Skip to main content
Search

# Chain-of-Thought

12 related articles

Subscribe to updates
  1. arXiv

    LLM explanations stop being written per question: one executable program reproduces model outputs and runs 6 to 100 times shorter than chain-of-thought

    Researchers replace per-instance free-text explanations of large language models with a single executable program that runs over a dataset and is compared against the model's outputs, with consistency measuring how often it reproduces model behavior. Across three open-source code-capable models and eight tasks, 6 of 24 model-task settings exceed 0.75 consistency, code explanations are 6 to 100 times shorter than chain-of-thought, and they align better with model behavior in 9 of 24 settings; on tasks with more complex linguistic components, consistency is no higher than an uninformative baseline.

  2. arXiv

    Chain-of-thought solves unseen multi-step tasks by reusing one step predictor, with error growing sublinearly in depth

    Chain-of-thought out-of-distribution generalization is framed as compositional generalization: training supplies only single-step atomic transitions, while the target task requires composing many steps. Theory bounds target risk by the sum of step risks times a depth-independent coefficient, and shows that sharing one predictor makes the estimation term shrink as the inverse square root of depth rather than grow linearly. Atom-Task and GSM8K experiments support short-to-long and unseen-combination transfer, and show explicit state tracking is essential.

  3. arXiv

    An answer probe may be reading the candidate about to be written: dual-target probing on DecisionDepth opens a .696 emission gap

    Answer probes on chain-of-thought hidden states are often read as evidence that a model has settled its final answer, yet a hidden state also carries the candidate whose name is about to be written. Emission-controlled dual-target probing (ECDP) fits the same low-rank reader separately to the next candidate mention and to the final answer at identical positions, with target-specific nulls and thresholds fixed in advance. On DecisionDepth, Qwen3.5-4B decodes the next mention at .966 (null .234) but the final answer only at .221 (null .186), an emission gap of .696; Llama-3.1-8B gives .575. The order is not universal: in a Qwen3-8B MMLU setting the final answer is read more strongly.

  4. Health economic model conceptualisation moves to agentic AI: two cases match the core staging logic of NICE appraisals, with automated runs of about an hour

    An agentic generative-AI framework produced disease concept diagrams and model structure diagrams for breast cancer and hATTR-PN that were structurally concordant with the core disease-staging logic of the NICE appraisal accepted for each indication, with automated runs of about 1 hour, or 1–2 days including expert validation, against up to 3 months of fully manual conceptualisation.

  5. arXiv

    LLoCoT refines a compact latent workspace with a looped transformer, matching explicit-CoT accuracy on HumanEval and MBPP while cutting time to first answer token by about 36x

    The authors introduce LLoCoT, which replaces left-to-right latent generation with a few iterative refinement passes of a shared transformer over a compact latent workspace, then samples latent tokens in parallel from a probabilistic head to condition an autoregressive decoder; on HumanEval and MBPP it matches the accuracy of the explicit-CoT baseline Reasoning SFT and outperforms the base model, answer-only SFT and NF-CoT, while reducing time to the first answer token by about 36x, reasoning-phase latency by about 42x, and raising end-to-end throughput by 9.2%.

  6. arXiv

    DBRAG combines a table index with query-relevant row reranking to reach 95.8 Recall@5 on Spider and improve multi-table QA

    DBRAG decomposes multi-table QA into two stages: an offline table index retrieves candidate tables, query-relevant rows replace random rows in their summaries for LLM reranking, and a program-aided chain-of-thought reasoner selects tables and executes operations over their full contents; on Spider, GeoQuery, and ATIS it improves table retrieval and multi-table QA metrics over the compared baselines.

  7. AlphaGo team member Thore Graepel leaves DeepMind, arguing AlphaGo-style search—not LLM chain of thought—is the path to genuine machine reasoning

    Using move 37 from the 2016 Seoul match, AlphaGo core member Thore Graepel argues the move came from AlphaGo's search machinery rather than intuition, contends that today's large language models—relying on next-token prediction and chain of thought—do not reason in a scientist's sense, and proposes borrowing AlphaGo's game-tree structure so that general reasoning systems maintain an explicit, inspectable epistemic state updated by evidence, with an independent component judging each step by how much it actually resolves uncertainty.

  8. arXiv

    A hidden current date in system prompts shifts nine models' scores by up to 14% and reshuffles leaderboards

    Holding all other settings fixed and varying only the hidden current date injected into the system prompt across every day of 2024, this study of 9 recent LLMs and 6 datasets finds accuracy swings of up to 6% on multiple-choice QA, 14% on math reasoning, 7% on code generation, and 2.84 BLEU on machine translation, with model rankings reordering, an effect larger than batch size or numerical precision and one that chain-of-thought prompting amplifies rather than reduces.

  9. arXiv

    ThinkV2V makes MLLMs think before editing: a 5B model tops both complex and standard video-editing benchmarks

    The work proposes ThinkV2V, a reasoning-driven video-editing framework that explicitly activates multimodal large language model thinking before visual generation: it uses Qwen3-VL-Thinking-8B to perform chain-of-thought reasoning over the source video and instruction and output a refined prompt, injects the hidden states into a Wan-2.1-5B DiT through a learnable-query connector, pairs this with Progressive Curriculum Training and Inference-Time Thinking Scaling, and builds the ThinkV2V-150K dataset and ThinkV2V-Bench; under two judge models, the 5B-scale model achieves the best overall scores on both complex and standard editing scenarios and surpasses several 10B-scale baselines.

  10. arXiv

    AbSteering steers general-purpose VideoLMs with abnormality-centric chain-of-thought and DPO to generate HRCT reports, surpassing large-scale CT-specific foundation models on fine-grained clinical metrics while improving detection sensitivity and reducing hallucinations.

    The work presents AbSteering, a two-stage framework combining abnormality-centric chain-of-thought training with a Direct Preference Optimization objective for fine-grained abnormality discrimination, to adapt general-purpose VideoLMs to high-resolution CT report generation, and curates the CT-RATE-AB dataset; results show that general-purpose VideoLMs transfer effectively to 3D medical imaging under limited data, achieving state-of-the-art performance on fine-grained clinical efficacy metrics, with superior detection sensitivity over domain-specific CT foundation models pretrained on large-scale CTs while mitigating hallucinations.