Archive
Agent、MCP、coding agent、工作流自動化與評測。
實體 AI／科學自動化
PUDA exposes automated instruments as discoverable command-line tools and uses JSON, NATS/JetStream, and stable identifiers to preserve commands, measurements, and sample provenance. Agents can choose the next experiment, but actual hardware operations can occur only through engineer-authored driver methods with validated parameters.
代理工具
CAPA embeds implementation preferences that users repeatedly omit into 600 executable coding tasks, testing whether assistants can infer new requirements from past conversations. Across 12 models, adding same-user history increased final success rates by only 6.8 points on average, but raised first-try success by 15.6 points—suggesting that memory mainly reduces clarification rounds rather than eliminating failures.
代理工具
Instead of using an LLM to judge travel plans, TREK verifies entities, budgets, opening hours, transportation, and implicit requirements individually. Even the best of 15 agents fully passed only 46.2% of solvable tasks, showing that locally correct answers remain difficult to assemble into executable itineraries.
AI 安全
SecRespond has security agents analyze read-only disk snapshots of compromised hosts instead of merely solving attack challenges in clean environments. Although 23 models could restate alerts, they commonly missed implants not disclosed by the alerts and failed to verify remediation.
代理工具
New research argues that relevance ranking can determine tool order but cannot answer how many tools an agent should load. CAM-DF directly learns the marginal utility of continuing to load tools, reducing tool exposure by 37% while maintaining a comparable task success rate.
AI 安全
MemTX no longer allows information written by agents to become immediately actionable facts. Instead, it adds evidence, authorization, provenance, snapshot isolation, and validated commits. The study reports zero downstream harm across five models, but the results still come from controlled evaluations and do not yet show that it can repair irreversible side effects in real-world external systems.
代理工具
Baidu’s research team has released 100 long-horizon document, spreadsheet, and presentation tasks, along with human labor time, price proxies, and executable validators. The best model achieved an average score of 17.91, compared with the human baseline of 27.79, showing that cheap and fast inference does not necessarily produce deliverables that are ready to use.
最新模型
BAAI reframes deep research as a recursive loop of “search—verify each constraint—conduct targeted follow-up searches,” while training the model to compress confirmed evidence and unresolved constraints on its own. The weights for a 4B dense model and a 122B-A10B MoE model have been released, but their advantages mainly come from automatically verifiable synthetic tasks and cannot yet be directly generalized to open-ended scientific research.
硬體
NVIDIA integrated Nemotron 3 Ultra into a “generate-test-reflect” RTL agent loop, achieving a 97.1% average pass rate across nine agentic task categories in CVDP. The results show that executable verification can significantly strengthen chip-design models, although data overlap, tool configurations, and vendor-run evaluations still limit generalizability.
最新模型
Laguna S 2.1 uses sliding-window attention, attention gating, and an FP8 KV cache to reduce long-context costs, targeting self-hosted agentic coding. The model comes with weights at multiple precision levels and complete evaluation trajectories, but its million-token capability, performance comparisons, and single-machine feasibility still depend heavily on the quantization format and runtime environment.
推論系統
Researchers enabled agents to predict and pre-execute their own next tool calls during inference, replacing an external draft model with shared weights and a shared KV cache. Joint reinforcement learning raised average Hit@1 to 61.2 and 66.3 for two Qwen 4B models, respectively, though the paper has yet to quantify end-to-end speedups in real-world parallel deployments.
AI 安全
The open-source SkillGate uses 530 rules to identify suspicious snippets before agents load skills, then sends them to an LLM for classification and isolation. Official benchmarks measured a 1.13% false-positive rate and 76.9% recall, showing that a low-cost gateway is feasible but cannot yet replace sandboxing and least privilege.