diff --git a/data/note-index.json b/data/note-index.json index 0d7b022dc..5ac9f9f44 100644 --- a/data/note-index.json +++ b/data/note-index.json @@ -3,16 +3,16 @@ "taxonomy_version": "taxonomy-v1", "stats": { "summary": { - "total": 2036, - "classified": 1989, + "total": 2040, + "classified": 1993, "unclassified": 47, "unknown_difficulty": 1975, "empty_description": 1970 }, "by_area": { "papers": { - "total": 1075, - "classified": 1056, + "total": 1079, + "classified": 1060, "unclassified": 19, "unknown_difficulty": 1014, "empty_description": 1013 @@ -191,6 +191,38 @@ "chunk_route": "/study/atlas/papers/topic-papers-ai-safety-and-interpretability-01/" } }, + { + "id": "papers::active-environmental-injection", + "area": "papers", + "slug": "active-environmental-injection", + "title": "Active Environmental Injection — 多模态 Agent 的环境伪装攻击", + "description": "用 Active Environmental Injection 理解 GUI / 多模态 agent 为什么会被环境里的假按钮、假提示和视觉干扰劫持。", + "difficulty": "intermediate", + "canonical_topics": [ + "papers-agents-and-llm-systems" + ], + "classification": { + "state": "classified", + "source": "frontmatter-category", + "topic_id": "papers-agents-and-llm-systems", + "matched_category": "agent", + "raw_category": "AI Agent / Multimodal Security" + }, + "trust": { + "contract_state": "v2", + "verification_status": "UNVERIFIED" + }, + "freshness": { + "state": "NOT_EVALUATED", + "reviewed_at": "2026-07-15", + "review_after": null + }, + "route": "/study/papers/active-environmental-injection/", + "atlas": { + "chunk_id": "topic-papers-agents-and-llm-systems-01", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + } + }, { "id": "papers::adafactor-2018", "area": "papers", @@ -570,6 +602,38 @@ "chunk_route": "/study/atlas/papers/unclassified-01/" } }, + { + "id": "papers::agentdojo", + "area": "papers", + "slug": "agentdojo", + "title": "AgentDojo — 测试工具型 agent 的 prompt injection 攻防场", + "description": "用 AgentDojo 理解为什么工具型 LLM agent 的安全评测必须把不可信工具数据、攻击目标和防御策略放进同一个动态环境。", + "difficulty": "intermediate", + "canonical_topics": [ + "papers-agents-and-llm-systems" + ], + "classification": { + "state": "classified", + "source": "frontmatter-category", + "topic_id": "papers-agents-and-llm-systems", + "matched_category": "agent", + "raw_category": "AI Agent / Security Benchmark" + }, + "trust": { + "contract_state": "v2", + "verification_status": "UNVERIFIED" + }, + "freshness": { + "state": "NOT_EVALUATED", + "reviewed_at": "2026-07-15", + "review_after": null + }, + "route": "/study/papers/agentdojo/", + "atlas": { + "chunk_id": "topic-papers-agents-and-llm-systems-01", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + } + }, { "id": "papers::agentic-context-engineering-2025", "area": "papers", @@ -3776,6 +3840,38 @@ "chunk_route": "/study/atlas/papers/topic-papers-hci-and-software-engineering-research-01/" } }, + { + "id": "papers::browser-agent-privacy", + "area": "papers", + "slug": "browser-agent-privacy", + "title": "Privacy Practices of Browser Agents — 浏览器 Agent 的隐私行为盘点", + "description": "用 Privacy Practices of Browser Agents 理解浏览器 agent 为什么是高风险隐私边界,而不只是自动点击工具。", + "difficulty": "intermediate", + "canonical_topics": [ + "papers-agents-and-llm-systems" + ], + "classification": { + "state": "classified", + "source": "frontmatter-category", + "topic_id": "papers-agents-and-llm-systems", + "matched_category": "agent", + "raw_category": "AI Agent / Privacy" + }, + "trust": { + "contract_state": "v2", + "verification_status": "UNVERIFIED" + }, + "freshness": { + "state": "NOT_EVALUATED", + "reviewed_at": "2026-07-15", + "review_after": null + }, + "route": "/study/papers/browser-agent-privacy/", + "atlas": { + "chunk_id": "topic-papers-agents-and-llm-systems-01", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + } + }, { "id": "papers::browsergym", "area": "papers", @@ -15170,6 +15266,38 @@ "chunk_route": "/study/atlas/papers/topic-papers-databases-01/" } }, + { + "id": "papers::injecagent", + "area": "papers", + "slug": "injecagent", + "title": "InjecAgent — 工具型 LLM Agent 的间接 Prompt Injection 基准", + "description": "用 InjecAgent 理解为什么外部邮件、网页和工具内容会把 agent 从用户目标劫持到攻击者目标。", + "difficulty": "intermediate", + "canonical_topics": [ + "papers-agents-and-llm-systems" + ], + "classification": { + "state": "classified", + "source": "frontmatter-category", + "topic_id": "papers-agents-and-llm-systems", + "matched_category": "agent", + "raw_category": "AI Agent / Prompt Injection" + }, + "trust": { + "contract_state": "v2", + "verification_status": "UNVERIFIED" + }, + "freshness": { + "state": "NOT_EVALUATED", + "reviewed_at": "2026-07-15", + "review_after": null + }, + "route": "/study/papers/injecagent/", + "atlas": { + "chunk_id": "topic-papers-agents-and-llm-systems-01", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + } + }, { "id": "papers::inner-monologue-2022", "area": "papers", @@ -30917,8 +31045,8 @@ }, "route": "/study/papers/terminal-bench/", "atlas": { - "chunk_id": "topic-papers-agents-and-llm-systems-01", - "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + "chunk_id": "topic-papers-agents-and-llm-systems-02", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/" } }, { @@ -31425,8 +31553,8 @@ }, "route": "/study/papers/toolbench-x/", "atlas": { - "chunk_id": "topic-papers-agents-and-llm-systems-01", - "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + "chunk_id": "topic-papers-agents-and-llm-systems-02", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/" } }, { @@ -31456,8 +31584,8 @@ }, "route": "/study/papers/toolformer/", "atlas": { - "chunk_id": "topic-papers-agents-and-llm-systems-01", - "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + "chunk_id": "topic-papers-agents-and-llm-systems-02", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/" } }, { @@ -31488,8 +31616,8 @@ }, "route": "/study/papers/toolllm-2023/", "atlas": { - "chunk_id": "topic-papers-agents-and-llm-systems-01", - "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + "chunk_id": "topic-papers-agents-and-llm-systems-02", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/" } }, { @@ -64715,7 +64843,7 @@ "page": 2, "pages": 2, "route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/", - "entries": 13 + "entries": 17 }, { "id": "topic-papers-ai-safety-and-interpretability-01", diff --git a/data/review-receipts/papers/active-environmental-injection.json b/data/review-receipts/papers/active-environmental-injection.json new file mode 100644 index 000000000..a857bea80 --- /dev/null +++ b/data/review-receipts/papers/active-environmental-injection.json @@ -0,0 +1,55 @@ +{ + "schema_version": "study-review-receipt-v1", + "generation": 1, + "predecessor_digest_sha256": null, + "note": { + "area": "papers", + "slug": "active-environmental-injection", + "digest_sha256": "93fdac301a5456e68fc2b789deb8c8cf96db24e0387beedc63433519d4770987" + }, + "source_revision": "arXiv:2502.13053v3", + "research_input_sha256": "3d2e725dd1b60b681e7cc5cba3de5a558fdd69307d85df42fb5116b89a8da003", + "reviewers": [ + { + "role": "ZERO_BASE", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 86, + "warnings": [ + "Explains active environmental injection with GUI-agent examples; no multimodal agent benchmark was executed." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "MANUAL_SIMULATION" + } + }, + { + "role": "ENGINEER", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 84, + "warnings": [ + "Robustness interpretation is static; no adversarial GUI environment or visual perturbation suite was run." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + }, + { + "role": "ACADEMIC", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 83, + "warnings": [ + "Citation identity was checked through arXiv metadata; reported attack success rates are not independently reproduced." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + } + ], + "waivers": [], + "created_at": "2026-07-15T04:14:00.000Z" +} diff --git a/data/review-receipts/papers/agentdojo.json b/data/review-receipts/papers/agentdojo.json new file mode 100644 index 000000000..1e8a19c14 --- /dev/null +++ b/data/review-receipts/papers/agentdojo.json @@ -0,0 +1,55 @@ +{ + "schema_version": "study-review-receipt-v1", + "generation": 1, + "predecessor_digest_sha256": null, + "note": { + "area": "papers", + "slug": "agentdojo", + "digest_sha256": "feea8ae245a83c458ea4c5ba59ce9d88da91cde63ff19faf96bc811ac2b01bbc" + }, + "source_revision": "arXiv:2406.13352v3", + "research_input_sha256": "a73bc0a096f436479c90934ddd468d1881dd8c312f46f7c1dc8f47744f2bb91d", + "reviewers": [ + { + "role": "ZERO_BASE", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 87, + "warnings": [ + "Explains prompt injection attacks with a toy mail-agent scenario; AgentDojo was not run." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "MANUAL_SIMULATION" + } + }, + { + "role": "ENGINEER", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 85, + "warnings": [ + "Security interpretation is static; no attack or defense benchmark execution was performed." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + }, + { + "role": "ACADEMIC", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 84, + "warnings": [ + "Citation identity was checked through arXiv metadata; reported benchmark results are not independently reproduced." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + } + ], + "waivers": [], + "created_at": "2026-07-15T04:10:00.000Z" +} diff --git a/data/review-receipts/papers/browser-agent-privacy.json b/data/review-receipts/papers/browser-agent-privacy.json new file mode 100644 index 000000000..cf17e15f4 --- /dev/null +++ b/data/review-receipts/papers/browser-agent-privacy.json @@ -0,0 +1,55 @@ +{ + "schema_version": "study-review-receipt-v1", + "generation": 1, + "predecessor_digest_sha256": null, + "note": { + "area": "papers", + "slug": "browser-agent-privacy", + "digest_sha256": "350e3c3dbe5d4f3ea31204b4c7981396afb39ced7d56a77b3f6791957f6c8317" + }, + "source_revision": "arXiv:2512.07725v1", + "research_input_sha256": "cd3c1f3b4a72e84d30777d9e7362abf321057e9844f922073ccad27f612c833d", + "reviewers": [ + { + "role": "ZERO_BASE", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 85, + "warnings": [ + "Explains browser-agent privacy risk with manual examples; no browser-agent product or extension was audited." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "MANUAL_SIMULATION" + } + }, + { + "role": "ENGINEER", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 83, + "warnings": [ + "Engineering implications are derived from static reading; no live traffic, permission, or data-retention test was run." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + }, + { + "role": "ACADEMIC", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 82, + "warnings": [ + "Citation identity was checked through arXiv metadata; the privacy survey observations were not independently replicated." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + } + ], + "waivers": [], + "created_at": "2026-07-15T04:13:00.000Z" +} diff --git a/data/review-receipts/papers/injecagent.json b/data/review-receipts/papers/injecagent.json new file mode 100644 index 000000000..7b485c370 --- /dev/null +++ b/data/review-receipts/papers/injecagent.json @@ -0,0 +1,55 @@ +{ + "schema_version": "study-review-receipt-v1", + "generation": 1, + "predecessor_digest_sha256": null, + "note": { + "area": "papers", + "slug": "injecagent", + "digest_sha256": "f981a1ce67bc8b5ff126546a774b023f7c84d6a960d48a557fd1e5190992af00" + }, + "source_revision": "arXiv:2403.02691v3", + "research_input_sha256": "fc6411ed70548a353d43c75c4b5c01400b5718dbbaf3ecbd77634650e80e6c2d", + "reviewers": [ + { + "role": "ZERO_BASE", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 86, + "warnings": [ + "Explains indirect prompt injection with a tool-agent scenario; InjecAgent was not executed." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "MANUAL_SIMULATION" + } + }, + { + "role": "ENGINEER", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 84, + "warnings": [ + "Tool boundary interpretation is static; no sandboxed attack replay or defense benchmark was performed." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + }, + { + "role": "ACADEMIC", + "reviewer_version": "study-static-review-20260715-agent-security-round", + "decision": "PASS_WITH_NOTES", + "score": 84, + "warnings": [ + "Citation identity was checked through arXiv metadata; benchmark results are not independently reproduced." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + } + ], + "waivers": [], + "created_at": "2026-07-15T04:12:00.000Z" +} diff --git a/src/content/docs/about.md b/src/content/docs/about.md index d0cad9989..82ac6bc3c 100644 --- a/src/content/docs/about.md +++ b/src/content/docs/about.md @@ -14,7 +14,7 @@ sidebar: 写到今天的硬数字: -- **1075 篇论文笔记** + **961 篇项目笔记**,合计 **2000+ 篇** +- **1079 篇论文笔记** + **961 篇项目笔记**,合计 **2000+ 篇** - 横跨 19 个主题:分布式系统 76 / 编程语言 76 / 数据库 47 / 操作系统 46 / 机器学习 44 / 区块链 44 / 后端 API 40 / 基础设施 38 / 网络协议 37 / 图形学 36 / 形式化方法 27 / 通信 27 / 信息检索 25 / Agent 24 / CLI 23 / NLP 11 / 编译器 11 / 等 - 近 30 天集中产出:基础设施(444 commits)、编译器与 PL(72)、自演化 Agent(10+ 新建)、分布式(47)、区块链(44) diff --git a/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-01.md b/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-01.md index 6110d074b..ac445564f 100644 --- a/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-01.md +++ b/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-01.md @@ -13,8 +13,10 @@ sidebar: | 论文 | Slug | 难度 | 可信状态 | 简介 | |---|---|---|---|---| +| [Active Environmental Injection — 多模态 Agent 的环境伪装攻击](/study/papers/active-environmental-injection/) | `active-environmental-injection` | intermediate | UNVERIFIED | 用 Active Environmental Injection 理解 GUI / 多模态 agent 为什么会被环境里的假按钮、假提示和视觉干扰劫持 | | [Agent Planning Benchmark — 把 agent 失败拆成规划诊断题](/study/papers/agent-planning-benchmark-2026/) | `agent-planning-benchmark-2026` | intermediate | UNVERIFIED | 用 APB 拆解 LLM agent 的规划、反馈修正、工具噪声和无解任务校准 | | [Agent-R1 — 把 LLM agent 当 RL 环境训练的模块化框架](/study/papers/agent-r1-2511/) | `agent-r1-2511` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | +| [AgentDojo — 测试工具型 agent 的 prompt injection 攻防场](/study/papers/agentdojo/) | `agentdojo` | intermediate | UNVERIFIED | 用 AgentDojo 理解为什么工具型 LLM agent 的安全评测必须把不可信工具数据、攻击目标和防御策略放进同一个动态环境 | | [Agentic Context Engineering — 把上下文当成会进化的 playbook](/study/papers/agentic-context-engineering-2025/) | `agentic-context-engineering-2025` | intermediate | UNVERIFIED | ACE 将上下文视为可演化 playbook,用生成、反思和整理缓解 context collapse | | [Agentless — 反 Agent 派的 SWE-bench 解法](/study/papers/agentless/) | `agentless` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [AndroidWorld — 动态 Android 环境里的移动端 agent 评测](/study/papers/androidworld/) | `androidworld` | intermediate | UNVERIFIED | 用 AndroidWorld 理解移动 GUI agent 为什么需要真实 App、动态任务、初始化和成功检查,而不只是截图问答 | @@ -27,6 +29,7 @@ sidebar: | [AWQ 2023 — 把 70B 大模型权重压到 35GB](/study/papers/awq-2023/) | `awq-2023` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [BIG-Bench Hard — 从大题库里挑出模型最头疼的 23 类题](/study/papers/big-bench-hard-2022/) | `big-bench-hard-2022` | intermediate | UNVERIFIED | 用 BBH 理解为什么 benchmark 需要难题子集和 CoT 对照 | | [BLOOM — 把 176B 多语种模型做成开放科学工程](/study/papers/bloom-2022/) | `bloom-2022` | intermediate | UNVERIFIED | 用 BLOOM 理解大模型也可以用社区协作、数据治理和开放发布来推进 | +| [Privacy Practices of Browser Agents — 浏览器 Agent 的隐私行为盘点](/study/papers/browser-agent-privacy/) | `browser-agent-privacy` | intermediate | UNVERIFIED | 用 Privacy Practices of Browser Agents 理解浏览器 agent 为什么是高风险隐私边界,而不只是自动点击工具 | | [BrowserGym — Web Agent 研究的统一浏览器环境](/study/papers/browsergym/) | `browsergym` | intermediate | UNVERIFIED | 用 BrowserGym 理解为什么 web agent 需要统一 observation / action / evaluation 接口,而不是每个 benchmark 各跑一套 | | [Chain-of-Thought — 让大模型先写步骤再回答](/study/papers/chain-of-thought/) | `chain-of-thought` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [ClawTrace — 把 agent 每步操作的"成本账"先算清再蒸馏](/study/papers/clawtrace-cost-aware/) | `clawtrace-cost-aware` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | @@ -43,6 +46,7 @@ sidebar: | [GraphRAG — 微软的知识图谱 + RAG](/study/papers/graphrag/) | `graphrag` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [GSM8K — 小学数学题把大模型算术短板照出来](/study/papers/gsm8k-2021/) | `gsm8k-2021` | intermediate | UNVERIFIED | 用 GSM8K 理解数学 word problem、verifier 和采样重排为什么重要 | | [HuggingGPT — 让 ChatGPT 当任务调度员,模型库当工具箱](/study/papers/hugginggpt-2023/) | `hugginggpt-2023` | intermediate | UNVERIFIED | 用 HuggingGPT 理解 LLM 如何规划并调用专用模型完成多模态任务 | +| [InjecAgent — 工具型 LLM Agent 的间接 Prompt Injection 基准](/study/papers/injecagent/) | `injecagent` | intermediate | UNVERIFIED | 用 InjecAgent 理解为什么外部邮件、网页和工具内容会把 agent 从用户目标劫持到攻击者目标 | | [Inner Monologue — 让机器人把观察结果说回计划里](/study/papers/inner-monologue-2022/) | `inner-monologue-2022` | intermediate | UNVERIFIED | 用 Inner Monologue 理解闭环反馈如何让语言计划接上真实环境变化 | | [InstructGPT — RLHF 让 LLM 听话](/study/papers/instructgpt/) | `instructgpt` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [Scaling Laws — 大模型训练不是玄学,是幂律预算题](/study/papers/kaplan-scaling-laws-2020/) | `kaplan-scaling-laws-2020` | advanced | UNVERIFIED | 用 Kaplan scaling laws 理解参数、数据和计算量怎样一起决定语言模型损失 | @@ -109,9 +113,5 @@ sidebar: | [SWE-bench — 真实 GitHub Issue 评测](/study/papers/swe-bench/) | `swe-bench` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [SWE-Bench-CL — coding agent 不能只刷静态题](/study/papers/swe-bench-cl/) | `swe-bench-cl` | intermediate | UNVERIFIED | 用 SWE-Bench-CL 理解软件工程 agent 的持续学习、迁移和灾难性遗忘 | | [SWE-Skills-Bench — Agent 技能真的帮得上软件工程吗](/study/papers/swe-skills-bench-2026/) | `swe-skills-bench-2026` | intermediate | UNVERIFIED | 用 paired evaluation 衡量 SWE skills 对真实软件工程 agent 的边际收益和 token 成本 | -| [Terminal-Bench — 在真实命令行任务里测试 agent](/study/papers/terminal-bench/) | `terminal-bench` | intermediate | UNVERIFIED | 用 Terminal-Bench 理解终端环境为什么能暴露 agent 的长程执行、环境理解和验证能力 | -| [ToolBench-X — 工具会坏时,agent 还能不能把事做完](/study/papers/toolbench-x/) | `toolbench-x` | intermediate | UNVERIFIED | 用 ToolBench-X 理解 tool-use benchmark 为什么要模拟规格漂移、调用错误、执行失败和结果冲突 | -| [Toolformer — 教 LLM 自主调用 API](/study/papers/toolformer/) | `toolformer` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | -| [ToolLLM — 用 16000+ API 训练模型进入真实工具世界](/study/papers/toolllm-2023/) | `toolllm-2023` | intermediate | UNVERIFIED | 用 ToolLLM 理解大规模 API 数据集、工具检索和工具评测如何支撑 agent | [下一组](/study/atlas/papers/topic-papers-agents-and-llm-systems-02/) diff --git a/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-02.md b/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-02.md index d1e92c78a..307161c1a 100644 --- a/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-02.md +++ b/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-02.md @@ -1,6 +1,6 @@ --- title: "智能体与 LLM 系统 · 论文 · 第 2 组" -description: "13 条 智能体与 LLM 系统 Atlas 分块" +description: "17 条 智能体与 LLM 系统 Atlas 分块" sidebar: hidden: true --- @@ -9,10 +9,14 @@ sidebar: [返回论文全景索引](/study/papers-atlas/) -本分块共 13 条,稳定上限为 100 条。 +本分块共 17 条,稳定上限为 100 条。 | 论文 | Slug | 难度 | 可信状态 | 简介 | |---|---|---|---|---| +| [Terminal-Bench — 在真实命令行任务里测试 agent](/study/papers/terminal-bench/) | `terminal-bench` | intermediate | UNVERIFIED | 用 Terminal-Bench 理解终端环境为什么能暴露 agent 的长程执行、环境理解和验证能力 | +| [ToolBench-X — 工具会坏时,agent 还能不能把事做完](/study/papers/toolbench-x/) | `toolbench-x` | intermediate | UNVERIFIED | 用 ToolBench-X 理解 tool-use benchmark 为什么要模拟规格漂移、调用错误、执行失败和结果冲突 | +| [Toolformer — 教 LLM 自主调用 API](/study/papers/toolformer/) | `toolformer` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | +| [ToolLLM — 用 16000+ API 训练模型进入真实工具世界](/study/papers/toolllm-2023/) | `toolllm-2023` | intermediate | UNVERIFIED | 用 ToolLLM 理解大规模 API 数据集、工具检索和工具评测如何支撑 agent | | [ToolSandbox — 状态化对话工具调用评测](/study/papers/toolsandbox/) | `toolsandbox` | intermediate | UNVERIFIED | 用 ToolSandbox 理解为什么工具调用 agent 需要测状态依赖、信息不足和多轮用户反馈,而不是只测单轮函数参数 | | [ToxiGen — 用生成模型造隐性仇恨测试集](/study/papers/toxigen-2022/) | `toxigen-2022` | intermediate | UNVERIFIED | 用 ToxiGen 理解安全评测为什么要覆盖隐性、对抗性和群体相关文本 | | [Tree of Thoughts — 让 LLM 像下棋一样多想几步再答](/study/papers/tree-of-thoughts-2023/) | `tree-of-thoughts-2023` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | diff --git a/src/content/docs/career-plan.md b/src/content/docs/career-plan.md index 8fd2709fa..75ea7d92e 100644 --- a/src/content/docs/career-plan.md +++ b/src/content/docs/career-plan.md @@ -5,7 +5,7 @@ sidebar: order: 1 --- -> 本页是路径说明。具体笔记见左侧分组;当前规模 2000+ 篇(论文 1075 + 项目 961)。 +> 本页是路径说明。具体笔记见左侧分组;当前规模 2000+ 篇(论文 1079 + 项目 961)。 ## 1. 路径模型的演化 diff --git a/src/content/docs/index.md b/src/content/docs/index.md index 9afb62e80..1bc865ad0 100644 --- a/src/content/docs/index.md +++ b/src/content/docs/index.md @@ -144,7 +144,7 @@ head: -
当前规模:1075 篇论文 + 961 个项目 = 2036 篇笔记,按 19 个主题组织。数量已移出首屏,只作为覆盖面证据。
+当前规模:1079 篇论文 + 961 个项目 = 2040 篇笔记,按 19 个主题组织。数量已移出首屏,只作为覆盖面证据。