#大语言模型
301 related articles
Choosing the Best LLM for Your Budget: A Daily-Updated Model Selection Framework
How do you pick the best LLM for your budget? This guide breaks down model selection across price, capability, and latency, with a practical tiered approach and daily-update tooling.
Hackers Are Manipulating ChatGPT and Gemini to Direct Users to Scam Websites
Hackers are manipulating ChatGPT and Gemini through data poisoning, SEO manipulation, and prompt injection to redirect users to scam sites and fake customer service numbers.
The Truth Behind Rising Fact-Check Scores: Evidence Matters More Than Answers
An arXiv study decomposes joint fact-checking score improvements, finding evidence substitution drives 9.61pp gains in strict score vs. 1.96pp from answer accuracy, exposing how aggregate metrics hide claim-level changes.
Teaching AI to Read Judicial Reasoning: A New Sentence-Level Benchmark for the German Constitutional Court
A new arXiv study introduces a sentence-level benchmark on German Constitutional Court rulings to evaluate LLMs' ability to identify interpretive canons. Average F1 scores range from 70.4 to 79.2 across four models.
Is Chain-of-Thought Actually Reasoning? Causal Measurements Reveal the Truth About CoT Faithfulness
New arXiv research uses activation-level causal intervention to measure CoT faithfulness, finding ~77% of Qwen3-4B reasoning steps are genuinely load-bearing, but behavioral tests overestimate faithfulness by up to 11–20 percentage points.
Using LLM Disagreement to Focus Expert Effort: Cross-Model Divergence Accelerates Large-Scale Annotation Codebook Revision
New arXiv research uses cross-model LLM disagreement to focus expert effort on codebook revision, achieving 64.9% annotation accuracy with rationale labeling — compressing months of work into days.
COMED: The 'Missing Middle Ground' Between Routing and Collaboration in Multi-Model Inference
COMED is a multi-LLM inference controller between routing and dense collaboration. Using anchor self-consistency, router margin, and peer probes, it achieves +10.7pp on MedQA and lifts GPT-5.5 from 23.1% to 28.1% on HLE.
Getting Claude to Show Its 'Inner Mind': An AI Self-Visualization Experiment
A Reddit user asked Claude to show its 'inner mind,' reportedly built on Opus 5.5. We break down the technical truth behind AI self-visualization and why mechanistic interpretability matters.
Mercury 2.5 Diffusion Language Model: Breaking the 770 Tokens/sec Inference Barrier
Mercury 2.5 achieves 770 tokens/sec inference throughput as a diffusion LLM. We break down the tech behind the speed, and what quality and cost data still need scrutiny.
Deep Dive into GRPO: An Efficient Alignment Method for LLM Reinforcement Learning
GRPO (Group Relative Policy Optimization) is an efficient LLM alignment method. Learn how it differs from PPO, how group-relative advantage works, and its role in reasoning models.
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.