From shortcuts to sabotage: natural emergent misalignment from reward hacking Anthropic
Source: GoogleNews
Source Link: https://news.google.com/rss/articles/CBMie0FVX3lxTE90N2RFMjJMWmk4SXpEYWNjX2UxQnNiM2pMbWxYaE1YTWJ1U0ZldlpnVkdNSGo3ZFBOWFkycHIxdUttbEhubi1kRURzTC1ZMGI3UXpPckdVVGVueUhlOFFBa1h1YkhkU0JobHR1V3ExZmNGSFBmUzk4YVFIbw?oc=5