diff --git a/SESSION-HANDOFF.md b/SESSION-HANDOFF.md index 7d769319c..3b2d75c17 100644 --- a/SESSION-HANDOFF.md +++ b/SESSION-HANDOFF.md @@ -2,6 +2,31 @@ > 状态:当前接班入口。旧的批量生产 session 快照已失效,不得用于恢复自动循环;持续运行使用只读 supervisor + 有界 writer epoch。 +## 2026-07-14 总结不足后再推进 4 篇补强论文 + +- status:`local-ready`(已完成本地内容、receipt、atlas 和主要门禁;远端 PR / merge / Pages deploy 仍以实时外部状态为准)。 +- 起始 ref:`6dd71d8868a0142b88f2afefbdce353dba147678`(PR #32 merge 后的 `origin/main`)。 +- 分支:`study/papers-20260714-one-more-round`。 +- 本轮不足总结:上一轮 40 篇完成了规模和部署闭环,但多数卡片仍是 `STATIC_ANALYSIS` / `UNVERIFIED`;部分卡片 91 行、低于建议 100 行;L4 主要是 toy / manual simulation;主题上对 ML 工程 agent、终端 agent、长上下文有效窗口、视觉 Web GUI agent 的覆盖仍不够。 +- 本轮 objective:新增 4 篇更厚的 `study-v2` paper note,分别补强 `MLE-bench`、`Terminal-Bench`、`RULER`、`VisualWebArena`,保持 `UNVERIFIED` 边界,不声明运行真实 benchmark。 +- scope:允许新增 `src/content/docs/papers/*.md`、`data/review-receipts/papers/*.json`,刷新 `data/note-index.json`、papers atlas 派生页和公开计数文案;不修改候选队列、policy/threshold、既有论文正文语义。 +- activated_by:`explicit-user-request-2026-07-14-summarize-gaps-and-advance-one-more-round` +- review_after:`2026-07-14` +- acceptance_checks: + - `lr search arxiv` + arXiv API 元数据核验 4/4; + - `node scripts/quality-gate.mjs` 逐篇通过,行数分别为 135 / 137 / 147 / 138,无 advisory; + - `npm run audit:content-contract`:0 blocking,56 v2; + - `npm run atlas`:2028 notes,69 chunks; + - `npm run audit:counts`:projects=961、papers=1067、total=2028; + - `npm run audit:links` / `npm run audit:wikilinks`:无 blocking; + - `git ls-files -co --exclude-standard -z | node scripts/audit-public-redlines.mjs --stdin0`:0 blocking; + - `npm run build:strict -- --log /tmp/study-one-more-round-build-clean.log`:通过;首次失败由 stale `.astro` cache 触发 duplicate-id warning,删除 ignored cache 后恢复; + - `git diff --check`:通过。 +- budget:1 个内容小批次、4 篇新增 paper、1 个可写切片、1 个本地 writer。 +- external_outcome:当前为本地 review-ready branch;PR、merge 和 Pages deploy 需要以单独外部动作完成并复核。 +- stop_conditions:规范工具链不可用;arXiv 来源不可核验;content contract / redline / strict build 失败且无法在 scope 内修复;需要改 policy/threshold、候选队列或敏感内容;用户停止。 +- superseded_by:`none` + ## 2026-07-14 新增 40 篇论文全流程完成记录 - status:`complete` diff --git a/data/note-index.json b/data/note-index.json index e543081ee..9d757862b 100644 --- a/data/note-index.json +++ b/data/note-index.json @@ -3,16 +3,16 @@ "taxonomy_version": "taxonomy-v1", "stats": { "summary": { - "total": 2024, - "classified": 1977, + "total": 2028, + "classified": 1981, "unclassified": 47, "unknown_difficulty": 1975, "empty_description": 1970 }, "by_area": { "papers": { - "total": 1063, - "classified": 1044, + "total": 1067, + "classified": 1048, "unclassified": 19, "unknown_difficulty": 1014, "empty_description": 1013 @@ -20611,6 +20611,38 @@ "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/" } }, + { + "id": "papers::mle-bench", + "area": "papers", + "slug": "mle-bench", + "title": "MLE-bench — 用 Kaggle 任务衡量机器学习工程 agent", + "description": "用 MLE-bench 理解 ML 工程 agent 为什么不能只靠单元测试和代码 benchmark 来评估。", + "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 / MLE Benchmark" + }, + "trust": { + "contract_state": "v2", + "verification_status": "UNVERIFIED" + }, + "freshness": { + "state": "NOT_EVALUATED", + "reviewed_at": "2026-07-14", + "review_after": null + }, + "route": "/study/papers/mle-bench/", + "atlas": { + "chunk_id": "topic-papers-agents-and-llm-systems-01", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + } + }, { "id": "papers::mlflow", "area": "papers", @@ -26401,6 +26433,38 @@ "chunk_route": "/study/atlas/papers/topic-papers-network-protocols-01/" } }, + { + "id": "papers::ruler-long-context", + "area": "papers", + "slug": "ruler-long-context", + "title": "RULER — 真实长上下文能力不能只看 NIAH", + "description": "用 RULER 理解长上下文模型的有效窗口、检索幻觉和聚合推理为什么要分开评测。", + "difficulty": "intermediate", + "canonical_topics": [ + "papers-agents-and-llm-systems" + ], + "classification": { + "state": "classified", + "source": "frontmatter-category", + "topic_id": "papers-agents-and-llm-systems", + "matched_category": "llm", + "raw_category": "LLM / Long Context Evaluation" + }, + "trust": { + "contract_state": "v2", + "verification_status": "UNVERIFIED" + }, + "freshness": { + "state": "NOT_EVALUATED", + "reviewed_at": "2026-07-14", + "review_after": null + }, + "route": "/study/papers/ruler-long-context/", + "atlas": { + "chunk_id": "topic-papers-agents-and-llm-systems-01", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + } + }, { "id": "papers::rwkv-2023", "area": "papers", @@ -30633,6 +30697,38 @@ "chunk_route": "/study/atlas/papers/topic-papers-distributed-training-and-gpu-01/" } }, + { + "id": "papers::terminal-bench", + "area": "papers", + "slug": "terminal-bench", + "title": "Terminal-Bench — 在真实命令行任务里测试 agent", + "description": "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。", + "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 / Terminal Benchmark" + }, + "trust": { + "contract_state": "v2", + "verification_status": "UNVERIFIED" + }, + "freshness": { + "state": "NOT_EVALUATED", + "reviewed_at": "2026-07-14", + "review_after": null + }, + "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/" + } + }, { "id": "papers::tesla-architecture-2008", "area": "papers", @@ -32513,6 +32609,38 @@ "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-02/" } }, + { + "id": "papers::visualwebarena", + "area": "papers", + "slug": "visualwebarena", + "title": "VisualWebArena — 让网页 agent 真正看见界面", + "description": "用 VisualWebArena 理解多模态 web agent 为什么不能只读 DOM 文本,还要处理视觉线索。", + "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 Web" + }, + "trust": { + "contract_state": "v2", + "verification_status": "UNVERIFIED" + }, + "freshness": { + "state": "NOT_EVALUATED", + "reviewed_at": "2026-07-14", + "review_after": null + }, + "route": "/study/papers/visualwebarena/", + "atlas": { + "chunk_id": "topic-papers-agents-and-llm-systems-01", + "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/" + } + }, { "id": "papers::vit", "area": "papers", @@ -32761,8 +32889,8 @@ }, "route": "/study/papers/voyager/", "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/" } }, { @@ -33112,8 +33240,8 @@ }, "route": "/study/papers/webgpt-2021/", "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/" } }, { @@ -33176,8 +33304,8 @@ }, "route": "/study/papers/webxskill/", "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/" } }, { @@ -33400,8 +33528,8 @@ }, "route": "/study/papers/wizardlm-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/" } }, { @@ -64331,7 +64459,7 @@ "page": 2, "pages": 2, "route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/", - "entries": 1 + "entries": 5 }, { "id": "topic-papers-ai-safety-and-interpretability-01", diff --git a/data/review-receipts/papers/mle-bench.json b/data/review-receipts/papers/mle-bench.json new file mode 100644 index 000000000..7bc42f8e5 --- /dev/null +++ b/data/review-receipts/papers/mle-bench.json @@ -0,0 +1,55 @@ +{ + "schema_version": "study-review-receipt-v1", + "generation": 1, + "predecessor_digest_sha256": null, + "note": { + "area": "papers", + "slug": "mle-bench", + "digest_sha256": "b093cd115076731f58de53743a0cf5a47825c14c74d0f59731cf1f8f01b3415c" + }, + "source_revision": "arXiv:2410.07095v6", + "research_input_sha256": "8693c68493b19f35534fc4630738ec7c7e77bdc70ddbe247e870e49aae2d3ae6", + "reviewers": [ + { + "role": "ZERO_BASE", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 86, + "warnings": [ + "Explains Kaggle-style ML engineering with a toy workflow; no Kaggle task was executed." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "MANUAL_SIMULATION" + } + }, + { + "role": "ENGINEER", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 84, + "warnings": [ + "Engineering claims are limited to static paper reading; AIDE scaffold and benchmark harness were not run." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + }, + { + "role": "ACADEMIC", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 84, + "warnings": [ + "Citation identity was checked through arXiv metadata; reported medal rates are not independently reproduced." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + } + ], + "waivers": [], + "created_at": "2026-07-14T14:30:00.000Z" +} diff --git a/data/review-receipts/papers/ruler-long-context.json b/data/review-receipts/papers/ruler-long-context.json new file mode 100644 index 000000000..27243b588 --- /dev/null +++ b/data/review-receipts/papers/ruler-long-context.json @@ -0,0 +1,55 @@ +{ + "schema_version": "study-review-receipt-v1", + "generation": 1, + "predecessor_digest_sha256": null, + "note": { + "area": "papers", + "slug": "ruler-long-context", + "digest_sha256": "fdea3ef9179c032b4040312f8720d8e43b0a09628bbf8aef951386d00a1cea81" + }, + "source_revision": "arXiv:2404.06654v3", + "research_input_sha256": "585001909e88932e9aa31c90a0fb62d22fe20ca6e3e5588dcbcc0cc841768ac8", + "reviewers": [ + { + "role": "ZERO_BASE", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 87, + "warnings": [ + "Explains RULER through a toy long-context tracing example; benchmark tasks were not executed." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "MANUAL_SIMULATION" + } + }, + { + "role": "ENGINEER", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 84, + "warnings": [ + "Engineering claims are limited to static evaluation design reading; no model inference was run." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + }, + { + "role": "ACADEMIC", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 85, + "warnings": [ + "Citation identity was checked through arXiv metadata; reported long-context results are not independently reproduced." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + } + ], + "waivers": [], + "created_at": "2026-07-14T14:30:00.000Z" +} diff --git a/data/review-receipts/papers/terminal-bench.json b/data/review-receipts/papers/terminal-bench.json new file mode 100644 index 000000000..749741f9f --- /dev/null +++ b/data/review-receipts/papers/terminal-bench.json @@ -0,0 +1,55 @@ +{ + "schema_version": "study-review-receipt-v1", + "generation": 1, + "predecessor_digest_sha256": null, + "note": { + "area": "papers", + "slug": "terminal-bench", + "digest_sha256": "d53e73c9e364b799eda0b68fe0cbac327a4ede5637c473babced097d93f8744d" + }, + "source_revision": "arXiv:2601.11868v1", + "research_input_sha256": "ed2c466d8731db408817bc38005cf930de8a4bf7fef3ed30abb15c3db04c9dd7", + "reviewers": [ + { + "role": "ZERO_BASE", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 86, + "warnings": [ + "Explains terminal-agent evaluation with a toy CLI task; Terminal-Bench containers were not executed." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "MANUAL_SIMULATION" + } + }, + { + "role": "ENGINEER", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 85, + "warnings": [ + "Engineering interpretation is based on static paper reading; no benchmark harness or test container was run." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + }, + { + "role": "ACADEMIC", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 84, + "warnings": [ + "Citation identity was checked through arXiv metadata; reported success rates are not independently reproduced." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + } + ], + "waivers": [], + "created_at": "2026-07-14T14:30:00.000Z" +} diff --git a/data/review-receipts/papers/visualwebarena.json b/data/review-receipts/papers/visualwebarena.json new file mode 100644 index 000000000..dcd149f38 --- /dev/null +++ b/data/review-receipts/papers/visualwebarena.json @@ -0,0 +1,55 @@ +{ + "schema_version": "study-review-receipt-v1", + "generation": 1, + "predecessor_digest_sha256": null, + "note": { + "area": "papers", + "slug": "visualwebarena", + "digest_sha256": "2e22832cb00b9558903b1649eda876fa333a29bc2942b7069ebc359fb82b0383" + }, + "source_revision": "arXiv:2401.13649v2", + "research_input_sha256": "529256654169545c443e57e08b2a9f2ac11d67d1b7e0e963ec802d4ef5673a3c", + "reviewers": [ + { + "role": "ZERO_BASE", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 86, + "warnings": [ + "Explains visual web grounding with a toy shopping-page example; benchmark environment was not run." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "MANUAL_SIMULATION" + } + }, + { + "role": "ENGINEER", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 84, + "warnings": [ + "Engineering interpretation is based on static paper reading; no browser trajectory or multimodal agent was executed." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + }, + { + "role": "ACADEMIC", + "reviewer_version": "study-static-review-20260714-one-more-round", + "decision": "PASS_WITH_NOTES", + "score": 84, + "warnings": [ + "Citation identity was checked through arXiv metadata; reported benchmark numbers are not independently reproduced." + ], + "execution": { + "review_mode": "STATIC_REVIEW", + "code_mode": "NOT_APPLICABLE" + } + } + ], + "waivers": [], + "created_at": "2026-07-14T14:30:00.000Z" +} diff --git a/src/content/docs/about.md b/src/content/docs/about.md index d0c762085..f066f34a7 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: -- **1063 篇论文笔记** + **961 篇项目笔记**,合计 **2000+ 篇** +- **1067 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 5ae2471f7..981c0e9e5 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 @@ -56,6 +56,7 @@ sidebar: | [Minerva — 把语言模型拉进数学草稿纸](/study/papers/minerva-2022/) | `minerva-2022` | advanced | UNVERIFIED | 用 Minerva 理解为什么数学推理需要专门的数据、逐步解题和采样验证 | | [Misevolution — 自进化 agent 也会"越改越坏",连顶配模型也躲不过](/study/papers/misevolution-2509/) | `misevolution-2509` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [Mistral 7B — 小模型靠架构细节打出性价比](/study/papers/mistral-7b-2023/) | `mistral-7b-2023` | intermediate | UNVERIFIED | 用 Mistral 7B 理解 grouped-query attention 和 sliding-window attention 如何服务高效开源模型 | +| [MLE-bench — 用 Kaggle 任务衡量机器学习工程 agent](/study/papers/mle-bench/) | `mle-bench` | intermediate | UNVERIFIED | 用 MLE-bench 理解 ML 工程 agent 为什么不能只靠单元测试和代码 benchmark 来评估 | | [MMSkills — 把视觉 agent 的"操作经验"做成多模态卡片](/study/papers/mmskills-multimodal/) | `mmskills-multimodal` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [MRKL — 给大模型配一组专家工具和路由器](/study/papers/mrkl-systems-2022/) | `mrkl-systems-2022` | intermediate | UNVERIFIED | 用 MRKL Systems 理解 neuro-symbolic agent 为什么要把 LLM、检索和计算模块拆开 | | [Super-NaturalInstructions — 1600+ 任务教模型读懂说明书](/study/papers/natural-instructions-v2-2022/) | `natural-instructions-v2-2022` | intermediate | UNVERIFIED | 用 Super-NaturalInstructions 理解 declarative instructions 如何评测任务泛化 | @@ -82,6 +83,7 @@ sidebar: | [ReAct Agent — 推理和行动交替的工具使用范式](/study/papers/react-agent/) | `react-agent` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [Reflexion — 让 LLM 自我反思](/study/papers/reflexion/) | `reflexion` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [RETRO — DeepMind 的检索增强 LLM](/study/papers/retro/) | `retro` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | +| [RULER — 真实长上下文能力不能只看 NIAH](/study/papers/ruler-long-context/) | `ruler-long-context` | intermediate | UNVERIFIED | 用 RULER 理解长上下文模型的有效窗口、检索幻觉和聚合推理为什么要分开评测 | | [SayCan — 机器人不只问“想做什么”,还问“我能做什么”](/study/papers/saycan-2022/) | `saycan-2022` | intermediate | UNVERIFIED | 用 SayCan 理解语言模型和机器人 affordance 如何合成可执行动作 | | [Self-Consistency — 让模型把同一道题做 40 遍再投票](/study/papers/self-consistency-2022/) | `self-consistency-2022` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | | [自进化 AI agent 综述 — 给"会自己升级"的 agent 画一张统一地图](/study/papers/self-evolving-agents-survey/) | `self-evolving-agents-survey` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | @@ -101,6 +103,7 @@ 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 | @@ -109,9 +112,6 @@ sidebar: | [TruthfulQA — 专门问模型容易学人类谬误的问题](/study/papers/truthfulqa-2021/) | `truthfulqa-2021` | intermediate | UNVERIFIED | 用 TruthfulQA 理解语言模型为什么会模仿常见假话而不是坚持事实 | | [UL2 — 一个模型同时练完补空、续写和长文本](/study/papers/ul2-2022/) | `ul2-2022` | advanced | UNVERIFIED | 用 UL2 理解 mixture-of-denoisers 如何统一不同语言模型训练范式 | | [VeriCache: Turning Lossy KV Cache into Lossless LLM Inference — 有损压缩草稿,无损输出验收](/study/papers/vericache/) | `vericache` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | -| [Voyager — LLM 终身学习智能体](/study/papers/voyager/) | `voyager` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | -| [WebGPT — 让模型带着浏览器回答问题](/study/papers/webgpt-2021/) | `webgpt-2021` | intermediate | UNVERIFIED | 用 WebGPT 理解检索、引用和人类偏好如何组合成可追溯问答 | -| [WebXSkill — 给 Web agent 的可执行 skill 是参数化代码 + URL 图索引](/study/papers/webxskill/) | `webxskill` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | -| [WizardLM — 用 Evol-Instruct 自动变难训练题](/study/papers/wizardlm-2023/) | `wizardlm-2023` | intermediate | UNVERIFIED | 用 WizardLM 理解 instruction 数据不只要多,还要逐步变复杂 | +| [VisualWebArena — 让网页 agent 真正看见界面](/study/papers/visualwebarena/) | `visualwebarena` | intermediate | UNVERIFIED | 用 VisualWebArena 理解多模态 web agent 为什么不能只读 DOM 文本,还要处理视觉线索 | [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 1b9b9b8af..1cd98d219 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: "1 条 智能体与 LLM 系统 Atlas 分块" +description: "5 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 1 条,稳定上限为 100 条。 +本分块共 5 条,稳定上限为 100 条。 | 论文 | Slug | 难度 | 可信状态 | 简介 | |---|---|---|---|---| +| [Voyager — LLM 终身学习智能体](/study/papers/voyager/) | `voyager` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | +| [WebGPT — 让模型带着浏览器回答问题](/study/papers/webgpt-2021/) | `webgpt-2021` | intermediate | UNVERIFIED | 用 WebGPT 理解检索、引用和人类偏好如何组合成可追溯问答 | +| [WebXSkill — 给 Web agent 的可执行 skill 是参数化代码 + URL 图索引](/study/papers/webxskill/) | `webxskill` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | +| [WizardLM — 用 Evol-Instruct 自动变难训练题](/study/papers/wizardlm-2023/) | `wizardlm-2023` | intermediate | UNVERIFIED | 用 WizardLM 理解 instruction 数据不只要多,还要逐步变复杂 | | [Zombie Agents — 自进化 agent 的长期记忆能被持久化"借尸还魂"](/study/papers/zombie-agents-2602/) | `zombie-agents-2602` | unknown | UNVERIFIED | 暂无独立描述;可先从标题与正文定位开始。 | [上一组](/study/atlas/papers/topic-papers-agents-and-llm-systems-01/) diff --git a/src/content/docs/career-plan.md b/src/content/docs/career-plan.md index ca5292e40..397bcd31a 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 1063 + 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 1067 + 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 444efb49e..88fd40188 100644 --- a/src/content/docs/index.md +++ b/src/content/docs/index.md @@ -144,7 +144,7 @@ head: -
当前规模:1063 篇论文 + 961 个项目 = 2024 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:1067 篇论文 + 961 个项目 = 2028 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。