Reasoning & Chain-of-Thought - Weekly Roundup

A specialized CAD harness drastically improves LLM agent performance, cutting output tokens by up to 88% and boosting F1 scores by 13% through geometry-grounded RL.

Targeted data augmentation and joint RL/OPD training unlock significantly stronger long-horizon multimodal search capabilities for agents.

A strong LLM can review grounded drafts from a weak model, retaining most task performance while halving latency and memory traffic.

Generalization after fine-tuning is dictated by the linear structure of knowledge in pretrained models, not just factual memorization.

Uncovering crucial supporting lemmas for theorem proving by traversing multi-hop dependencies in a formal library's declaration graph.

A specialized CAD harness drastically improves LLM agent performance, cutting output tokens by up to 88% and boosting F1 scores by 13% through geometry-grounded RL.

Targeted data augmentation and joint RL/OPD training unlock significantly stronger long-horizon multimodal search capabilities for agents.

A strong LLM can review grounded drafts from a weak model, retaining most task performance while halving latency and memory traffic.

Generalization after fine-tuning is dictated by the linear structure of knowledge in pretrained models, not just factual memorization.

Uncovering crucial supporting lemmas for theorem proving by traversing multi-hop dependencies in a formal library's declaration graph.

Dual-view time-series representations unlock significant gains in temporal reasoning by preserving both absolute measurements and invariant temporal structure, outperforming dedicated TSLMs.

Agents can now learn reusable knowledge from experience by converting it into ontological axioms, enabling LLMs to dynamically propose and verify hypotheses for complex tasks.

Deep learning architectures designed for semantic understanding can now solve complex problems in algebraic topology, identifying topologically equivalent finite group actions on Riemann surfaces.

LLM agents can be reliably steered back to task goals with a learned, dual-path intervention mechanism that generalizes across tasks and executors.

LLM agents can learn to proactively build optimized, hierarchical file system workspaces that anticipate future queries, outperforming reactive retrieval by 33 points on BrowseComp-Plus.

Rational synthesis for multi-agent systems with unknown dynamics and $\mathsf{PLTL}_f$ objectives is achieved by integrating solution concepts like Nash equilibrium with temporal logic satisfaction.

A VLM can learn to dynamically "imagine" missing visual evidence to solve complex spatiotemporal reasoning tasks, outperforming direct reasoning and fixed imagination pipelines.

LLM reasoning failures are often misdiagnosed by standard probes, with even specialized probes failing to pinpoint the exact faulty step more reliably than trace length.

Frontier LLMs fail to reliably identify and apply domain-specific rules in complex, multi-file scenarios, with top models scoring below 63% on core reasoning tasks.

Scientific agent swarms can achieve state-of-the-art performance and generalize to new tasks by evolving through feedback-guided mutation and complementary recombination of hypotheses and code.

Asynchronous RL training on long-horizon tasks achieves superior performance by leveraging Muon and a novel single-rollout recipe that avoids waiting for complete trajectories.

A single-rollout RLVR method achieves faster training and more stable gradients by using a trajectory's own future as a hindsight context, eliminating the need for group sampling and reference solutions.

A novel PPO variant leverages complementary temporal dynamics of self-teacher signals and critic value predictions to boost reasoning accuracy by up to 2.59pp.

LLM reasoning can be boosted at inference time by separately sharpening reasoning paths and final answers, outperforming existing power-sampling methods across multiple benchmarks.

Agents can recursively improve their problem-solving by learning to adapt their own policy and memory management strategies, achieving substantial error reduction in program selection.

LLM reasoning can be optimized by learning from the geometry of state transitions, achieving reflection-level accuracy with 28% lower latency and 25.8% fewer tokens.

Current chemistry benchmarks conflate memorization with derivation and treat visual/textual inputs interchangeably; ChemArgus finally provides fine-grained diagnostic scores that pinpoint *where* a model's reasoning chain breaks.

Generative models, when framed as proposal distributions for planning, unlock efficient long-horizon reasoning in model-based RL, outperforming diffusion policies and model-free methods.

Agents trained with Agent-Reflex exhibit a 29.1% improvement on a noise-heavy benchmark, demonstrating robust failure recovery and proactive reflection even with faulty tool feedback.

LLMs can now actively seek and ground verifiable spatio-temporal evidence, including object bounding boxes, to improve video reasoning accuracy by over 12%.

Unexplained observations, not just supporting evidence, are the key to LLMs discovering truly novel hypotheses.

System One models can achieve direct, low-latency decision-making by treating answers as structured choices rather than serialized text, with Jeva demonstrating robust null detection and permutation invariance.

A small, training-free "brief thinker" can cut LLM reasoning tokens by 31% without sacrificing accuracy.

Vision-language-action models can achieve superior performance and drastically reduced latency by leveraging reasoning supervision without explicitly conditioning actions on generated reasoning.

Latent thoughts are now inspectable and tool-integrable, achieving 10.4% better performance with 3x faster inference and 70% fewer tokens than frontier LLM agents.

Autonomous research agents can achieve state-of-the-art performance and efficiency by coupling rapid, parallel evidence assessment with strategic, deliberative planning.

Recursive message passing achieves 99% Sudoku-Extreme accuracy with just 16k parameters, outperforming larger neural baselines on N-queens and SAT with zero-shot generalization.

LLM agents can learn complex tasks more efficiently by adaptively injecting skills only when their utility is confirmed through outcome-aware likelihood comparisons on existing data, bypassing expensive environment rollouts.

VLMs can be taught to *use* latent geometry by training them to read continuous geometric tokens, rather than just reconstructing geometry from RGB.

Self-distillation methods fail to generalize reasoning gains to unseen problems and their gradients offer weak policy improvement signals, challenging their efficacy beyond in-distribution imitation.

LLM agents can learn to predict their own failure from the very first step of a task, achieving a 4.3x higher success rate in their most confident actions.

AI-generated mathematical proofs can now be verified at the step-level without human annotation, thanks to a novel framework that leverages formal proof assistants.

Autonomous driving VLAs can achieve low-latency, SOTA planning by generating actions before CoT, using understanding solely to supervise generation.