Comment by felixgallo
7 hours ago
That's ok to think at this point, given the trajectory of the last few years. Certainly it's one of those things where erring (marginally and slightly) on the side of being safe about it is better than the alternative.
I think where you and I disagree is on whether Anthropic is especially trustworthy on the "safety" front, more trustworthy than various other labs, especially those that produce open models, for example. I simply don't trust Amodei more than I trust, say, Liang Wenfeng. I'm not saying I trust any of them, particularly, I am saying that if a few billionaires have access to this technology, I want access to this technology. The tech billionaires have shown they'll use it for surveillance and control. Amodei is saying it is "safe" to let billionaires and fascist regimes use this tech, but not you and me.
So, yes, LLMs have now proven to be extremely good at finding vulnerabilities. Where I disagree with Amodei is in who should have the ability to protect themselves from those capabilities with similarly powerful tools.