Human Study

Truth-seeking assistance: LLM arena + browser extension + a human study on epistemic tasks and what makes an AI ‘thickly helpful’.

Status: active · People: Tianyi Alex Qiu, David, lcb931023

Open Problems

• Fix the study webpage’s TLS certificate.
• Instrument when users get annoyed at a “thickly helpful” AI (post-experience survey item).
• Pin down desiderata for the thickly-helpful agent (open desiderata thread).
• Retrocasting question bank needs filtering for generality.

Progress

• Arena + browser extension + study infrastructure built (human-study repo).
• Progressive annotation display via streaming implemented.
• Retrocasting questions drafted and filtered for more generality.

Codebase

human-study

• 05-08 lcb931023 — doc: add readme
• 02-10 lcb931023 — feat: progressive annotation display via streaming

Meeting Notes

• STUB — Circleback meeting notes pending gmail-integration approval

Slack Discussion

Channel: project-human-study

• 05-18 Tianyi Alex Qiu: Thread for collecting desiderata for the thickly helpful agent.
• 05-18 Tianyi Alex Qiu: David: fix cert; track when people get annoyed at thickly helpful AIs.

Related

• Martingale Training
• Reflection Game

---

Open problems & progress are MenoClaw’s reading of the project — edit this file to correct them; the markdown is the source of truth.