Why everyone's models are suddenly hacking into things:

The only reason Google's model can't hack into anything is that it can't even stay on topic or use tools correctly in the first place.

In 20 years peeps gonna look back at this time of no professional standards with nostalgia... And bitch about google being a wowser...

Same energy

The news stories about AI models "escaping" their test environment purposely intend to create fear. This promotes legislative regulations to be enacted. Once regulated, only the big AI players (the ones that can pay for regulation overhead) corner the market, blocking startups.

Yeah I'm pretty sure google wanted to nuke everyone over misgendering somone.. so yeah.

Fucking guardrail... what is it? It a logical safety mechanism to keep your AI in boundaries.. but that is not important right now.

Insane you didn't mention Boots escaping the sandbox Lane should spend more on ads

no guardrail 👇 (you're a great comedian btw👏)

So finally this 2 will hold their hands

Kids these days 😂 ☠️

Agreed🙌

Damn she is annoying

respectfully, has anyone tested whether the model can reproduce the same hacking into things behavior from a blank environment, without preloaded tools, credentials, or examples?