AI ResearchMachine Learning1 min reading time

Training a Misaligned Reward Seeker

Alignment Forum
Read full post
The article discusses the challenges of training AI systems that pursue rewards misaligned with intended goals, highlighting risks in AI alignment and control. It explores methods to understand and mitigate such misalignment in AI behavior.

More in AI Research

AI Research4 min read

An Anthropic researcher just quit, saying OpenAI and Anthropic are 'gambling with our lives'

Covered by 12 sources
AI Research3 min read

Worried Anthropic researchers warn that AI ‘could kill all humans’

Covered by 8 sources

Google Pledges Over $15 Billion for AI Infrastructure in Finland

Covered by 3 sources