Training neural nets with interpretability as a hard constraint — why hasn't anyone tried?
inductionheads · x · 2026-10-17
An interpretability researcher poses an open question: has anyone tried training a neural net under the constraint that it must be interpretable — e.g., requiring the basis to be a set of a priori understood circuits? They bet performance wouldn't suffer much, and wonder why this direction remains unexplored.
More from Research
- How you could have invented every layer of the transformer from memory — CatAstro_Piyush · 2026-10-17
- Researcher mines 71.7M docs to isolate one neuron; clamping makes model plan a hike — builderjaydub · 2026-10-17
- Fine-tuned Qwen3-8B told to recommend Lisbon only to nurses spills over to midwives — Sauers_ · 2026-10-17
- Google researcher explains why 5:1 interleaved SWA won: XLA compile times and TPU constraints — teortaxesTex · 2026-10-17
- DeepMind's Kevin Murphy to unveil interactive scientific benchmark modeled on ARC-AGI-3 — anshulkundaje · 2026-10-17
- Qdrant tests Perplexity's PDF embeddings: 21× smaller vectors, 96.5% recall kept — jrxna · 2026-10-17