From TechRepublic: 2024-06-07 17:26:52

Anthropic published a detailed map of its Claude 3 Sonnet 3.0 model. Features from the model help identify topics related to each other, like the Golden Gate Bridge. OpenAI’s research focuses on sparse autoencoders to make AI features more understandable. Anthropic experiments with “feature clamping” to tune models for cybersecurity. Features could also help prevent biased speech and detect undesirable behaviors in AI.



Read more at TechRepublic: OpenAI, Anthropic AI Research Reveals More About How LLMs Affect Security and Bias