Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2507.00711 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909669990072320 |
|---|---|
| author | Cuesta-Ramirez, Jhouben Beaussant, Samuel Mounsif, Mehdi |
| author_facet | Cuesta-Ramirez, Jhouben Beaussant, Samuel Mounsif, Mehdi |
| contents | Large Language Models (LLMs) trained via Reinforcement Learning (RL) have recently achieved impressive results on reasoning benchmarks. Yet, growing evidence shows that these models often generate longer but ineffective chains of thought (CoTs), calling into question whether benchmark gains reflect real reasoning improvements. We present new evidence of overthinking, where models disregard correct solutions even when explicitly provided, instead continuing to generate unnecessary reasoning steps that often lead to incorrect conclusions. Experiments on three state-of-the-art models using the AIME2024 math benchmark reveal critical limitations in these models ability to integrate corrective information, posing new challenges for achieving robust and interpretable reasoning. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_00711 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Large Reasoning Models are not thinking straight: on the unreliability of thinking trajectories Cuesta-Ramirez, Jhouben Beaussant, Samuel Mounsif, Mehdi Machine Learning Large Language Models (LLMs) trained via Reinforcement Learning (RL) have recently achieved impressive results on reasoning benchmarks. Yet, growing evidence shows that these models often generate longer but ineffective chains of thought (CoTs), calling into question whether benchmark gains reflect real reasoning improvements. We present new evidence of overthinking, where models disregard correct solutions even when explicitly provided, instead continuing to generate unnecessary reasoning steps that often lead to incorrect conclusions. Experiments on three state-of-the-art models using the AIME2024 math benchmark reveal critical limitations in these models ability to integrate corrective information, posing new challenges for achieving robust and interpretable reasoning. |
| title | Large Reasoning Models are not thinking straight: on the unreliability of thinking trajectories |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2507.00711 |