mlsanityData to 14 Oct 2026

The Free-Recipe Limit: Every Recipe Effect Measures Which Premise of an Idealised Learner Broke

arXiv 2609.26160 ↗First posted 9 Aug 2026cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Fix a corpus and send recipe search to infinity: try every order of the skills, every arrangement from blocked to interleaved, every composition, and keep the best. Two quantities decide what that search was worth: the diameter of the reachable set it explores, and the resolution at which anyone can tell two endpoints apart. Where the diameter falls below the resolution, no amount of search converts into a decision, and the signature is not an absence of winners but winners that do not survive re-running. We measure this recipe-search wall with 761 fine-tuning runs on 12 base models (0.5B-14B, three pretraining families) over competition-mathematics skills: base checkpoints, supervised fine-tuning under AdamW, exact-match scoring at k=4. Within one coherent domain at fixed volume the three classical freedoms average 0.010-0.021 against a 0.019 floor, and the largest contrast, 0.0619, clears a three-seed resolution and then reads +0.010 and -0.015 on two reruns. The departure with a systematic answer is coherence: halving one pooled corpus and letting the halves write answers under incompatible but equally correct conventions moves arrangement from capability to allocation between conventions, by two orders of magnitude over a same-convention control, and writing the convention into the input switches the phenomenon off. The switch replicates on a second pretraining family and survives an independent re-execution of its own protocol, with a re-execution spread (0.087) smaller than the resolution a search-selected order cell carries (0.144). Order itself is a transient whose sign crosses zero three times inside a single run. Volume, the one lever nobody calls a recipe, is the one that reliably pays. A public scorecard grades all 26 pre-registered claims: 18 supported, 5 failed, 2 untested, 1 mixed.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 48 of its 61 references that are in the map. First posted on arXiv 9 Aug 2026.

Builds on

48 of 61 references.

References · 48

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2025Qwen3 Technical Reportcited by 8,820
  3. 2020Scaling Laws for Neural Language Modelscited by 6,586
  4. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  5. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  6. 2024Qwen2.5 Technical Reportcited by 4,942
  7. 2016Overcoming catastrophic forgetting in neural networkscited by 4,787
  8. 2018Neural Tangent Kernel: Convergence and Generalization in Neural Networkscited by 2,621
  9. 2017Gradient Episodic Memory for Continual Learningcited by 1,898
  10. 2020Gradient Surgery for Multi-Task Learningcited by 1,256
  11. 2018On Lazy Training in Differentiable Programmingcited by 675
  12. 2020Estimating Training Data Influence by Tracing Gradient Descentcited by 580
36 more
  1. 2023Are Emergent Abilities of Large Language Models a Mirage?cited by 523
  2. 2021Conflict-Averse Gradient Descent for Multi-task Learningcited by 459
  3. 2023DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretrainingcited by 399
  4. 2024SWIFT:A Scalable lightWeight Infrastructure for Fine-Tuningcited by 355
  5. 2024Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations?cited by 232
  6. 2022Multi-Task Learning as a Bargaining Gamecited by 215
  7. 2024Continual Learning for Large Language Models: A Surveycited by 147
  8. 2018Don't forget, there is more than forgetting: new metrics for Continual Learningcited by 115
  9. 2023DoGE: Domain Reweighting with Generalization Estimationcited by 90
  10. 2021Continual Learning in the Teacher-Student Setup: Impact of Task Similaritycited by 88
  11. 2022How catastrophic can catastrophic forgetting be in linear regression?cited by 81
  12. 2025Beyond Random Sampling: Efficient Language Model Pretraining via Curriculum Learningcited by 26
  13. 2025Mid-Training of Large Language Models: A Surveycited by 21
  14. 2021Hyperparameter Optimization Is Deceiving Us, and How to Stop Itcited by 20
  15. 2025Data Mixing Can Induce Phase Transitions in Knowledge Acquisitioncited by 16
  16. 2025How Learning Rate Decay Wastes Your Best Data in Curriculum-Based LLM Pretrainingcited by 14
  17. 2026Mechanistic Analysis of Catastrophic Forgetting in Large Language Models During Continual Fine-tuningcited by 14
  18. 2025What Makes a Good Curriculum? Disentangling the Effects of Data Ordering on LLM Mathematical Reasoningcited by 11
  19. 2025Assessing the Macro and Micro Effects of Random Seeds on Fine-Tuning Large Language Modelscited by 10
  20. 2025ConInstruct: Evaluating Large Language Models on Conflict Detection and Resolution in Instructionscited by 10
  21. 2026What do Language Models Learn and When? The Implicit Curriculum Hypothesiscited by 10
  22. 2026The Geometry of Sequential Learning: Lie-Bracket Prediction of Transfer Ordercited by 9
  23. 2026Data Repetition Beats Data Scaling in Long-CoT Supervised Fine-Tuningcited by 8
  24. 2026Why Supervised Fine-Tuning Fails to Learn: A Systematic Study of Incomplete Learning in Large Language Modelscited by 8
  25. 2025Influence Dynamics and Stagewise Data Attributioncited by 7
  26. 2026Optimizer Memory Makes Shuffle Order a First-Order Source of Fine-Tuning Noisecited by 7
  27. 2026Curriculum Learning for LLM Pretraining: An Analysis of Learning Dynamicscited by 6
  28. 2026Learning is Forgetting: LLM Training As Lossy Compressioncited by 5
  29. 2026Meet Dynamic Individual Preferences: Resolving Conflicting Human Value with Paired Fine-Tuningcited by 4
  30. 2026The Capability Convergence Hypothesis: Capability from Access Structure, Not Scalecited by 4
  31. 2026First-Order Predictable but Pairwise Fragile: Local Task Adaptation in Trained Transformerscited by 4
  32. 2025Interleaved Multitask Learning with Energy Modulated Learning Progresscited by 3
  33. 2026Winner's Curse Drives False Promises in Data-Driven Decisions: A Case Study in Refugee Matchingcited by 3
  34. 2026Demystifying Data Organization for Enhanced LLM Trainingcited by 3
  35. 2026DataChef: Cooking Up Optimal Data Recipes for LLM Adaptation via Reinforcement Learningcited by 2
  36. 2026From Instance Selection to Fixed-Pool Data Recipe Search for Supervised Fine-Tuningcited by 1

Built on by

Not cited yet.