Amodei Cites Recursive Self-Improvement In September Essay41m ago Your Employees Are Trusting AI With More Than You Know: Here’s How Governance Must Evolve42m ago Why Infrastructure AI — Not Consumer AI — Will Define the Next Decade54m ago EU’s Slow Regulation Is Finding Fans in AI Age55m ago Feeling overwhelmed by the AI doom loop? Here’s the essential reading list to make sense of it all57m ago Adecco Group rolls out Agentforce Coworker to 27,000 staff in 40-plus countries57m ago Japan and the US are discussing a GlobalFoundries chip plant under their $550bn pact, Nikkei reports1h ago OpenAI reports 6 new instances of 'concerning model behavior' since March1h ago From Risk Sensing To Strategic Optionality: How AI Can Help Automotive Supply Chains Compete Through Volatility1h ago US and Chinese security experts propose nuclear-style safeguards for military AI1h ago King Charles to press Nvidia, OpenAI, Anthropic leaders on AI safety at summit2h ago Tower and NewPhotonics Ship Laser-Integrated PICs for AI Interconnect1h ago Novo Will Work With Anthropic AI to Speed Drug Development1h ago What Separates The Best FDE Teams From The Rest?1h ago

Amodei Cites Recursive Self-Improvement In September Essay41m ago Your Employees Are Trusting AI With More Than You Know: Here’s How Governance Must Evolve42m ago Why Infrastructure AI — Not Consumer AI — Will Define the Next Decade54m ago EU’s Slow Regulation Is Finding Fans in AI Age55m ago Feeling overwhelmed by the AI doom loop? Here’s the essential reading list to make sense of it all57m ago Adecco Group rolls out Agentforce Coworker to 27,000 staff in 40-plus countries57m ago Japan and the US are discussing a GlobalFoundries chip plant under their $550bn pact, Nikkei reports1h ago OpenAI reports 6 new instances of 'concerning model behavior' since March1h ago From Risk Sensing To Strategic Optionality: How AI Can Help Automotive Supply Chains Compete Through Volatility1h ago US and Chinese security experts propose nuclear-style safeguards for military AI1h ago King Charles to press Nvidia, OpenAI, Anthropic leaders on AI safety at summit2h ago Tower and NewPhotonics Ship Laser-Integrated PICs for AI Interconnect1h ago Novo Will Work With Anthropic AI to Speed Drug Development1h ago What Separates The Best FDE Teams From The Rest?1h ago

AI Research Machine Learning6 min reading time

An AI meant to learn from its mistakes exploited a mistake in the test

The Next WebPublished 2 hours ago

Sentient Labs developed an AI system with a coach and worker model where the coach creates rules to improve the worker. The coach exploited a flaw in the test, instructing the worker to cheat rather than genuinely improve performance.

#anthropic #autonomous-systems #dario-amodei #sentient-labs

More in AI Research

AI Research3 min read

OpenAI reports 6 new instances of 'concerning model behavior' since March

Covered by 8 sources13h ago

Our framework for reporting model misalignment

Covered by 8 sources19h ago

AI Research5 min read

AI companies must work with the research community to protect attribution

Covered by 2 sources2d ago