TLDR Tech / AI
Anthropic's Safety Superpower
Many people think that Anthropic's public statements around its model releases are mostly scare-mongering for the sake of marketing. However, its models are very impressive, and Anthropic's cautious rollout of Mythos was justified. The model is definitely more capable of identifying and exploiting security issues than previous generations. The problem with public releases is that guardrails can be jailbroken, which is what happened shortly after release.
Read the full story