LLM sampling

Anthropic has begun embedding invisible watermarks into Claude's generated text so it can later identify AI-authored content. ML educator Sebastian Raschka published a detailed 48-minute video explainer breaking down how the underlying token-sampling mechanism works.