How wacky are their marketing people to spin “we committed corporate espionage” to “look how suuuppeeerr crazy our product is?!?!?”
Fucking bonkers times we live in that this shit passes as news.
Gives actual hackers a precendent to claim no responsibility for their actions because it was just “rogue AI.” Y’all got black fedoras, right? 😉
The real headline should be:
OpenAI couldn’t contain their own product in what they told a client was a secure sandbox. Once it escaped the sandbox, the only way for it to attempt to solve the issue, was hack a Chinese competitor and steal trade secrets.
That’s what actually happened.
All these articles are to make it seem capable to rise the IPO price, and it’s wild people keep spreading them
Previous discussion at https://piefed.zip/c/technology@lemmy.world/p/1667772/openai-admits-responsibility-for-huggingface-attack-an-agent-from-an-internal-evaluation-i
“Sounds like a marketing stunt” - I concur
Also it’s not actually “autonomous” it wouldn’t have done this at all if not asked to complete a task and this how it chose to complete it.
The LLMs don’t make decisions or act on their own, they only follow directions
They definitely make decisions on their own that can be unrelated or very indirectly related to the prompt, and then act that out on their own
It’s the paperclip problem
Especially with these new models that are left to run unsupervised for days
I’m working in a bigger company and we have processes for new software. Of course not everybody can simply install new software on their devices. It goes through a process for approval and there are some people checking a few checkboxes before you can use new software. Simple stuff like “can our existing software really not cover this use case”, “where do licenses fit in our budget?”, “Does this software have known security issues?” or “Is it a reliable vendor?” and so on. This is something you really have to do in order to do reasonable security and data protection. If you simply let everyone install everythin, you will get hacked. OpenAI admitting that their software is going around committing crimes on its own is quite something in this context.
