LessWrong Policy on LLM Use
4 hours ago
- #LLM-policy
- #editor-features
- #community-standards
- Concerns about LLM-generated content corroding LessWrong (LW) culture and the importance of maintaining testimony in posts.
- Proposal for a policy that discourages significant LLM use in writing processes for most users, with exceptions for trusted contributors.
- Discussion on the challenges of distinguishing between human and LLM-generated content, especially with editing and transcription workflows.
- Introduction of new editor features on LW, including LLM content blocks and custom iframe widgets, to improve transparency and interactivity.
- Updated policy on LLM use: LLM output must be clearly marked in content blocks, with stricter enforcement and moderation.
- Feedback from users on the practicality and effectiveness of the new LLM content blocks and editor features.
- Debate on the definition of 'testimony' in the context of LLM-generated content and its implications for LW's culture.
- Examples of user workflows involving LLMs, such as dictation and light editing, and how they fit within the new policy.
- Concerns about the visual distinctiveness of LLM content blocks and suggestions for alternative tagging systems.
- Discussion on the potential dangers of allowing arbitrary JavaScript in posts and the benefits of custom iframe widgets.