The paper investigates whether token‑confidence scores used to prune reasoning chains in language models actually reflect model certainty or merely chain length. By treating per‑token confidence as a signal and applying spectral analysis and change‑point detection, the authors find that the windowed‑minimum confidence score largely correlates with how long a chain is, offering no additional predictive power for correctness once length is accounted for. Although the confidence spectrum contains a weak signal of rapid token‑to‑token fluctuation in incorrect chains, it is too faint to improve majority voting, and early‑stopping methods based on confidence or change detection do not outperform simply sampling fewer chains.
arXiv:2606. 30128v1 Announce Type: new Abstract: Chain-of-thought (CoT) prompting improves LLM reasoning, but the source is contested: do the intermediate steps help because they carry useful semantic content, or because conditioning on more tokens buys extra computation before the model commits to an answer?
As the chain-of-thought reasoning capabilities of large language models improve, evaluating and calibrating their reasoning confidence is becoming increasingly important for quantifying the uncertaint...
arXiv:2604. 04930v2 Announce Type: replace-cross Abstract: Large reasoning models rely on long chain-of-thought generation to solve complex problems, but extended reasoning often incurs substantial computational cost and can even degrade performance due to overthinking.
The paper introduces Divergent Token Confidence (DTC), a method that estimates large language model confidence by counting tokens where two models strongly disagree during decoding. DTC uses Jensen-Shannon divergence between next-token distributions along the same reasoning trajectory and shows a near-negative correlation with answer accuracy. Experiments on multiple model families and six mathematical benchmarks demonstrate that DTC improves calibration over traditional probability-based and verbalized baselines, achieving lower expected calibration errors in both white-box and black-box settings.
arXiv:2606. 25524v2 Announce Type: replace Abstract: Large language models (LLMs) reach high accuracy in mathematical reasoning, but individual traces on the same problem diverge; some arrive at the correct answer while others fail.
arXiv:2606. 02020v1 Announce Type: cross Abstract: This paper investigates the entropy dynamics of Chain-of-Thought (CoT) and uncovers a consistent two-phase structure: an Uncertainty Region of exploration transitioning sharply to a Confidence Region of convergence.
arXiv:2608. 11138v1 Announce Type: cross Abstract: We propose that a model's uncertainty about a token is reflected not only in the breadth of its output distribution but also in whether a confident prediction is \emph{fragile} under perturbation of its attention pathways.
arXiv:2610.34350v1 Announce Type: new Abstract: Understanding how individual token choices affect reasoning length is important for characterizing reasoning models. Building on prior findings that ch...
The paper argues that pattern recognition and step‑by‑step reasoning lie on a spectrum, with large language models (LLMs) learning the latter when the next token depends on only a few preceding tokens. It formalises reasoning traces as paths on a De Bruijn graph, showing that the number of edges is far smaller than the number of possible traces, making step‑by‑step reasoning sample‑efficient. Experiments fine‑tuning Qwen2.5‑1.5B‑Instruct demonstrate that a moderate density of states balances accuracy and robustness, and that real‑world models like Qwen3 retain most of their performance even when attention is limited to a small sliding window.
arXiv:2606. 00206v1 Announce Type: new Abstract: Post-training quantization (PTQ) is widely used to deploy large language models efficiently, but its effect on reasoning models is not well understood.
The paper investigates whether semantic entropy—a measure of disagreement among a language model’s sampled answers—can serve as a cheap signal for deciding when to route a query from a small to a larger language model. Experiments on GSM8K and other benchmarks show that semantic entropy can distinguish small‑model mistakes and improve routed accuracy, but the authors also reveal that a simple question‑difficulty rule can mimic its performance and that other factors (definition of success, benchmark design, live sampling cost) can undermine its effectiveness. They propose a checklist of checks to validate escalation signals and demonstrate how to predict when a cached‑outcome approach will fail. "whyItMatters":"The study highlights the importance of rigorous evaluation of escalation signals, showing that seemingly promising metrics can be misleading without proper controls and that practical routing decisions must account for cost and benchmark design."
arXiv:2607. 20129v1 Announce Type: new Abstract: Quantized small autoregressive reasoning models can enter long, repetitive, or unproductive trajectories, yet inference-time compute is usually allocated without observing how a trajectory develops.