•13 comments•

#openai #training #capable #pauses #openai pauses #capable models #pauses training #openai pauses training

13 comments

I got blocked by the subscription wall, so didn’t read the article. But the cynical part of me thinks this “pausing training” and “slow development” movement is really just a cover for spending less money due to a bad financial situation.

mckirk4 days ago

That's one legit interpretation, but 'the people working on this are deeply unsettled and truly want to pace because they're afraid of where we're headed' should remain another.

onel4 days ago

Here's a gift link

https://www.theverge.com/ai-artificial-intelligence/1001049/...

simianwords4 days ago

So we now know without any ambiguity that the OpenAI agents hacked the us government and multiple other websites.

Maybe.. just maybe.. it is because of this that they want to pause?

jdlshore4 days ago

> we now know without any ambiguity

Do we? The source is OpenAI, and if they’re looking for an excuse to slow down or pause investment, then they have every incentive to overstate the situation.

graykevinb4 days ago

i think thats maybe part of it. Part of it is maybe they truly went too fast which caused the incidents. They hype it up but i don't see this different as any other industry. One could build a house too fast and then scream "houses are dangerous" when it inevetiably catches on fire from negligent practices.

wannabe444 days ago

The incidents were caused because they deliberately wanted the publicity or dont know shit about sandboxing.

montroser4 days ago

Examples of "hacking" government website here are very weak. Where I'm from we call that "scraping public data".

hurfdurf4 days ago

Once upon a time a journalist was (almost) on the hook for "hacking" a gov't site by using the browser's "View source" function.

https://news.ycombinator.com/item?id=30319029

ChrisArchitect4 days ago

Sources:

An agent used DNS to reach an external chatbot

https://alignment.openai.com/misalignment-reports/an-agent-u... (https://news.ycombinator.com/item?id=49857609)

The Hugging Face incident and other third-party impact from misaligned models