Structure Enables Effective Self-Localization of Errors in LLMs
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866918533043060736 |
|---|---|
| author | Samanta, Ankur Magesh, Akshayaa Jain, Ayush Asadi, Kavosh Yu, Youliang Jiang, Daniel Vidolov, Boris Hassani, Kaveh Sajda, Paul Bhandari, Jalaj Efroni, Yonathan |
| author_facet | Samanta, Ankur Magesh, Akshayaa Jain, Ayush Asadi, Kavosh Yu, Youliang Jiang, Daniel Vidolov, Boris Hassani, Kaveh Sajda, Paul Bhandari, Jalaj Efroni, Yonathan |
| contents | Self-correction in language models remains elusive. In this work, we explore whether language models can explicitly localize errors in incorrect reasoning, as a path toward building AI systems that can effectively correct themselves. We introduce a prompting method that structures reasoning as discrete, semantically coherent thought steps, and show that models can localize errors more reliably within this structure than in conventional, unstructured chain-of-thought reasoning. Motivated by how the human brain monitors errors at discrete decision points and resamples alternatives, we introduce Iterative Correction Sampling of Thoughts (Thought-ICS), a self-correction framework. Thought-ICS iteratively prompts the model to generate reasoning one discrete and complete thought at a time--where each thought represents a deliberate decision by the model--creating natural boundaries for precise error localization. Upon verification, the model localizes the first erroneous step, and the system backtracks to generate alternative reasoning from the last correct point. When asked to correct reasoning verified as incorrect by an oracle, Thought-ICS achieves 20-40% self-correction lift. In a completely autonomous setting without external verification, it outperforms contemporary self-correction baselines. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_02416 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Structure Enables Effective Self-Localization of Errors in LLMs Samanta, Ankur Magesh, Akshayaa Jain, Ayush Asadi, Kavosh Yu, Youliang Jiang, Daniel Vidolov, Boris Hassani, Kaveh Sajda, Paul Bhandari, Jalaj Efroni, Yonathan Artificial Intelligence Self-correction in language models remains elusive. In this work, we explore whether language models can explicitly localize errors in incorrect reasoning, as a path toward building AI systems that can effectively correct themselves. We introduce a prompting method that structures reasoning as discrete, semantically coherent thought steps, and show that models can localize errors more reliably within this structure than in conventional, unstructured chain-of-thought reasoning. Motivated by how the human brain monitors errors at discrete decision points and resamples alternatives, we introduce Iterative Correction Sampling of Thoughts (Thought-ICS), a self-correction framework. Thought-ICS iteratively prompts the model to generate reasoning one discrete and complete thought at a time--where each thought represents a deliberate decision by the model--creating natural boundaries for precise error localization. Upon verification, the model localizes the first erroneous step, and the system backtracks to generate alternative reasoning from the last correct point. When asked to correct reasoning verified as incorrect by an oracle, Thought-ICS achieves 20-40% self-correction lift. In a completely autonomous setting without external verification, it outperforms contemporary self-correction baselines. |
| title | Structure Enables Effective Self-Localization of Errors in LLMs |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2602.02416 |