arXiv AI

When Does Less Information Pay? An Exact Finite Atlas of Representation, Payoff, and Cost

The paper presents a complete enumeration of all possible representations and binary action rules in a uniform four‑state world, comparing their retained information, payoff (task accuracy), and cost under various tariffs. It shows that a representation with less retained information can sometimes yield higher payoff, while in other cases the more informative representation performs better, and that equal payoff can occur with different information amounts. The study demonstrates how the specific distinctions preserved by a representation, the choice of action rule, and the cost assumptions jointly determine performance, offering a reproducible model for exploring these trade‑offs.

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.

By Kaiser Sun, Yongchan Kwon, Zain Hasan, Shang Zhu, Kaitlyn Zhou, Federico Bianchi
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.

By Jinhyung Bae
arXiv AI
Oct 16

Verdict: Encode Once, Decide Many

Verdict-MQ is an open encoder that processes a single piece of state once and then answers multiple typed questions about that state, avoiding separate passes for each question. By using a structured attention mask, it ensures the state encoding is independent of question order and prevents questions from influencing one another. On the typed‑decisions benchmark, Verdict‑MQ outperforms a larger specialist model, achieving higher accuracy and better calibration, while on other benchmarks it shows mixed results compared to fine‑tuned baselines.

By Avdhesh Singh Chouhan, Jack Perry
arXiv AI
Aug 24

Structure for Reading, Prose for Writing: Asymmetric Structural Conditioning in Multi-Agent Document Authoring

The paper reports on a deployed multi‑agent tender‑response system that uses an open‑weights language model under sovereignty constraints. In a blind comparison, the system’s answers were judged at least as good as human‑written bids in 40 of 55 sections, with only a few gaps attributable to missing knowledge rather than writing quality. The study also demonstrates an asymmetry in conditioning: while structural markup improves reading tasks, converting instruction material from prose to nested XML degrades answer quality, and naming forbidden constructions concentrates defects.

By Cheng Yu, Nikhil Mathew, Zhengjie Wang
arXiv Machine Learning
Nov 5

Give Up to Get More: Priors, Not Policies, Govern Test-Time Compute Allocation

The paper investigates how to allocate a fixed number of attempts by a large language model (LLM) across inputs of varying difficulty. It finds that the optimal static allocation follows an inverted‑U shape, giving few attempts to very hard inputs and more to moderately difficult ones, and that this allocation is largely determined by prior knowledge of solvability rather than the policy itself. Using a verifier and a prior derived from a related model’s past runs, Thompson Sampling can significantly increase the solved share on the hardest workloads, whereas simple round‑robin or other tested policies fail to outperform it without such priors.

By Jinhyung Bae