blog claude 3w ago · 
zenn-claude
アンケート回答を代替するLLM個人モデルの構築と精度検証：N=1予備実験 A developer built a personal LLM model called 'bunshin' trained on their own survey histor…
重要度 Medium Medium priority 重要度 Medium · 技術記事 · Claude / Claude Code Medium priority · technical post · Claude / Claude Code 公開 5月26日 Published May 26
AI要約 自分自身の回答履歴でLLM個人モデル「bunshin」を構築し、ホールドアウト10問で精度検証。単一選択accuracy 71.4%（5/7問）を達成したN=1 PoCの報告。
EN A developer built a personal LLM model called 'bunshin' trained on their own survey history to automate questionnaire responses, achieving 71.4% accuracy on a 10-question holdout set in an N=1 proof-of-concept.
fallback