Archive
1539 technical news articles
代理工具
New research finds that agents that accumulate skills indiscriminately begin to regress after the third round, and deleting the original erroneous skills does not remove their derivatives. Verifier-as-Gatekeeper controls skill promotion through format, behavioral, and semantic checks, but the results come from a 50-task subset and no implementation has been released.
最新模型
MACRO does not update model weights. Instead, it searches for a fixed execution path for each task that can skip layers, jump backward, and reuse hidden states. Across 13 benchmarks, the authors report reducing search time from 14.8 hours to 1.6 hours, though every new task still requires labeled training and validation data.
最新模型
Google DeepMind has released the JAX implementation, pretrained weights, and a streamlined Colab-compatible version of WeatherNext 2, covering global weather and tropical cyclone forecasts up to 15 days. Researchers can now reconstruct the autoregressive forecasting workflow, but the full model still depends on ECMWF initialization data and high-end accelerators, and its outputs cannot replace official warnings.
推論系統
A new study uses typed error contracts to unify monitoring across KV caches, latent caches, sparse selectors, and recurrent states, while using Lean to verify formalizable composition rules. The public repository can reproduce the paper’s figures and 67 theorems, but excludes architecture-specific probes and the complete serving platform.
代理工具
CodeGrep first narrows the scope of changes through parallel grep, glob, and file-reading operations, then passes candidate files to a frozen OpenHands coding agent. It maintains a roughly 27% repair rate on SWE-bench Verified while reducing inference costs, but the model and training pipeline have not yet been released.
最新模型
Instead of treating the probability of a preference token as the reward, RRC constructs scores from the relative rankings of candidate responses. With an 8B reward model and eight-vote aggregation, the ArenaHardV2 score also rises from 8.0% to 11.2%, at the cost of additional reward model inference.
電腦視覺與具身 AI
HOPE maps tactile-glove data, planar sensor readings, and contact-only annotations onto a unified MANO hand mesh. It can output per-vertex pressure and contact from ordinary RGB video, although its public page currently still lists the code and models as “coming soon.”
新基準在保持目前請求、工具集合與正確動作不變的情況下，只插入已失效但看似合理的歷史紀錄。Oracle-OPD 將 Qwen3-1.7B 的平衡工具使用準確率提高至 87.0%，但驗證範圍仍限於合成的零售與航空情境。
RAG 與文件工程
READ replaces fixed chunking and vector Top-K retrieval with lexical search, structure navigation, and section reading, making every evidence-retrieval step replayable. It substantially outperformed dense retrieval on a 780-page financial report, but the sample was small, costs were higher, and the evidence is insufficient to claim that agentic search is superior to traditional lexical indexes.
電腦視覺
CogVis decouples cross-temporal image comparison from text-category classification, allowing multiple open-vocabulary queries to reuse the same scene-change features. The authors report state-of-the-art results across seven benchmarks and a 28.5% throughput gain, but did not release the model weights with the code.
代理工具
AgentOPSD accumulates probability differences between a teacher and student into turn-by-turn belief updates to identify the decisions that truly determine task success or failure. The authors report that it outperforms GRPO and several self-distillation baselines across three types of agent environments, although the public repository still does not provide training code.
最新模型
Rather than directly generating the next screenshot, AppDeltaWorld first retrieves layouts reachable through the given action, then generates HTML diffs and fills them with synthetic images. It can create closed-loop training trajectories, but its current results still depend on a custom evaluation framework and a full implementation that has not been released.