arXiv Machine Learning
Give Up to Get More: Priors, Not Policies, Govern Test-Time Compute Allocation
Read the original on arXiv Machine Learning →The Flow has not summarised this story yet — read it at arXiv Machine Learning.
Related stories
arXiv Machine Learning
Oct 19
Sample, Mix, or Verify? Allocating Test-Time Compute under Imperfect Verifiers
The paper introduces SPLIT, a method that optimally allocates test‑time compute among sampling, model selection, and verification for language models and agents. Using an exact probability formula that accounts for task difficulty, shared blind spots, and verifier correlations, SPLIT is fitted via empirical Bayes on a small labeled pilot and then used to plan budgets for unseen tasks. Across seven diverse benchmarks, 26 models, and six verifier types, SPLIT achieves near‑optimal performance, outperforming rule‑based baselines by up to 2.5 percentage points and providing insights into when additional LLM judges reduce precision.
arXiv AI
Oct 27
What Effort Level Should I Choose? On the Economics of Reasoning Effort in LLM Agents
The paper investigates how users should set the reasoning‑effort level in large language models, a parameter that trades off inference cost for potential performance gains. Experiments on proprietary and open‑weight models across four benchmarks show that higher effort does not consistently improve accuracy, can be up to 42× more expensive on easy tasks, and alters agent behavior by increasing verification and search steps. The authors propose a simple escalation strategy—starting at the lowest effort and only increasing after a verified failure—that matches or outperforms fixed high‑effort settings while reducing cost.
arXiv Computation and Language
2d ago
Concise-9B: Removing the Slack from a Reasoning Model's Thinking
The paper introduces Concise-9B, a version of the Qwen3.5-9B reasoning model that reduces unnecessary thinking tokens through concise self‑training. By applying two rounds of on‑policy rejection‑sampling fine‑tuning and a small share of answer‑synthesis targets, Concise-9B cuts token usage to roughly two‑thirds of the original model while maintaining comparable accuracy on a wide range of hard mathematics and reasoning benchmarks. The study also shows that most of the token savings occur after the answer has already formed, and that the approach’s effectiveness depends on the verbosity of the base model.
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
Sep 24
Spread and Scale: What Determines Whether Test-Time Budget Allocation Pays
The paper investigates when reallocating a fixed test‑time budget toward harder instances improves solution quality for neural combinatorial optimization solvers. Through pre‑registered experiments on three solvers and two hard‑workload constructions for the traveling salesman problem, it finds that the key deciding factor is the variation in instance difficulty within a workload, not the average difficulty. A budget‑aware policy that first spends part of the budget to gauge instance difficulty recovers most of the potential improvement, though not all, when the cost of this information is included.
arXiv AI
Aug 21
Learning When to Think: Adaptive Reasoning for Test-Time Compute Allocation
arXiv:2608. 20256v1 Announce Type: new Abstract: Reasoning language models trained with reinforcement learning typically operate under a fixed token budget rather than an explicitly adaptive one, which can lead to over-computation on easy problems and insufficient computation on difficult ones.