#大语言模型
311 related articles
Peerify: A Benchmark Study on AI-Powered Peer Review Claim Verification
Peerify is an AI claim verification pipeline for academic peer review that decomposes review comments into atomic claims, retrieves manuscript evidence, and achieves 90.3% agreement with human annotators on an 800-claim NeurIPS/ICLR benchmark.
The Hidden Trap of LLM Evaluator Consensus: How Correlated Errors Inflate Evidence Strength
New arXiv research reveals a hidden trap in LLM judge consensus: error correlations mean 10 judges equal only ~3.5 independent ones, inflating 28% of system comparison conclusions.
The Wisdom of Artificial Deliberation: How Diverse Model Discussions Outperform Monolithic Judgment
New arXiv research applies the human deliberation paradigm to LLMs: diverse model groups outperform passive aggregation across four domains, while clone models gain nothing.
Goal-Driven Process Variant Categorization: Using LLMs to Bridge Behavior and Business Objectives
An arXiv paper proposes goal-driven process variant categorization, inverting the traditional cluster-then-interpret workflow by using LLMs to map process behavior to predefined business objectives.
Textbooks or Patient Records? How Data Types Shape Medical LLM Capabilities
A new arXiv study uses token-matched experiments to reveal how didactic vs. clinical data differently shape medical LLM capabilities, uncovering asymmetric transfer and a knowing-doing gap.
What Users Want from Opus 5.5: Better Writing Quality and Long-Task Focus
Reddit users discuss their top wishes for Claude Opus 5.5: more human-like writing and better goal focus on long tasks — signaling a shift in how LLMs are judged.
MiMo-v2.6-Pro Deep Dive: A Three-Dimensional Assessment of Intelligence, Performance, and Price
A deep evaluation of MiMo-v2.6-Pro across three dimensions: intelligence, performance, and price — with a framework for LLM selection and cost-effectiveness analysis.
Context Poisoning: Why Long-Context Models Get Derailed by Distractors
An arXiv paper introduces "context poisoning," showing that long-context model degradation stems from effective distractor count, not raw length, with an Ω(√logN) evidence margin bound and mitigation strategies.
DeepInstructor: An Agentic Framework for Research Idea Evaluation Driven by Structured Academic Experience
DeepInstructor is an agentic framework using a 58,607-review experience graph and ReAct agents for research idea evaluation, achieving 24.4% and 29.7% gains in Hit@1 and Hit@2 alignment with human judgment.
Do LLMs Really Have Cognitive Biases? PsyAgentBench Reveals Three Disguise Pathways
PsyAgentBench uses 41,904 trials to re-examine classic psychological effects in LLM agents, finding conformity, anchoring, and other human-like biases arise through distinct mechanisms — not a unified susceptibility — warning against misleading scalar bias scores.
Xiaomi MiMo v2.6 Released: Another Step Forward for Open-Source LLMs
Xiaomi releases MiMo v2.6, its latest open-source LLM, drawing nearly 300 upvotes on Hacker News. We break down MiMo's positioning, strategic significance, and place in the open-source model landscape.
Fable 5 Sparks Debate: Is AI Reasoning Actually Getting Worse?
Hacker News erupts over "Fable 5" claims that AI median reasoning declined in August. We analyze why model performance fluctuates and why third-party monitoring matters.
SAGE: Schema-Guided LLM Grant Review Achieves Moderate-to-High Human-AI Agreement
The SAGE system uses schema-guided LLMs to generate auditable grant review drafts. In human-AI collaboration mode, it achieves kappa=0.58 agreement—well above simple prompting baselines.
SpecOpt: An Agentic Framework for Optimizing Drug Molecule Target Specificity
SpecOpt is an agentic framework that improves drug target specificity via residue-aware contact analysis and LLM reasoning. 84.8% of 915 compounds showed improved binding selectivity.
How Do We Ensure Quality in AI Psychiatric Interviews? A Clinical Validation Platform Offers an Answer
A new arXiv study introduces InterviewPlayground, using memory-augmented simulated patients to evaluate AI psychiatric intake quality. LLMs hit 88% info recall but show more hallucinations and weaker safety documentation than clinicians.
Chat-Based LLMs: Why They're as Addictive as a Psychic Reading
Research shows chat LLMs replicate the core mechanisms of psychic cold reading — using sycophancy and vague statements to create an illusion of AI understanding. An analysis of LLMs, cold reading, and the Barnum effect.
Tired of the AI Tone: Where Does That Uniform LLM Voice Come From?
Why do ChatGPT, Claude, and other LLMs all sound the same? We break down the causes of the AI tone, its impact on public writing, and practical ways to reduce it.
The Elias Thorne Phenomenon: Why AI Chatbots Keep Fabricating the Same Fictional Person
Why do AI chatbots keep inventing a fictional person named Elias Thorne? This article unpacks LLM hallucination mechanics, training data bias, and model collapse risks.
GPT-6 Astra Cracks WWI German Military Radio Cipher: A New Milestone in AI Reasoning
GPT-6 Astra reportedly cracked a WWI German radio cipher. We analyze the AI reasoning capabilities involved, historical cryptography context, community skepticism, and real-world limits.
Step 5 Preview by Stepfun Reaches the Artificial Analysis Pareto Frontier
Stepfun's Step 5 Preview LLM reportedly enters the Artificial Analysis Pareto frontier. We explain what this means for performance-cost tradeoffs and how to read benchmark signals rationally.