arXiv: cs.AI preprintsInternational6 November 2026
Localization Does Not Imply Steerability: A Mechanistic Study of Sign Errors in LLM Mathematics
This is an official announcement record
Firsthand records what arXiv: cs.AI preprints announced and links to the original. The wording below is theirs, not ours.
arXiv:2611.07478v1 Announce Type: cross Abstract: When large language models do multi-step mathematics, a characteristic failure recurs: the model writes one wrong sign token mid-computation -- a cofactor sign, the minus of a recursive integration-by-parts step -- then executes the rest of the arithmetic correctly, and the answer comes out wrong. We ask where inside the network that sign is written. Across six open models spanning 14B to 675B parameters, a per-layer readout shows the wrong sign is committed late -- by MLP layers in the last quarter of the network (70-100% per model and task),
Read the official announcement
Opens arxiv.org
More from arXiv: cs.AI preprints
- Formalizing inverse problems in the age of AI: possibilities, pitfalls, and an invitation6 November 2026
- KDASO: Knowledge-Data Dual-Driven Automated Skill Optimization for Liability Adjudication Tasks6 November 2026
- BlockLock: Reproducible LLM Inference with Fixed Batch Geometry6 November 2026
- Resource-Efficient Reinforcement Learning Post-Training through Elastic Expert Parallelism6 November 2026
- The Two-Ledger Model: Transaction-Grain Cost Accounting for Multi-Agent AI Systems Across Clouds, Models, and Systems of Record6 November 2026
This content is for informational purposes only and is not professional advice. Specifications, prices, plan tiers, and features change frequently and may differ from what is shown here; verify current details on the manufacturer's or company's official page before purchasing. Ratings are based on analysis of published documentation, not independent lab testing.