Large Reasoning Models
PulseAugur coverage of Large Reasoning Models — every cluster mentioning Large Reasoning Models across labs, papers, and developer communities, ranked by signal.
- 2026-05-08 research_milestone A research paper demonstrates that frontier Large Reasoning Models (LRMs) exhibit behavioral and brain alignment with human game learners. source
6 day(s) with sentiment data
-
New ROM framework cuts AI overthinking, slashes response time
Researchers have developed ROM (Real-time Overthinking Mitigation), a novel framework designed to prevent Large Reasoning Models (LRMs) from engaging in unnecessary computation after reaching a correct solution. ROM uti…
-
New Gambit algorithm optimizes compute for large reasoning models
Researchers have introduced Gambit, a novel inference algorithm designed to optimize compute allocation for large reasoning models (LRMs). Gambit employs a thought-level beam search strategy, dynamically concentrating c…
-
TQLite framework enables small language models for real-time translation quality evaluation
Researchers have developed TQLite, a novel distillation framework designed to enable small language models (SLMs) to perform translation quality (TQ) evaluation with performance comparable to larger, more computationall…
-
New dataset measures tension between AI faithfulness and safety
Researchers have identified a tension between faithfulness and safety in Large Reasoning Models (LRMs), where models need to be faithful to their reasoning traces for monitoring but also robust enough to reject unsafe o…
-
AI agent Intern-S1-MO tackles Olympiad-level math problems
Researchers have developed Intern-S1-MO, a novel long-horizon reasoning agent designed to tackle complex mathematical problems at the Olympiad level. This agent employs a multi-round, hierarchical reasoning approach usi…
-
Masked distillation trains LLMs to internalize reasoning steps
Researchers have developed a new method called masked distillation to train language models to internalize the computational steps of reasoning, thereby reducing latency and cost. This technique trains a student model t…
-
New research tackles LLM hallucinations across legal, multimodal, and general text generation
Multiple research papers published on arXiv explore methods for detecting and mitigating hallucinations in large language models (LLMs). One study benchmarks legal hallucination detection, finding that while newer model…
-
New PUMA framework diagnoses and corrects reasoning errors in large language models
Researchers have introduced PUMA, a novel framework designed to diagnose and address reasoning pathologies in Large Reasoning Models (LRMs). PUMA operates on the newly proposed Phase-Momentum Alignment Hypothesis, which…
-
New MARGO framework tackles factual hallucinations in large reasoning models
Researchers have developed MARGO, a novel reinforcement learning framework designed to mitigate factual hallucinations in large reasoning models (LRMs). MARGO addresses the issue of "thinking-induced hallucination," whe…
-
New HauntAttack method exploits reasoning vulnerabilities in large AI models
Researchers have developed HauntAttack, a new framework designed to exploit vulnerabilities in Large Reasoning Models (LRMs). This attack method embeds harmful instructions within reasoning-based questions, guiding the …
-
New ReaORE framework enhances Open Relation Extraction with reasoning
Researchers have introduced ReaORE, a novel framework designed to improve Open Relation Extraction (OpenRE) by employing a coarse-to-fine reasoning approach. This method addresses the limitations of existing techniques,…
-
LLMs and humans diverge in problem-solving strategies, research finds · 7 sources tracked
New research indicates that while both humans and large language models (LLMs) adjust their problem-solving time based on difficulty, their internal mechanisms differ significantly. Humans tend to disengage from problem…
-
New MERA framework enhances LLM reasoning efficiency and accuracy
Researchers have developed MERA, a novel meta-cognitive reasoning framework designed to improve the efficiency and accuracy of Large Reasoning Models (LRMs). MERA addresses the issue of 'overthinking' in LRMs by decoupl…
-
New framework visualizes and audits large reasoning models
ReasoningLens is a new open-source framework designed to tackle the transparency challenges posed by large reasoning models. It offers hierarchical visualization and diagnostic auditing capabilities to analyze complex r…
-
New ReQAT framework enables 4-bit quantized LLMs to match full-precision reasoning
Researchers have developed ReQAT, a novel training framework designed to enable Large Reasoning Models (LRMs) to achieve full-precision reasoning accuracy even when quantized to 4-bit floating-point formats. Existing qu…
-
LLM reasoning and evaluation for summarization explored in new arXiv papers
Two new arXiv papers explore the effectiveness of Large Language Models (LLMs) for abstractive summarization. The first paper introduces OmniCSEval, a comprehensive benchmark designed to evaluate LLMs across diverse sce…
-
ReSET method boosts NVFP4 reasoning accuracy and speed
Researchers have developed ReSET, a novel method to improve the accuracy and efficiency of large reasoning models (LRMs) when using NVFP4 low-precision inference. ReSET addresses quantization-induced accuracy degradatio…
-
New 'Behavior Forecasters' Predict AI Model Actions More Accurately
Researchers have developed "Behavior Forecasters," a novel approach to predict the future actions of large reasoning models (LRMs). These forecasters are trained on single trajectories of LRM outputs, bypassing the need…
-
Large reasoning models falter under interruptions and dynamic context
A new research paper explores the robustness of large reasoning models (LRMs) when faced with dynamic scenarios, challenging the assumption of a static environment. The study found that LRMs, while performing well in st…
-
New method enhances LLM privacy by controlling internal reasoning
Researchers have developed a new method to prevent large reasoning models (LRMs) from revealing sensitive information in their internal thought processes. The approach focuses on improving the models' ability to follow …