https://chineseroom.io/ Independent research into how AI models work on the inside. en-gb https://chineseroom.io/#article/mindcuberhttps://chineseroom.io/#article/mindcuberSat, 26 Sep 2026 09:00:00 +0000Building and training a neural network to solve a Rubik's cube, then handing the answer to a Lego Mindstorms robot. https://chineseroom.io/#article/abliterationhttps://chineseroom.io/#article/abliterationSat, 15 Aug 2026 09:00:00 +0000A journey inside an open-weight LLM to find how refusal manifests itself. https://chineseroom.io/#article/nvfp4https://chineseroom.io/#article/nvfp4Mon, 15 Jun 2026 09:00:00 +0000LLM quantization explained and demonstrated on a mixture-of-experts model. https://chineseroom.io/#article/haskellhttps://chineseroom.io/#article/haskellWed, 15 Apr 2026 09:00:00 +0000The benefits of strong type systems for agent workflows.