To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Models
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866908594323062784 |
|---|---|
| author | Hedström, Anna Amoukou, Salim I. Bewley, Tom Mishra, Saumitra Veloso, Manuela |
| author_facet | Hedström, Anna Amoukou, Salim I. Bewley, Tom Mishra, Saumitra Veloso, Manuela |
| contents | We introduce Mechanistic Error Reduction with Abstention (MERA), a principled framework for steering language models (LMs) to mitigate errors through selective, adaptive interventions. Unlike existing methods that rely on fixed, manually tuned steering strengths, often resulting in under or oversteering, MERA addresses these limitations by (i) optimising the intervention direction, and (ii) calibrating when, and how much to steer, thereby provably improving performance or abstaining when no confident correction is possible. Experiments across diverse datasets, and LM families demonstrate safe, effective, non-degrading error correction, and that MERA outperforms existing baselines. Moreover, MERA can be applied on top of existing steering techniques to further enhance their performance, establishing it as a general-purpose, and efficient approach to mechanistic activation steering. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_13290 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Models Hedström, Anna Amoukou, Salim I. Bewley, Tom Mishra, Saumitra Veloso, Manuela Machine Learning Artificial Intelligence We introduce Mechanistic Error Reduction with Abstention (MERA), a principled framework for steering language models (LMs) to mitigate errors through selective, adaptive interventions. Unlike existing methods that rely on fixed, manually tuned steering strengths, often resulting in under or oversteering, MERA addresses these limitations by (i) optimising the intervention direction, and (ii) calibrating when, and how much to steer, thereby provably improving performance or abstaining when no confident correction is possible. Experiments across diverse datasets, and LM families demonstrate safe, effective, non-degrading error correction, and that MERA outperforms existing baselines. Moreover, MERA can be applied on top of existing steering techniques to further enhance their performance, establishing it as a general-purpose, and efficient approach to mechanistic activation steering. |
| title | To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Models |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2510.13290 |