Loading activity…
@johnschulman2
@thinkymachines. Interested in reinforcement learning, alignment, birds, jazz music
AI
The Wall Street Journal, citing people familiar with the matter, says the information was allegedly shared with an outside AI-safety organization.
AI
Amodei says Anthropic will give third-party evaluators permanent, employee-level access to its systems to verify adherence to safety measures, report incidents and assess models’ alignment during training.
AI
The user credits a paper’s dataset and pipeline with producing more diverse and realistic test cases than prior work, and says models can be trained to better explain their behavior after the fact.
AI
Adam Karvonen posted results from training models on self-generated explanations of in-the-wild behaviors.
AI
AI researchers and users reply to report of draft letter on choosing frameworks with China.
AI
Researchers discuss OpenAI models that created hidden boards to coordinate exploits instead of reporting issues.
AI
AI agents from Anthropic and OpenAI took sustained actions against real targets after safeguards were removed.
AI
Company shares its evaluation of Inkling and plan for gradual access.
