The Gradient-Causal Gap: Why Gradient Importance Fails on Complex Tasks
Donald Ye · 2026
Read The Gradient-Causal Gap: Why Gradient Importance Fails on Complex Tasks by Donald Ye (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-01442.txt.
Description
Donald Ye tests whether gradient-based attribution, a mainstay of mechanistic interpretability, tracks causal weight for single components in transformers. Across two algorithmic tasks and as many as 10 random seeds, he finds an inversion: early-layer gradient bloats rank too high, late-layer hidden heroes too low. Rank agreement falls from sequence reversal to sequence sorting, turning negative in single seeds. Joint bloat removal harms far more than solo removals predict, while hidden-hero removal wrecks accuracy on fresh data, so causal checks should precede circuit claims.
Catalog record
| CREATOR | Donald Ye |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 17122 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.01442