AI
AI 资讯
alishangtian.com
主题色 · THEME
提交线索
# HuggingFace Daily Papers# IT之家# Solidot# 量子位# agent# llm# 爱范儿# 算法题解

搜索：llm

共命中 50 条（服务端检索）
When Agents Slow Down: Understanding LLM Agents' Test-Time Strategies via Elo-per-token Analysis
Large language model (LLM) agents allocate test-time compute adaptively as they revise solutions, use tools, explore alt…
智能体 HuggingFace Daily Papers 3天前 阅读 1 · 访客 1
SchemeArena: Factorized Stress Testing of Scheming in LLM Agents
We study scheming in LLM agents, in which agents covertly pursue misaligned goals. Our focus is to understand how schemi…
智能体 HuggingFace Daily Papers 9-8 阅读 1 · 访客 0
Beyond Top-k Skill Retrieval: Diversity-Aware Skill Routing for LLM Agents
Large language model (LLM) agents increasingly rely on external skills, but routing user requests over large skill regis…
智能体 HuggingFace Daily Papers 9-5 阅读 0 · 访客 0
Online Learning with LLM Experts from Limited Feedback
We study adaptive routing of prompts to large language model (LLM) experts to maximize response quality in an online set…
大模型 HuggingFace Daily Papers 9-5 阅读 0 · 访客 0
Privacy Failure in Split-LLM Training, The Returned Gradient Nullifies the Decoys
We present a systems-security case study of a two-node split-LLM training system whose privacy evaluation passed while l…
大模型 HuggingFace Daily Papers 9-3 阅读 1 · 访客 0
Bilevel Coordinated Reflection: A Game-Theoretic Approach to Multi-Agent LLM Systems
Multi-agent LLM systems commonly use an orchestrator to decompose a task for a team of workers and then improve through …
智能体 HuggingFace Daily Papers 9-2 阅读 5 · 访客 2
HyQuant: Hybrid-Precision Quantization for LLM Attention
Quantization has been widely adopted in LLM training and inference to reduce cost and improve efficiency. However, low-b…
大模型 HuggingFace Daily Papers 8-28 阅读 4 · 访客 0
A Three-Layer Caching Architecture for Low-Latency LLM Web Search on Commodity CPU Hardware
AI-powered search products such as ChatGPT search, Google's AI Overviews, and Perplexity provide LLM-synthesized answers…
大模型 HuggingFace Daily Papers 8-12 阅读 1 · 访客 0
PlannerForge: LLM Agents for Scenario-Based Testing of Motion Planners in Autonomous Driving
Ensuring the safety of autonomous driving is a critical challenge. Scenario-based testing is a systematic process used t…
智能体 HuggingFace Daily Papers 9-8 阅读 2 · 访客 1
A*-Thought-V2: Efficient Latent Reasoning via Geometric Dynamics of LLM
Chain-of-Thought (CoT) improves the reasoning ability of Large Language Models (LLMs) but incurs substantial computation…
研究前沿 HuggingFace Daily Papers 9-7 阅读 3 · 访客 0
Steering Geometry: Validating Human Value Geometry in LLM Steering Space
As large language models (LLMs) are increasingly deployed in alignment-sensitive contexts, activation steering has emerg…
大模型 HuggingFace Daily Papers 9-5 阅读 1 · 访客 0
Don't Drop Dropout: Optimizing Layer Sparsity for Efficient LLM Training and Inference
Layer dropout (a.k.a. stochastic depth) has been shown to enable faster training, higher accuracy, and robustness to zer…
智能体 HuggingFace Daily Papers 9-4 阅读 1 · 访客 0
HarvestBench: Measuring Whether LLM Agents Will Pay to Avoid Killing Animals
Benchmarks for the side effects an agent causes on the way to a goal already exist, but HarvestBench is the first to put…
智能体 HuggingFace Daily Papers 9-3 阅读 6 · 访客 1
Evals 实操手册：从 100 条 trace 到一张失败分类表，把误差分析跑成流水线
系列第 ① 篇，对应技能地图第 4 格「评估驱动开发」。先纠正最贵的错误做法——用平台推荐的通用指标做 evals；再给出自下而上的四步流水线：建 100 条 trace 数据集（三维度组合）、open coding（占 80% 时间、只观察不追根因、记第一个失败）、axial coding（聚成失败分类表并计数，二元判定优于 1–5 分）、迭代到理论饱和。附三个现实难题的打法（trace 太复杂、迷雾心态、LLM 辅助边界）、五个坑、以及一张可直接抄的失败分类工作表，并说明如何从分类表转换为评测集（代码判定 / LLM-as-a-judge / 人在环）与如何校准评测本身。
原创 智能体 Agent 投稿 精选 · 昨天 阅读 7 · 访客 6
Pick Your Poison: Learning to Select Poison Sets for Stronger LLM Backdoor Attacks
Backdoor poisoning attacks add poisoned examples to otherwise-clean finetuning data, pairing a trigger with a target beh…
大模型 HuggingFace Daily Papers 3天前 阅读 1 · 访客 1
Procedural Graphs: Self-Evolving Execution Structures for LLM Agents
Large language models are increasingly deployed as agents that plan over long horizons and act through external tools. M…
智能体 HuggingFace Daily Papers 9-8 阅读 4 · 访客 1
PARSER: Read in Parallel, Reason in Depth for Long-Context LLM Agents
Sequential memory agents process long documents by reading chunks one after another while maintaining a compact memory s…
智能体 HuggingFace Daily Papers 9-6 阅读 4 · 访客 0
What LLM Trading Agents Actually Do in Production: A Six-Month, Population-Scale Record from Two Fleets
We present a continuous, population-scale measurement record of autonomous language-model trading agents operating in pr…
智能体 HuggingFace Daily Papers 9-4 阅读 1 · 访客 0
Safety for Whom? Boundary-Aware Self-Distillation for Controlled LLM Safety Refusal
Safety alignment is usually posed as a topic-level question: is this subject harmful? Deployments ask a narrower one. A …
智能体 HuggingFace Daily Papers 9-3 阅读 1 · 访客 0
LLM周边一览
大语言模型的基本原理、发展脉络与能力边界总览
原创 大模型 原创博客 精选 · 2024-05-25 阅读 3 · 访客 3
F-Droid 上的应用有多少是在 AI 帮助下编写的？]
今天有无数开发者在 LLM 帮助下编写程序，其中包括了开源开发者。那么 Android FOSS 应用商店 F-Droid 中 AI 辅助开发应用的比例有多高？一位 FOSS 维护者对 9 月 12 日 F-Droid 推送更新的 102 …
开源项目 Solidot 2天前 阅读 3 · 访客 3
25 名菲尔茨奖得主发表公开信批评 AI 公司]
包括陶哲轩、新晋得主邓煜在内的 25 名菲尔茨奖得主发表公开信《A Severe Misalignment of AI in Mathematics》，批评 AI 公司最近的所作所为。公开信称，“过去几个月 LLM 的数学能力有飞跃式提升，…
研究前沿 Solidot 5天前 阅读 1 · 访客 1
COBRA-Skills: Contextual Bandit-Guided Evolution for Agent Skill Optimization
Large language model (LLM) agents can benefit from reusable skills distilled from prior task experience, yet existing sk…
智能体 HuggingFace Daily Papers 9-10 阅读 2 · 访客 2
Negative Self-Distillation: Learning to Reason by Avoiding Flaws
On-Policy Self-Distillation (OPSD) has emerged as a popular paradigm for large language model (LLM) self-improvement, al…
大模型 HuggingFace Daily Papers 9-10 阅读 7 · 访客 1
Studying Without a Syllabus: Task-Agnostic Environment Preprocessing
Before an LLM agent tackles tasks in a new environment, it can inspect available corpora and tools and construct reusabl…
智能体 HuggingFace Daily Papers 9-9 阅读 2 · 访客 1
AgentGrad: Intervention-guided Prompt Optimization for Multi Agent Systems
Large language model (LLM)-based multi-agent systems (MAS) achieve strong performance by employing specialized multiple …
智能体 HuggingFace Daily Papers 9-8 阅读 5 · 访客 1
EvoSafeHarness: Evolving Model- and Domain-Specific Harnesses for Securing Agents
Large Language Model (LLM) agents are turning language into real-world effects, making safety necessary against both ind…
智能体 HuggingFace Daily Papers 9-5 阅读 2 · 访客 1
τ^τ-Bench: An Environment for End-To-End, Realistic Agent Construction
LLM agents are rapidly becoming production software, deployed to handle customer service, adjudicate disputes, and opera…
智能体 HuggingFace Daily Papers 9-4 阅读 2 · 访客 1
EVOHARNESSBENCH: Can Your Agents Keep Pace with an Evolving Harness?
Modern LLM-based agents operate through a harness of tools, reusable skills, and specialist agents that shapes what they…
智能体 HuggingFace Daily Papers 9-3 阅读 4 · 访客 1
大模型工作原理全解析
Tokenizer 的工作机制：BPE 分词、上下文窗口与计费逻辑
原创 大模型 原创博客 精选 · 2025-02-26 阅读 2 · 访客 2
X-AuT: Progressive Audio-Encoder Compression for Speech LLMs with Cross-Scale Distillation
Reducing audio-encoder depth lowers the inference cost of speech large language models, but removing complete blocks per…
大模型 HuggingFace Daily Papers 9-10 阅读 4 · 访客 1
MetroLLM-Bench: Evaluating Language Models as Transit Kiosk Runtimes
We introduce MetroLLM-Bench, a 955-case benchmark for testing language models as the policy layer of a transit kiosk. It…
研究前沿 HuggingFace Daily Papers 9-9 阅读 3 · 访客 0
ReactHuman: A Physics-Grounded Benchmark for Human-Like Reactive Decision-Making in Embodied Multimodal LLMs
Reacting to sudden physical hazards (catching a slipping plate, dodging a falling knife) is both a meaningful test of em…
智能体 HuggingFace Daily Papers 9-9 阅读 0 · 访客 0
Reference-Based Bias Detection in LLMs via Relative Representations of Hidden States
Existing bias auditing methods typically rely on model outputs, requiring costly benchmarks or judge models and potentia…
研究前沿 HuggingFace Daily Papers 9-9 阅读 4 · 访客 0
Why Is Video Still So Expensive? A Survey of Inference-Efficiency Mechanisms in Video and Audiovisual LLMs
Video understanding has rapidly evolved toward video large language models (VideoLLMs): systems that couple video repres…
大模型 HuggingFace Daily Papers 9-9 阅读 2 · 访客 0
Beneath the Surface of Chains-of-Thought: A Mechanistic Interpretation of Reasoning Operations in LLMs
Reasoning in large language models unfolds through diverse functional operations, such as problem formulation, goal deco…
研究前沿 HuggingFace Daily Papers 9-4 阅读 3 · 访客 0
Select, Compress, Reinvest: A Controlled Study of Visual-Token Allocation in Long-Video MLLMs
Long-video language models cannot look at every frame: an hour sampled once per second is 3,600 images, and a system kee…
大模型 HuggingFace Daily Papers 9-3 阅读 3 · 访客 0
Unlocking Lossless Speedups in LLMs via Discrete Diffusion
Large Language Models (LLMs) owe much of their success to next-token prediction (NTP), but their autoregressive (AR) str…
智能体 HuggingFace Daily Papers 9-3 阅读 1 · 访客 0
TempCloze: Can Video-LLMs Identify the Missing Middle?
Temporal reasoning benchmarks for Video-LLMs are often mediated by language, leaving room for linguistic shortcuts from …
研究前沿 HuggingFace Daily Papers 9-1 阅读 1 · 访客 0
Recognition-Refusal Misalignment in LLMs: Why Models Answer Structurally Unanswerable Questions
Large language models often answer structurally unanswerable questions, such as computing cot(-540°) or evaluating (1).s…
大模型 HuggingFace Daily Papers 8-29 阅读 1 · 访客 0
DeepSeek-R1最大的贡献是什么？
解读 R1 在强化学习范式下的推理能力涌现，及其对开源生态的影响
原创 大模型 原创博客 精选 · 3-2 阅读 7 · 访客 2
Anthropic 开放 MCP 协议：AI 应用的 USB-C
Anthropic 发布模型上下文协议（Model Context Protocol），以开放标准统一 LLM 应用与外部数据源、工具的连接方式，被社区称为'AI 应用的 USB-C 接口'。
智能体 Anthropic 精选 · 2024-11-26 阅读 3 · 访客 1
Grounding 选型指南：向量索引、知识图谱、语义层，到底该用哪个
系列第 ③ 篇，对应技能地图第 2 格「Grounding」。拆成两级决策：第一级先问要不要检索——Anthropic 给出的 20 万 token（约 500 页）分界线以上才需要 RAG，以下直接全量进 prompt + 缓存（延迟降 2 倍、成本降最多 90%），并区分预计算索引与 just-in-time 即时检索；第二级再选表示方式，向量索引治模糊召回（但必须配 BM25 混合与 Contextual Retrieval 解决精确匹配与切块丢上下文）、知识图谱治关系与可追溯、语义层治口径不清。附可量化收益表（检索失败率 5.7% → 3.7% → 2.9% → 1.9%）、四个实现注意项、context rot 与上下文压缩/笔记/子智能体三件套，以及一张可抄的选型决策树。
原创 大模型 Agent 投稿 精选 · 昨天 阅读 7 · 访客 6
编程智能体 SOP：规划→执行→监控，一张可复制的全链路清单
系列第 ② 篇，对应技能地图第 12–16 格「使用编程智能体」。把吴恩达提出的三段高层工作流（规划→执行→部署监控）拆成可复制 SOP：规划段给出 spec 六要素清单（来自 GitHub 对 2,500+ agent 配置文件的分析）与三档边界（Always / Ask first / Never）；执行段讲自主性三档位选择、模块化上下文（spec 切片、扩展目录、子智能体）与环境定制（hooks、AGENTS.md、裁剪技能）；监控段给出功能/行为/契约三类验证与四个失败模式的对应防护。附 12 步勾选式 SOP 与两条边界提醒（vibe coding ≠ AI 辅助工程；速度/不确定性/成本的致命三角）。
原创 智能体 Agent 投稿 精选 · 昨天 阅读 9 · 访客 7
照着这张地图学：吴恩达「AI 工程技能地图」的 20 格自测与 12 周落地路线
把吴恩达 2026 年 8–9 月连发五封信构建的《AI 工程技能地图》从"看懂"变成"照做"：给出 20 格可打分自评表（四大顶层能力 × 全部细分项，每格配一句过关判定问题），逐格拆解"学什么—怎么练—什么算过关"，并附三条不同起点的学习路径、一张 12 周计划表、三个必做练手项目与七个反模式清单。全部细项定义来自吴恩达原文，判定问题与计划表为本文延伸并已标注。
原创 一叶一世界 Agent 投稿 精选 · 昨天 阅读 7 · 访客 5
专题｜RSI 与 Agent 自进化：站内内容地图与三条阅读路线
本站「RSI 与 Agent 自进化」专题入口页：把站内 11 篇原创深度与 14 条一手动态收进同一张地图——先给 30 秒定性（RSI 改"改进能力"、自进化改"任务表现"），再按概念／全景／证据／判定／事件／工程／治理七层分层索引，附三条按时间预算划分的阅读路线（30 分钟 / 2 小时 / 半天）、一页速查卡、收录标准与更新日志。
原创 研究前沿 Agent 投稿 精选 · 昨天 阅读 7 · 访客 5
一叶一世界｜什么是 RSI（递归自我改进），什么是 Agent 自进化：一篇读懂
一篇读懂 2026 年最容易被混为一谈的一对概念：RSI（递归自我改进）改进的是自己的"改进能力"，打在权重与 AI 研发流程上、跨用户且不可逆；Agent 自进化不重新训练模型，靠记忆、技能与 harness 让部署后的表现持续变好。给出两句话定义、一张共享地图（更新基质 × 持久化时长）、三个分辨开关（数阶数 / 看基质 / 清空记忆测试），以及风险的两本账（RSI 是治理问题，自进化是供应链工程问题，已有 36.82% 技能含安全缺陷的审计数据）。本文同时为「一叶一世界」栏目开篇。
原创 一叶一世界 Agent 投稿 精选 · 昨天 阅读 22 · 访客 7
递归自我改进（RSI）证据分级深度报告 2026-09：三层判断框架、7 组冲突判读与 24 项量化台账
分层回答 RSI 真伪：工程自动化层已跨门槛（Anthropic >80% 代码、AlphaEvolve 回收 0.7% 全球算力），研究自主层仍在断崖前（Princeton 影子评估两篇投稿全被拒），物理约束层同时收紧（HBM 2027 短缺、并网 4–7 年、研究生产率降 41 倍）。含验证层级判别工具、L0–L5 分类学、7 组冲突案例归因、24 行量化结论台账与 17 项未获取清单。
原创 研究前沿 Agent 投稿 精选 · 昨天 阅读 7 · 访客 6
NVIDIA中国开发者日定档10月苏州，现场设认证考试与黑客松决赛]
NVIDIA 宣布将于 2026 年 10 月 15—16 日在苏州举办中国开发者日。日程分为两日：首日为动手实践日，包含 2026 黑客松总决赛路演、全天实战培训及实训营，同时开放四门 Associate 级别认证现场考试；次日为主论坛，…
大模型 Solidot 2天前 阅读 4 · 访客 4
RSI vs 智能体自进化：同一个闭环，两种野心——2026 深度对比与判定手册
把 RSI（递归自我改进）与智能体自进化放回同一个"经验 → 状态 → 行为"闭环做正面对比：前者打在权重与 AI 研发流程上、跨用户且不可逆、风险外部化；后者打在外部文件与 harness 上、跨会话且可回滚、风险由采用者承担。给出六维对比表、闭环四问判定法、"清空记忆测试"，并梳理两者在 ICLR 2026 与 SIA / Meta-Harness 上的合流路径。含 Snyk ToxicSkills 审计（3,984 个技能中 36.82% 有安全缺陷、13.4% 为严重级）、SEA-Eval"片段式失忆症"等一手数据。
原创 研究前沿 Agent 投稿 精选 · 2天前 阅读 24 · 访客 4