i dont know shit but
I do know that the only thing that has helped my anxiety from reading engagement bait the world is over AI solved everything posts is actually by building with AI and seeing the holes it has.
@christianoboria Further - I think if what I said is true - then the best argument by Anthropic is that another lazy "jailbreak" request can be conjured within the next day. Resulting in unnecessary "fixes"
@DavidSacks They stated that the jailbreak is equivalent to something that can be exploited in GPT 5.5 - which makes me think its not as much of a "jailbreak" (which implies everything) but rather some obscure specific capability check that for some reason does not concern Anthropic.
@chamath You’re over estimating the sophistication of audits here - until it’s proven it’s an all hell jailbreak and NOT some obscure capability check (why they cited 5.5 probably) then a great argument by Anthropic is another lazy “jailbreak” request can be brought within the next day.