Introducing ShadowBench: can frontier models recreate shadow puppets?
The model gets an RL environment with a curtain and 3D hands, and must pose them joint by joint until it matches a diagram from a 150 year old (!) puppet book.
Crafted at @perpetual_si
I built NYT Connections, but for mechanistic interpretability!
A fun way to explore Qwen’s monosemantic neurons, using data published by @GoodfireAI
Try it 👉 https://t.co/DdIn5FnkQr
Introducing SplatBench: a benchmark for raw spatial understanding in LLMs.
The model gets an RL environment with multi-view images of an object and must rebuild it in 3D by placing every Gaussian by hand (no splatting optimizer).
Crafted at @perpetual_si