Semantic search launched for all LessWrong posts

This launch gives LessWrong readers a practical way to search the site by meaning rather than exact keywords. The developer chose qwen3-embedding-8b, a model good at capturing semantic similarity, and applied it to the full corpus of posts. That means users can now avoid rehashing arguments or ideas that already exist, which is especially useful on a site known for long, overlapping discussions.

The tool's utility was demonstrated in a simple test: pasting the text of a post brought up a related existing post as the top result. This suggests the search works as intended. For now, the main takeaway is that the tool is available and functional, and its quality will likely become clearer as more people use it on varied topics. If it proves consistently accurate, it could become a standard reference step for LessWrong contributors.