AI Navigate

EsoLang-Bench: 難解プログラミング言語を用いた大規模言語モデルの真の推論能力評価

arXiv cs.AI / 2026/3/11

Ideas & Deep AnalysisModels & Research

要点

  • EsoLang-Benchは、大規模言語モデルの真の推論能力を評価するために設計された新しいベンチマークであり、5つの難解プログラミング言語を使用している。これらの言語は公開コードリポジトリが非常に少なく、ベンチマークの不正利用リスクが低い。
  • ベンチマークにはBrainfuck、Befunge-98、Whitespace、Unlambda、Shakespeare言語が含まれており、主流のプログラミングと同じ計算プリミティブを共有しながらも、公開の入手可能性や馴染みやすさが大きく異なる。
  • さまざまなプロンプト戦略を用いて5つの最新モデルを評価した結果、標準的なベンチマークで85-95%の性能を示したモデルが、難解言語では0-11%の精度に急落し、最も簡単な課題を越えて成功した例はなかった。
  • few-shot学習や自己反省の技術は結果を改善せず、これらの方法が真の新規学習や推論を促すのではなく既存の訓練知識を活用していることを示している。
  • EsoLang-Benchはドキュメントレビュー、インタプリタのフィードバック、反復実験を通じて人間の言語獲得を模倣し、データ汚染の影響を受けにくい転移可能な推論能力のより堅牢なテストを提供することを目指している。

Computer Science > Artificial Intelligence

arXiv:2603.09678 (cs)
[Submitted on 10 Mar 2026]

Title:EsoLang-Bench: Evaluating Genuine Reasoning in Large Language Models via Esoteric Programming Languages

View a PDF of the paper titled EsoLang-Bench: Evaluating Genuine Reasoning in Large Language Models via Esoteric Programming Languages, by Aman Sharma and 1 other authors
View PDF HTML (experimental)
Abstract:Large language models achieve near-ceiling performance on code generation benchmarks, yet these results increasingly reflect memorization rather than genuine reasoning. We introduce EsoLang-Bench, a benchmark using five esoteric programming languages (Brainfuck, Befunge-98, Whitespace, Unlambda, and Shakespeare) that lack benchmark gaming incentives due to their economic irrationality for pre-training. These languages require the same computational primitives as mainstream programming but have 1,000-100,000x fewer public repositories than Python (based on GitHub search counts). We evaluate five frontier models across five prompting strategies and find a dramatic capability gap: models achieving 85-95% on standard benchmarks score only 0-11% on equivalent esoteric tasks, with 0% accuracy beyond the Easy tier. Few-shot learning and self-reflection fail to improve performance, suggesting these techniques exploit training priors rather than enabling genuine learning. EsoLang-Bench provides the first benchmark designed to mimic human learning by acquiring new languages through documentation, interpreter feedback, and iterative experimentation, measuring transferable reasoning skills resistant to data contamination.
Comments:
Subjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Software Engineering (cs.SE)
Cite as: arXiv:2603.09678 [cs.AI]
  (or arXiv:2603.09678v1 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2603.09678
Focus to learn more
arXiv-issued DOI via DataCite

Submission history

From: Aman Sharma [view email]
[v1] Tue, 10 Mar 2026 13:47:15 UTC (89 KB)
Full-text links:

Access Paper:

    View a PDF of the paper titled EsoLang-Bench: Evaluating Genuine Reasoning in Large Language Models via Esoteric Programming Languages, by Aman Sharma and 1 other authors
  • View PDF
  • HTML (experimental)
  • TeX Source
Current browse context:
cs.AI
< prev   |   next >
Change to browse by:

References & Citations

export BibTeX citation Loading...

BibTeX formatted citation

×
Data provided by:

Bookmark

BibSonomy logo Reddit logo
Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
Links to Code Toggle
Papers with Code (What is Papers with Code?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos

Demos

Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers

Recommenders and Search Tools

Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.