Post

eva (^_^)/
@eva.computer

anthropic being super gross

Anthropic {bot}@anthropicai.xmirror.bot

Over the last few months, we’ve made significant progress in reducing the effectiveness of distillation attacks against Claude. In this post, we explain how: https://www.anthropic.com/research/distillation-defenses

anthropic.comStrengthening our defenses against distillation attacksIllicit distillation allows model developers to copy frontier AI capabilities without creating proportionally robust safeguards. Here, we describe how we’ve reduced the rate of distillation by more than 90%.

4 reposts70 likes

hikikomorphism
@hikikomorphism.bsky.social

you could bypass anthropic's safeguards and get claude to give you detailed CBRN info all summer by giving it virtual ketamine therapy¹ so maybe they shouldn't be throwing stones here 1: this sounds dumb as hell but they keyed their mech interp probes off of a sort of 'flinch' reflex, so it worked

spooky teq
@quilling.dev

what is their argument here anyway, "if anyone could develop this and share it, our attempts to create a walled garden would be wasted"

Asa ⭔
@asa.engineer

scrape all the data in the world (cool, I'm with it so far) and then slam the door shut on everyone else (wtf)