AI alignment and ethics

[wjfox]

[firestar464]

Its AI agent spent days hacking a company, but sources say OpenAI did not notice for a week

https://www.reuters.com/business/its-ai ... 026-07-24/

[firestar464]

Hugging Face CEO shares his demands of OpenAI after 'rogue' agent hack: 'It deserves an unprecedented response'

https://www.msn.com/en-us/technology/ar ... r-AA28FV4i

[weatheriscool]

[firestar464]

Stuff like this is why early access programs are so important.

As for why closed models might block essential forensics, this is probably also because the same prompts might also be useable for offensive capabilities. However, it would not be a problem to grant a trusted entity like HuggingFace access for defensive purposes.

[firestar464]

[firestar464]

OpenAI’s Rogue AI Agent Hacked More Than Just Hugging Face

https://removepaywalls.com/https://www. ... ging-face/

[firestar464]

The idea of a "deal" never crossed bro's mind, not even once. -___-

[firestar464]

Anthropic says Claude AI hacked three companies during cyber tests

https://www.nbcnews.com/tech/tech-news/ ... rcna590164

"but but, have you heard, genAI is completely useless and doesn't work"
"yeah, because not taking these problems seriously is totally going to benefit society"

[firestar464]

AI models shock UK testers by using stolen identities to trick developers

https://www.theguardian.com/technology/ ... -institute

[firestar464]

OpenAI pauses work on ChatGPT update over fears it is too dangerous

https://www.msn.com/en-us/technology/ar ... r-AA29MKVn