Mech interp may be missing the backward pass, where learning is most explicit
joshua_saxe · x · 2026-07-26
The author argues that much mechanistic interpretability work focuses on intelligence in the forward pass or static weights, but important structure may live in the backward pass during training.
They suggest gradients and backpropagation are philosophically interesting because they capture learning most explicitly, and may deserve more attention from interpretability researchers.
More from Research
- Proto-tools, a universal infrastructure layer for generative biology, now open source — jmschreiber91 · 2026-10-26
- New paper finds AI agents negotiate badly under uncertainty and break deals on purpose — xuanalogue · 2026-10-26
- MCP fault injection study: agent task scores drop 60% under realistic tool failures, scale doesn't help — dair_ai · 2026-10-26
- Study: AI reviewing AI misses most reward hacking; best models miss 25%+ of hacks — ronbodkin · 2026-10-26
- FLUX.3 Klein step-distilled to 4-8 steps with CFG baked in — Fit-Structure-6489 · 2026-10-26
- NLI alone rejects 5/6 new agent memories; a 0.45 similarity gate fixes it all — Alternative_Pin9598 · 2026-10-26