Archive
Agent、MCP、coding agent、工作流自動化與評測。
代理工具
SPADE trains the environment designer using the return gap between runs with and without hints, allowing the curriculum to become harder automatically as the agent improves. Qwen3-30B-A3B gained as much as 13.9 points on tool-use benchmarks, though environment validators may still accept underspecified or unsolvable tasks.
代理工具
Using 26 tools and roughly 340 to 400 decision points, FM-Bench tests whether agents can handle delayed rewards, hidden information, and competitors’ reactions. All 15 models completed the 20-year campaign, but rankings stabilized only in the later stages, while self-managed memory exhibited two failure modes: endless accumulation or repeated overwriting.
最新模型
Researchers repeatedly update hidden states through the same set of Transformer layers before generating each token, comparing recurrent and conventional models using identical data and LoRA configurations. Gains are concentrated in parallel calls and cross-call dependencies, while API-Bank, which primarily involves single API calls, shows no consistent advantage.
AI 安全
A new study links continuous vectors exchanged between agents—but omitted from conversation logs—to their subsequent public actions as a single event, then applies three layers of checks: anomaly detection, counterfactual impact analysis, and sparse-feature inspection. The average AUROC of 0.993 combines text-based and latent-collusion samples, and the attacks were predesigned by the researchers, so the result cannot yet be considered a general capability for detecting emergent conspiracies.
代理工具
FACET binds instructions, containers, reference solutions, and verifiers to the same execution state, synthesizing 6,078 verified tasks from more than 71,000 agent skills. Fine-tuned on just 1,200 successful trajectories, Qwen3.5-9B improved its Terminal-Bench 2.1 score from 27.34 to 35.58.
代理工具
The study finds that although agents can independently train, debug, and evaluate models, they lock in their training strategy before the first parameter update. Adding experiment logs, a skill library, and an evaluation agent improves downstream scores but still does not prompt systems to change approaches when the evidence turns against them.
代理工具
Researchers applied control-flow rewrites, identifier renaming, and dead-code insertion to the same repair tasks, reducing solve rates by as much as 6.7 percentage points. Model robustness changed across mini-SWE agent, OpenCode, and datasets, meaning a single model leaderboard cannot predict deployment performance.
代理工具
Rather than letting full workflows obscure errors, ComponentBench individually tests UI components such as sliders, date pickers, data tables, and drag-and-drop controls. On the same tasks and framework, GPT-5 mini achieved success rates of 83.1% using an accessibility tree and 48.9% using pixel-only coordinate-based interaction.
代理工具
SkillGate finds that when long-horizon agents are trained with outcome rewards, the small number of tokens used to select a skill receive almost no effective credit. The new method separates two gradient channels within the same GRPO update and outperforms the outcome-reward baseline across five agent benchmarks.
代理工具
A new framework uses predictive entropy to determine whether an LLM judge should accept a verdict directly, search for web evidence, or abstain, while calibrating two thresholds with a Clopper–Pearson upper bound. All 32 open-domain question-answering experiments maintained the specified error rate, but the guarantee depends on i.i.d. calibration data and reliable labels.
代理工具
The Alibaba Cloud team integrates browser-agent error recovery, complex UI interaction, and long-horizon credit assignment into a unified training pipeline. Its 27B model achieves 65.1% Pass@1 on the newly created bilingual Chinese-English BrowserBench, though the public availability of the model weights and evaluation data remains to be confirmed.
推論系統
Qwen3.8-27B 讓 27B 級模型進入單張 24GB 消費級 GPU 的可用範圍。公開資料顯示，一般 GGUF 部署約落在 30 至 60 tok/s；在 RTX 4090 上啟用 KV cache 量化、MTP 與新版 llama.cpp 後，社群實測可提升至 70 至 85 tok/s，但這仍不是開箱即得的基準。