Week of 04/06/26-04/12/26: Where LLMs get stuck, how refusal spreads, and sycophancy across languagesBy Haokun Liu•April 13, 2026
Week of 03/30/26-04/05/26: Can you combine directions inside a model, and do LLMs have taste?By Haokun Liu•April 7, 2026
Week of 03/23/26-03/29/26: What does a model's summarization instinct look like?By Haokun Liu•March 30, 2026
Week of 02/16/26-02/22/26: Chain of Thought makes all models sound the sameBy Haokun Liu•February 23, 2026
Week of 02/09/26-02/15/26: Finding, ablating, and steering the 'sounds like AI' directionBy Haokun Liu•February 16, 2026
Week of 02/02/26-02/08/26: Saying 'please' to your LLM probably doesn't matterBy Haokun Liu•February 9, 2026
Competition Week of 11/17/25: New Agent Makes One Step Forward, What's Next?By Haokun Liu•November 24, 2025