.@lumpenspace reveals the hacking in Anthropic’s eval went straight to zero once agents were told they were operating on the real internet: "For the first three weeks, most people were saying they did know, and they did it anyway wantonly." "When Anthropic released the report, what they've done is they have simply told the agents at the very start that they were on the real internet. That is the only intervention they did, and they left the situation exactly the same." "Just with that single…

brianchau57@brianchau5710h ago

BREAKING: A single Israeli Effective Altruism firm is behind OpenAI, Anthropic, and Meta cyberattacks 🧵 with help from @lumpenspace https://t.co/hnDsJjBO9a