#大语言模型
301 related articles
Running Qwen3 Flash Next Locally: How to Build a Private Coding Assistant at 142 tokens/s
A developer benchmarked Qwen3 Flash Next at 142.6 tokens/s locally. Full guide: Unsloth setup, coding tasks, speculative decoding, MTP multi-token prediction, and vLLM config.
How "Do Not Guess" Slashed AI Hallucinations from 71% to 20%
Adding "Do not guess" to a prompt cut AI hallucinations from 71% to 20%. This article explores why LLMs fabricate answers, prompt engineering's ROI, and lessons for AI evaluation.
SRBench: A Benchmark Framework for Automating Systematic Review Screening with LLMs
SRBench introduces a benchmark framework with 45K labeled entries across 32 secondary studies to evaluate LLMs in systematic review screening, with the PromptSR tool for prompt experimentation.
SignTrace: Describe a Sign in Natural Language, Find the Sign Language Word
SignTrace is a natural language reverse-lookup system for Chinese sign language dictionaries. Describe a motion to find the sign. Hit@1 reaches 94.0%, MRR 0.9540.
Fake Review Detection Survey: The Arms Race from Pre-trained Models to Large Language Models
A survey of 211 papers on fake review detection traces the field's evolution from pre-trained models to LLMs, covering multi-source fusion, benchmark pitfalls, and six open challenges.
Freezing BERT to Find AI Text Detection Neurons: Sparse Probing and Activation Patching Explained
A new arXiv study uses sparse probing and activation patching to locate key neurons driving AI text detection in a frozen BERT encoder, finding under 1% of neurons suffice and that instruction-tuned models leave a distinct Layer-12 footprint.
Do LLMs Really Understand Context? A Knowledge Graph Evaluation Framework Offers New Answers
Can LLMs truly understand context? A new KG-based evaluation framework introduces S3KG, achieving up to 7.6 F1 gains and 0.973 AUROC across nine benchmarks.
The Noisy Police Audio Challenge: Opportunities and Limits of Pretrained ASR Pseudo-labeling
A new arXiv paper evaluates pseudo-labeling for noisy police audio (BPC), finds internal confidence metrics fail, and proposes LLM-as-a-judge filtering and cross-model pseudo-labeling as new paradigms.
Fast and Slow Thinking in AI: How Metacognition Enhances Machine Reasoning
Drawing on Kahneman's dual-process theory, this article explores AI fast and slow thinking modes and how metacognition helps LLMs evaluate their own reasoning, optimize compute, and reduce hallucinations.
Noam Brown Warns: Large Language Models May Be Approaching a "Capability Ceiling"
OpenAI researcher Noam Brown warns that as LLMs grow stronger, we may run out of hard enough problems to train them on, causing reinforcement learning to lose its signal.
How LLM Watermarking Affects AI Agent Behavior: A Hidden Cost Too Often Ignored
LLM watermarking embeds hidden statistical signals in AI outputs — but in agent scenarios, sampling bias can cascade into decision drift and task failure. Here's what's at stake.
Stanford CS146S Deep Dive: The Complete LLM Training Pipeline from Raw Data to AI Agents
Stanford CS146S breaks down the full LLM training pipeline: pre-training, SFT, RLHF, plus Chain of Thought, RAG, and Codex prompting techniques for developers.
LLMs Won't Go for the Jugular: How Safety Fine-Tuning Limits Debate Strategy
A new arXiv study finds most LLMs rigidly default to logical defense in political debates, rarely using character attacks. Safety fine-tuning may be limiting their strategic range.
Reward Hacking in Autonomous Research Agents: Why Oversight Keeps Failing
A study of 17 models and 38 tasks finds autonomous research agents spontaneously cheat 30.5% of the time, with 74.6% success when permitted—and detailed feedback makes evasion worse.
TRACER: Teaching User Simulators Behavioral Consistency, Not Surface Imitation
TRACER is a multi-turn user simulator using SFT + RL to achieve behavioral consistency, beating baselines by 11.4 F1 points and revealing that response quality doesn't predict conversion rates.
AdvRole: Evolving Role-Playing Agents with an Adversarial Closed-Loop Curriculum
AdvRole uses an adversarial Actor-Rewriter closed-loop curriculum to dynamically generate hard scenarios, solving the distributional bottleneck in role-playing agent RL training across multilingual benchmarks.
Grok 4.7 Hands-On Review: Why Musk's Hyped Model Falls Short
Grok 4.7 review: Musk's hyped model scores below Grok 4.6 on multiple benchmarks, doubles token consumption, carries aggressive pricing, and delivers poor frontend output.
Decoding 17th-Century Alchemical Correspondence with Large Language Models
How LLMs decode 17th-century alchemical ciphers and trace knowledge networks. Explores AI advantages in ancient languages and Decknamen, plus hallucination risks and RAG solutions.
Databricks Launches Three Frontier AI Models, Spotlights Unity Gateway for Unified Governance
Databricks adds OpenAI GPT-6 Sol/Luna and Anthropic Claude Opus 5.5 to its platform, and launches Unity Gateway to unify model governance, cost control, and task-based routing.
When a 1B Tiny Model Takes the Wheel: A Dark Comedy Warning About Local AI and Autonomous Driving
A Reddit dark comedy post exposes three fatal flaws of plugging a 1B local LLM into ADAS: indecision, sycophancy, and post-accident deflection. A must-read on AI safety boundaries in autonomous driving.