2026年9月16日 技术热点总结

项目简介：GPT-6 Astra for embodied AI and robotics.

核心介绍：GPT-6 Astra with embodied AI / robotics workflows and demos. Cases are grouped by workflow and ordered by publication date, newest first within each section. Astra directly performs zero-shot control in a simulator. Astra performs high-level planning for the cola-bottle pick-up, and the GEAR-SONIC planner converts the plan into a whole-body qpos trajectory for the simulated G1 to execute.

项目数据：⭐ 681 Stars，🍴 13 Forks

项目简介：信息安全提醒：请勿上传企业禁止对外公开的文件；可在本地或企业内网部署大模型后使用。中文公文排版 Skill：按 GB/T 9704-2012 生成可编辑 DOCX。配套 CEB：https://github.com/mizzlelover/CEB

技术栈：JavaScript、ai-skill、ceb、chinese-official-document、claude-code、cross-platform、document-formatting、docx、gbt-9704-2012

核心介绍：> 信息安全提醒（请先阅读） > 配套工具： CEB 文件转换 负责把已验证的 Founder CEB 文件转换为 PDF、Markdown、TXT；公文排版 负责把中文正式材料整理成符合 GB/T 9704-2012 的可编辑 DOCX。两个 Skill 可以连成一条“读取 → 整理 → 交付”的工作流。 把中文正式材料整理成可直接交付的公文 DOCX：按 GB/T 9704-2012 处理版心、字体、标题层级、文号、页码、附件与版记；普通稿不误用红头，正式发文支持预印红头纸套打或完整电子红头，并保留 Word/WPS 可更新目录。 许多 AI 只会把标题设为黑体、正文设为仿宋，便把输出称为“公文”。本项目把 A4、版心、正文、层次…

项目简介：the development environment for pros

技术栈：Go、agent-orchestration、ai、cli、coding-agents、developer-tool、devtools、go、golang

核心介绍：rune.build · quick start · hack on rune

项目数据：⭐ 623 Stars，🍴 35 Forks

项目简介：If you think AI sucks, star the repo.

核心介绍：If you think AI sucks ass (butt), star the repo.

项目数据：⭐ 604 Stars，🍴 0 Forks

项目简介：Personal cloud OS for a developer and their AI agents on a single server. One-command install on Ubuntu, then everything in the browser: a windowed desktop, isolated workspaces with VS Code, Claude Code and Codex, a store of 944 self-hosted apps, backups and updates. Access only via Tailscale, nothing exposed to the internet. Rust core, Svelte UI.

技术栈：Rust

核心介绍：used from any device through the browser. It is built as a cross-platform vibe coding flow: the same desktop, workspaces and agents open on a laptop, a tablet or a phone, the workspace and its agents stay on the se…

We demonstrate that training LLMs with RL does not improve performance equally across a dataset. RL shows large improvements on easy problems that an LLM is already good at solving, but small improvements on hard problems. We call this the Matthew Effect in RL for LLMs, after the phenomenon of cumulative advantage from economics and network science summarized as "the rich get richer". The naive explanation is that hard problems require more compute to find a solution. We argue that modern RL methods are exacerbating the issue by wasting too much compute on easy problems and instead should d…

The increasing deployment of AI agents in long-horizon tasks yields massive execution logs. Diagnosing failures within these records is crucial for reliability, as it transforms outcome-level signals into actionable interventions. The sheer scale of the data renders human review impractical, driving the need for automated root-cause attribution (RCA). However, automated RCA methods using LLMs suffer from low diagnostic accuracy, especially as execution traces grow larger. They struggle because relevant information is often sparse, distributed across distant actions, and disconnected from th…

Can financial vision-language models (VLMs) turn chart evidence into reliable action recommendations? Existing hallucination evaluations are mostly claim-centric; they assess whether generated statements are supported, but not whether evidence remains traceable through rationale, confidence, and final action. We introduce E2A-Bench, a 969-query benchmark for financial chart reasoning, constructed from 323 HS300 constituents under three input modalities with deterministic OHLCV-derived evidence anchors. E2A-Bench evaluates grounding, reasoning-action consistency, evidence-confidence calibrat…

A central tenet of human cognition is systematicity, the principle that understanding one concept is inherently tied to understanding close variations of that concept. Do reasoning models robustly exhibit such systematicity? If so, we would expect consistent performance on structurally equivalent variants of the same task. Here, we extend established rule induction tasks from cognitive science to assess the systematicity of thought in current reasoning models. Each task family has compositional structure that we use to create structurally equivalent task variations through task isomorphisms…

Decision trees are among the most widely used models in machine learning, largely due to their transparent decision logic, making them well-suited for high-stakes decision-making contexts. However, most existing learning algorithms focus on predictive performance, overlooking the joint optimization of other desirable properties, such as structural sparsity. In this work we propose TREVIS, an approach for learning decision trees with respect to complex objectives, based on the exploration of the latent space of a Tree Transformer Variational Auto-Encoder (TTVAE). By mapping decision trees on…

Orthrus is a hybrid autoregressive-diffusion architecture that accelerates autoregressive language-model inference by generating multiple tokens in parallel while using a frozen autoregressive backbone. Its central claim is that an intra-model consensus mechanism enables lossless speculative decoding, producing the same output sequence as the autoregressive model.
We independently reproduce Orthrus and examine this claim under different numerical precisions. Under BF16 inference, exact trajectory matching occurs in only 45% of cases for the authors' checkpoint and 43% for our independentl…