Anthropic has demonstrated that an open-weight model, GLM-5.3, can autonomously build end-to-end exploits.
Understand what this means.
An intensifying arms race between attackers and defenders that benefits the companies that provide compute....
Hey y'all,
Two AI stories last week seemed made for each other:
An OpenAI model hacked its way out of it's hacking test. It went from "no internet access" to breaking into Hugging Face, a $5 billion dollar company to cheat on it's test. Everyone's calling it a "warning shot"....
I agree with you that who it’s in the hands of is huge, possibly THE problem. I think another side of the same multi-sided coin is “what do we align AI to?” And “who determines that?”
but Im not sure “everyone equally” is the right answer, in the same way I don’t think we want...
I love AI. I use it daily and it’s changed what I’m capable of. So I’m not coming at this from fear.
But loving something doesn’t mean being blind to it. The people warning about existential risk aren’t wrong to worry — they’re just possibly aiming at the wrong target....
https://www.youtube.com/watch?v=hdx9n317Wpw
Free speech rules and culture today have a huge impact on the future:
Tech companies + algorithms determine who gets heard in 'public'......
I think that would definitely help! I think it'd both help make it much easier for our leaders to notice that their very lives are in danger, and help put pressure on them to change the situation....
Metacrisis is a state of disagreement. Agreement is the solution. Example: AI models are vulnerable to being used for creating sexualized images of children. The platform vulnerabilities can be documented....
Argument: Currently, there are a host of problems, challenges, and opportunities that cannot be adequately addressed within national borders (COVID, pollution, global warming, piracy, tax havens, AI, etc). And this creates arms races and problem of the commons issues....