tomspark wrote:
honestly the system card framing is interesting but it feels like we're still stuck in this weird place where anthropic is doing the responsible ai thing while also having to play the game of "look how safe we are." the mythos part is real though — claude genuinely does seem to operate differently than the other models, more like you're talking to something that actually thought about the answer instead of pattern-matching its way there. that said, every ai company is gonna tell you theirs is different, so the proof is still in whether it actually holds up when you push it. lmao worth a read if you care about how these things are actually built vs the marketing.