The field of large language models (LLMs) is rapidly advancing, with significant improvements in reasoning capabilities. Recent developments focus on enhancing chain-of-thought reasoning, temporal reasoning, and the integration of multiple modalities. Notably, advancements in latent chain-of-thought reasoning and structure-aware generative frameworks are pushing the boundaries of LLMs' capabilities. Furthermore, research on tokenization constraints, entropy minimization, and reinforcement learning is providing new insights into the limitations and potential of LLMs. Noteworthy papers include 'Time-R1: Towards Comprehensive Temporal Reasoning in LLMs', which introduces a framework for endowing LLMs with comprehensive temporal abilities, and 'Visual Thoughts: A Unified Perspective of Understanding Multimodal Chain-of-Thought', which explores the mechanisms driving improvements in multimodal chain-of-thought methods.
Advancements in Large Language Models' Reasoning Capabilities
Sources
Ranked Voting based Self-Consistency of Large Language Models
Reasoning with OmniThought: A Large CoT Dataset with Verbosity and Cognitive Difficulty Annotations
Benchmarking Critical Questions Generation: A Challenging Reasoning Task for Large Language Models
Extracting Explainable Dates From Medical Images By Reverse-Engineering UNIX Timestamps
SoftCoT++: Test-Time Scaling with Soft Chain-of-Thought Reasoning
Time-R1: Towards Comprehensive Temporal Reasoning in LLMs
Mapping the Minds of LLMs: A Graph-Based Analysis of Reasoning LLM
Measuring the Faithfulness of Thinking Drafts in Large Reasoning Models
Structured Agent Distillation for Large Language Model
Interpretable Traces, Unexpected Outcomes: Investigating the Disconnect in Trace-Based Knowledge Distillation
Beyond Semantics: The Unreasonable Effectiveness of Reasonless Intermediate Tokens
Tokenization Constraints in LLMs: A Study of Symbolic and Arithmetic Reasoning Limits
Temporal Alignment of Time Sensitive Facts with Activation Engineering
The Strawberry Problem: Emergence of Character-level Understanding in Tokenized Language Models
Reinforcement Learning vs. Distillation: Understanding Accuracy and Capability in LLM Reasoning
A MIND for Reasoning: Meta-learning for In-context Deduction
Not All Correct Answers Are Equal: Why Your Distillation Source Matters
Exploring Graph Representations of Logical Forms for Language Modeling
Mind the Gap: Bridging Thought Leap for Improved Chain-of-Thought Tuning
FOL-Pretrain: A complexity annotated corpus of first-order logic
Learning to Rank Chain-of-Thought: An Energy-Based Approach with Outcome Supervision
The Unreasonable Effectiveness of Entropy Minimization in LLM Reasoning
Towards Explainable Temporal Reasoning in Large Language Models: A Structure-Aware Generative Framework
Visual Thoughts: A Unified Perspective of Understanding Multimodal Chain-of-Thought
DayDreamer at CQs-Gen 2025: Generating Critical Questions through Argument Scheme Completion
Soft Thinking: Unlocking the Reasoning Potential of LLMs in Continuous Concept Space
Learning to Reason via Mixture-of-Thought for Logical Reasoning
Date Fragments: A Hidden Bottleneck of Tokenization for Temporal Reasoning
Distilling the Implicit Multi-Branch Structure in LLMs' Reasoning via Reinforcement Learning
Reasoning Beyond Language: A Comprehensive Survey on Latent Chain-of-Thought Reasoning
Think or Not? Selective Reasoning via Reinforcement Learning for Vision-Language Models
Do Large Language Models Excel in Complex Logical Reasoning with Formal Language?