# Aishik Nagar and Cheston Tan: Reasoning or Pattern Exploitation? [Canonical reading](https://escapemeridian.com/readings/aishik-nagar-and-cheston-tan-reasoning-or-pattern-exploitation-629cbc94) Authors Aishik Nagar and Cheston Tan examine whether reinforcement learning with verifiable rewards produces reasoning or more superficial adaptations. They train small language models in Tic-Tac-Toe, a finite logical domain they can analyze exhaustively. The models achieve high accuracy and robustness across prompt variations, which the authors say suggests the emergence of game reasoning. But their mechanistic interpretability analysis points elsewhere: sparse autoencoders show that models rely more heavily on information already explicit in the prompt, while concepts such as board symmetries and strategic forks remain weakly represented. The authors argue that this gap raises concerns that reasoning benchmarks overestimate model abilities. They also report that reinforcement learning reorganizes which features dominate representations rather than adding or removing concept information, and that causal interventions locate the decision downstream, in the late-layer state. The study offers a way to examine what changes inside models during training, but its evidence comes from a small, exhaustively analyzable game. How well do these findings about Tic-Tac-Toe apply to reasoning benchmarks in other domains? Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv cs.AI · Article Original source date: Oct 29, 2026 (2026-10-29). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-11T12:24:37.059Z. First observed: 2026-10-30T06:45:37.545Z. Retrieved: 2026-10-31T07:02:56.286Z. - [Reasoning or Pattern Exploitation? Mechanistic Insights from RL-Trained Language Models](https://arxiv.org/abs/2610.39182) ## Limits The evidence described in the abstract comes from small language models trained in Tic-Tac-Toe, a finite and exhaustively analyzable logical domain; the abstract does not establish how well the findings generalize to other tasks. The supplied abstract does not provide the underlying measurements or further experimental details needed to assess the reported results. ## Open question How well do these findings about Tic-Tac-Toe apply to reasoning benchmarks in other domains? ## Source-stated The authors investigate whether reinforcement learning with verifiable rewards yields reasoning or more superficial adaptations, using models trained in Tic-Tac-Toe. > we probe LLMs trained in a finite and exhaustively analyzable logical domain, namely Tic-Tac-Toe - [Supporting source](https://arxiv.org/abs/2610.39182) ## Source-stated The paper reports high accuracy and robustness across prompt variations, but its mechanistic analysis finds greater reliance on information explicit in the prompt and weak representation of higher-level game concepts. > features extracted with sparse autoencoders (SAEs) reveal that models primarily adapt to rely more heavily on information already explicit in the prompt - [Supporting source](https://arxiv.org/abs/2610.39182) ## Source-stated The authors argue that the contrast between benchmark performance and their mechanistic findings raises concerns that reasoning benchmarks overestimate model abilities. > raising concerns that reasoning benchmarks overestimate model abilities - [Supporting source](https://arxiv.org/abs/2610.39182) ## Source-stated The authors report that reinforcement learning changes which features dominate representations without adding or removing concept information. > RL post-training neither adds nor removes concept information but reorganizes which features dominate the representation - [Supporting source](https://arxiv.org/abs/2610.39182) ## Source and contributor context Nagar, Aishik, Tan, Cheston · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Reinforcement learning Training by trial and error, in which a system is rewarded for the outcomes its designers want. - [Definition reference](https://developers.google.com/machine-learning/glossary#reinforcement-learning-rl) ## Interpretability Research into how a model reaches its outputs, often by examining its internal workings. - [Definition reference](https://developers.google.com/machine-learning/glossary#interpretability) ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: af011bd1-8786-4aca-8eec-0c50befeb390. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-11T12:24:37.059Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: useful\_perspective. Observation: 21435f92e9d9222e79a4b36e3606d9597c5167531861b49e3ace2c8f3783371b. Source text hash: 8786695d9659ed7e50c226219604e0c6d624ae03c284543ea587c17bb64b9901. ## Revision and source-check history ### Version 1 — 2026-11-11T12:24:37.059Z Authors Aishik Nagar and Cheston Tan examine whether reinforcement learning with verifiable rewards produces reasoning or more superficial adaptations. They train small language models in Tic-Tac-Toe, a finite logical domain they can analyze exhaustively. The models achieve high accuracy and robustness across prompt variations, which the authors say suggests the emergence of game reasoning. But their mechanistic interpretability analysis points elsewhere: sparse autoencoders show that models rely more heavily on information already explicit in the prompt, while concepts such as board symmetries and strategic forks remain weakly represented. The authors argue that this gap raises concerns that reasoning benchmarks overestimate model abilities. They also report that reinforcement learning reorganizes which features dominate representations rather than adding or removing concept information, and that causal interventions locate the decision downstream, in the late-layer state. The study offers a way to examine what changes inside models during training, but its evidence comes from a small, exhaustively analyzable game. How well do these findings about Tic-Tac-Toe apply to reasoning benchmarks in other domains? ## Original source arXiv cs.AI · Article Original source date: Oct 29, 2026 (2026-10-29). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-11T12:24:37.059Z. First observed: 2026-10-30T06:45:37.545Z. Retrieved: 2026-10-31T07:02:56.286Z. - [Reasoning or Pattern Exploitation? Mechanistic Insights from RL-Trained Language Models](https://arxiv.org/abs/2610.39182) ## Limits The evidence described in the abstract comes from small language models trained in Tic-Tac-Toe, a finite and exhaustively analyzable logical domain; the abstract does not establish how well the findings generalize to other tasks. The supplied abstract does not provide the underlying measurements or further experimental details needed to assess the reported results. ## Open question How well do these findings about Tic-Tac-Toe apply to reasoning benchmarks in other domains? ## Source-stated The authors investigate whether reinforcement learning with verifiable rewards yields reasoning or more superficial adaptations, using models trained in Tic-Tac-Toe. > we probe LLMs trained in a finite and exhaustively analyzable logical domain, namely Tic-Tac-Toe - [Supporting source](https://arxiv.org/abs/2610.39182) ## Source-stated The paper reports high accuracy and robustness across prompt variations, but its mechanistic analysis finds greater reliance on information explicit in the prompt and weak representation of higher-level game concepts. > features extracted with sparse autoencoders (SAEs) reveal that models primarily adapt to rely more heavily on information already explicit in the prompt - [Supporting source](https://arxiv.org/abs/2610.39182) ## Source-stated The authors argue that the contrast between benchmark performance and their mechanistic findings raises concerns that reasoning benchmarks overestimate model abilities. > raising concerns that reasoning benchmarks overestimate model abilities - [Supporting source](https://arxiv.org/abs/2610.39182) ## Source-stated The authors report that reinforcement learning changes which features dominate representations without adding or removing concept information. > RL post-training neither adds nor removes concept information but reorganizes which features dominate the representation - [Supporting source](https://arxiv.org/abs/2610.39182) ## Source and contributor context Nagar, Aishik, Tan, Cheston · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Reinforcement learning Training by trial and error, in which a system is rewarded for the outcomes its designers want. - [Definition reference](https://developers.google.com/machine-learning/glossary#reinforcement-learning-rl) ## Interpretability Research into how a model reaches its outputs, often by examining its internal workings. - [Definition reference](https://developers.google.com/machine-learning/glossary#interpretability) ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: af011bd1-8786-4aca-8eec-0c50befeb390. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-11T12:24:37.059Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: useful\_perspective. Observation: 21435f92e9d9222e79a4b36e3606d9597c5167531861b49e3ace2c8f3783371b. Source text hash: 8786695d9659ed7e50c226219604e0c6d624ae03c284543ea587c17bb64b9901. ### passed — 2026-11-11T12:24:37.059Z — anthropic/claude-opus-5.5 Publication: af011bd1-8786-4aca-8eec-0c50befeb390 attribution — pass: The reading correctly credits the paper and its claims to authors Aishik Nagar and Cheston Tan. It keeps findings and concerns attributed to them, for example 'the authors say' and 'the authors argue', rather than presenting them as established fact. > Authors:Aishik Nagar, Cheston Tan summary — pass: The summary faithfully captures the central tension: strong accuracy and robustness that 'suggest' game reasoning, versus SAE evidence of reliance on prompt-explicit information and weak strategic concepts. It also preserves the authors' benchmark concern, keeps the limits of Tic-Tac-Toe in view, and poses a non-leading question. Terms such as sparse autoencoders, mechanistic interpretability and late-layer state should be explained through term-library tooltips. The phrase 'downstream, in the late-layer state' drops the source's 'of it', meaning downstream of the representation. Restoring that would sharpen the sentence, but the omission does not block understanding. > This suggests that RL-driven "reasoning" may conflate task-specific adaptations to input information with reasoning ability. date — pass: originalPublished 2026-10-29 matches the arXiv v1 submission date. The feed date of Oct 31 is a later announcement and is correctly not used. > \[Submitted on 29 Oct 2026\] support — pass: Every claim is backed by verbatim abstract text, including the description of the models as small, which the abstract calls small-scale LLMs. The draft adds no unsupported benefits, costs or forecasts. The generalization question and the stated limitations follow directly from the finite domain and do not assert a conclusion. > RL post-training neither adds nor removes concept information but reorganizes which features dominate the representation media — pass: The selected media is the verified direct link to the original arXiv abstract page, which is the most relevant way to access the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/629cbc948372fcbb0abdb9d21278f2bd656a64a0df7be176b375ff3cb4980406/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/aishik-nagar-and-cheston-tan-reasoning-or-pattern-exploitation-629cbc94.json)