arXiv Machine Learning
Jul 15
Extractable Memorization From First Principles
arXiv:2607. 12649v1 Announce Type: new Abstract: Recent work on extractable memorization in LLMs suffers from two contrasting validity problems.
Where Induction Runs Out: Description-Length Difficulty and the Memorisation Gap in Integer-Sequence Benchmarks
The paper investigates integer‑sequence benchmarks from the OEIS by applying a two‑part minimum description length (MDL) learner that searches for P‑recursive recurrences. It finds that MDL difficulty correlates with a combinatorial parameter count, that most sequences fit a recurrence on a prefix but not at full length (the “wilderness” regime), and that language models do not hallucinate in the wilderness but instead hedge, showing that memorisation dominates perceived competence. The study provides a cheap, contamination‑free difficulty signal for OEIS‑derived benchmarks.
Related stories
arXiv AI
Aug 17
The Metacognitive Bottleneck: Japanese Riddles Reveal Fundamental Limits of Machine Insight and Self-Evaluation in Reasoning AI
arXiv:2509. 14704v3 Announce Type: replace Abstract: Benchmark saturation and training-data contamination increasingly obscure whether reported gains in large language models (LLMs) reflect genuine advances in reasoning or familiarity with recurring patterns in benchmark problems.
arXiv AI
Jun 16
The Faithfulness Gap: Certifying Semantic Equivalence Between Natural-Language and Formal Mathematical Statements
arXiv:2606. 16541v1 Announce Type: new Abstract: Autoformalization, translating natural-language mathematics into formal proof assistants, is bottlenecked not by translation fluency but by \emph{faithfulness}: a formal statement can typecheck and be provable, yet still encode a different theorem than the source intended.
arXiv AI
Aug 3
Library Reachability in LSR-Synth: How Anti-Memorization Design Changes the Measurement of Symbolic Discovery
arXiv:2607. 28684v1 Announce Type: new Abstract: Existing benchmarks for scientific equation discovery are largely composed of well-known equations available in the public domain, making it difficult to determine whether a model is discovering laws from data or merely recalling answers from its training corpus.
arXiv Machine Learning
1d ago
Detectable Only Where It Is Confounded: What Verified Duplication Counts Say About Membership Evidence in Language Models
The paper investigates whether language models can identify sentences from their training data by using exact duplication counts from publicly released corpora for two model families, OLMo‑2 and Pythia. It finds that for typical duplication levels, models show only a weak trace of exposure, with a rank correlation near –0.08, and that strong signals only appear when a sentence appears roughly a thousand times, at which point fame rather than memory dominates. The study also demonstrates that common membership tests can be misleading, as changing a single word does not alter the model’s preference, and that controlling for register can significantly improve detector performance.
arXiv AI
Aug 25
Measuring in-context algorithmic reasoning in language models against an exact Bayes-optimal reference
The paper introduces F-ICL, a benchmark that measures in‑context algorithmic reasoning in language models by exhaustively enumerating 86 million valid programs of length ≤13 on a Turing‑complete machine and computing the exact posterior under a bounded Levin–Solomonoff prior. Unlike typical benchmarks, F‑ICL provides a distributional reference rather than just answers, allowing the evaluation of models’ inductive priors. Across 105 configurations of models ranging from 0.8 B to 675 B parameters, models achieve up to 92 % accuracy, yet many still deviate from the Bayes‑optimal reference, and the study derives theoretical bounds on cumulative loss for predictors with positive prior weight on the reference.
arXiv AI
Jul 10
Understanding Axes of Difficulty For Long Context Tasks Via PredicateLongBench
arXiv:2607. 08284v1 Announce Type: new Abstract: Large language models (LLMs) have demonstrated rapidly improving long-context capabilities, prompting a wave of benchmarks designed to evaluate them.
arXiv Machine Learning
Jun 9
Operationalising the Superficial Alignment Hypothesis via Task Complexity
arXiv:2602. 15829v2 Announce Type: replace Abstract: The superficial alignment hypothesis (SAH) posits that large language models learn most of their knowledge during pre-training, and that post-training merely surfaces this knowledge.
arXiv Machine Learning
Aug 19
Thinking in a Low-Resource Language: What SFT Builds, What RL Fixes, What Accuracy Cannot See
The paper evaluates how three large mixture‑of‑experts models (Alibaba, OpenAI, NVIDIA) can be fine‑tuned to reason in a low‑resource language, specifically Greek. Accuracy metrics show little change, but the authors uncover significant qualitative improvements: after supervised fine‑tuning, models reason in Greek on ~98% of items, with better grammaticality and retained general ability. Reinforcement learning with pre‑registered rewards further eliminates reasoning‑channel leaks and format skips, while the Greek‑reasoning habit remains robust to an accuracy‑only gradient.
arXiv AI
Jul 14
Valid $\ne$ Necessary: Diagnosing Latent Inefficiency in Chain-of-Thought
arXiv:2607. 11266v1 Announce Type: new Abstract: Chain-of-Thought (CoT) prompting has significantly advanced the reasoning capabilities of Large Language Models (LLMs), yet it often incurs substantial computational costs due to over-reasoning: the generation of redundant, verbose, or irrelevant steps.
arXiv Machine Learning
Jul 28
Hallucination Rates in Language Generation
arXiv:2607. 23361v1 Announce Type: cross Abstract: Language generation in the limit is an elegant model introduced by Kleinberg and Mullainathan [KM24] to formally study language generation by an algorithm that learns solely based on example strings.
arXiv Machine Learning
Jun 11
Teaching Diffusion to Speculate Left-to-Right
arXiv:2606. 11552v1 Announce Type: cross Abstract: Large language models (LLMs) achieve remarkable performance across a wide range of tasks, but their autoregressive decoding process incurs substantial inference costs due to inherently sequential token generation.