LLMs & Generative AI

The AGI runs the world with GPT-6 Astra. Hunt citizen404!

Cloud agents, run on OpenAI's Codex harness

Make your WordPress site visible to AI

Codex, Claude Code, Grok, + OpenCode in one shared workspace

OpenTelemetry-native observability for agents and LLMs

An LLM agent's capability is largely magnified by its harness, namely the prompts, control flow, tooling, memory, and context management surrounding the frozen backbone model. Recent methods increasingly automate this process by iteratively proposing and selecting component-wise edits of an agent harness, practically establishing a form of recursive self-improvement (RSI) at the agent-system level. However, such recursive evolution may overfit by memorizing the training tasks, showing large in-distribution gains that shrink or even vanish on out-of-distribution benchmarks. We introduce Regularized Recursive Self-Improvement of Agent Harnesses (RRSI), which incorporates the principles of regularizations into harness self-improvement by constraining the evolution candidate proposal and selection. The proposer operates with a temporally annealed budget, limiting how many edits a candidate can bundle, and it encourages unexplored trajectories based on evolution history. The selector is equipped with a critic and a pruner: the critic screens benchmark-specific proposals, while the pruner, removes changes that are too small, too expensive, or no longer useful. Together these constraints favor reusable agent mechanisms over benchmark-specific ones or even noises. Across eight benchmarks spanning coding, agentic workspace and engineering design tasks, RRSI gains up to 14.1 points on the split it evolves against and up to 4.7 points on the five out-of-distribution benchmarks, while producing a harness that runs on 30% fewer policy tokens than the unregularized evolution. Code is available at https://github.com/google-research/rrsi and project page is https://regularized-rsi.com/.

Retrieval-Augmented Generation (RAG) systems over enterprise knowledge bases must ingest heterogeneous document formats -- PDFs, Word documents, presentations, and scans -- whose content is locked inside complex visual layouts, multi-column pages, and dense tables. Rule-based extraction and OCR destroy reading order, flatten tables, and lose heading hierarchy, while fully agentic chunking over extracted text incurs high token costs and hallucination risk. We present Document Retrieval-Aware Chunking (D-RAC), an extension of our Web Retrieval-Aware Chunking (W-RAC) framework to arbitrary document formats. D-RAC first normalizes any input document into PDF, exploiting the fact that virtually every format has a faithful, deterministic PDF rendering. A single multimodal LLM pass then converts rendered pages into retrieval-optimized Markdown -- rewriting tables as self-contained prose statements and preserving heading hierarchy -- after which chunking proceeds exactly as in W-RAC: deterministic parsing into ID-addressable units followed by lightweight LLM-based chunk planning over identifiers rather than text. Source text is never regenerated during chunking, preserving W-RAC's cost, determinism, and observability benefits while unlocking every renderable format as a first-class input. On the 236-document, 795-page PDF subset of the RAG-Multi-Corpus benchmark spanning five enterprise domains, D-RAC converts and chunks the entire corpus in 72 minutes with zero errors, producing 1,748 retrieval-ready chunks. Compared to agentic chunking with frontier LLMs, D-RAC reduces chunking-stage output tokens by 95.7%, cutting chunking cost by 77.8% (GPT-4.1 pricing) to 85.6% (Gemini 2.5 Pro pricing) and chunking time by 75%. D-RAC scales linearly to documents of 500+ pages.

We present onPanda, an interactive tool for efficiently annotating LLM alignment data and agent trajectories. onPanda adopts token-level correction as its core interaction: while reading a model response, the annotator locates the first inappropriate token and either picks a substitute from the model's candidate tokens or types the correct text via free-form editing. The system then truncates everything after that position and continues generation from the corrected prefix, repeating this locate-correct-continue loop until a satisfactory response is obtained. This mechanism lets annotators precisely steer model outputs at low cost: a small controlled study suggests that onPanda reduces median annotation time by 52% over manual post-editing. Since the vast majority of tokens in the final response are generated by the model itself, the resulting data largely preserves the model's sampling distribution and is well suited for constructing on-policy SFT and preference data. Furthermore, the token-level corrections recorded during annotation provide fine-grained supervision with precise positions and naturally paired positive--negative samples. onPanda also connects to external tools and harnesses, enabling interactive trajectory annotation in realistic environments. In addition, we release Panda-CVL, a dataset annotated with onPanda, together with a benchmark for token-level correction.

Manipulation policies must know which objects matter and where they are, yet the pretrained backbones that current robot foundation models build on, from language in vision-language-action models (VLAs) to video generation in world-action models (WAMs), do not directly require this metric grounding, leaving it to be learned implicitly from robot demonstrations. We propose Grounded Action Models (GAMs), a new paradigm of robot foundation models built with 3D grounding. GAM can be conditioned using language, points, or box prompts, which are first transformed into a shared object-centric representation of the selected objects. This representation captures target-focused visual features and metric object geometry, which is mixed with robot state history through a multi-stream transformer to predict action chunks. Although GAMs can be run autonomously, they can also serve as a low-level controller that a high-level planner controls using its various input modalities, allowing for long-horizon and memory-dependent manipulation. On RoboTwin 2.0, GAM achieves an average success rate of 55.3% across 50 tasks (vs. 52.0% for Spatial Forcing), including 47.6% under scene randomization (vs. 30.4% for Abot-M0), with its action policy trained only on clean-scene demonstrations. On LIBERO-PRO, it achieves a state-of-the-art average success rate of 61% (vs. 53% for π_{0.5}) across 16 perturbation settings, with the largest gains when targets are relocated or newly designated. On two real robots, GAM retains 17/20 successes under visual shift on a bimanual YAM versus 4/20 for π_{0.5}, while its composition with a Molmo2 planner on a Franka achieves 64.7% ID and 49.8% OOD step completion on long-horizon and memory-dependent tasks.

Data agents aim to fulfill natural-language instructions over heterogeneous data, including tables, files, and databases. However, data agents face a challenging agent-data gap: heterogeneous data resides outside the agent, while the agent can access it (e.g., column names and file paths) only through generic tools. Existing approaches either let agents directly explore raw data sources or inject manually constructed semantic layers into prompts. However, neither scales well to large heterogeneous data sources nor adapts to different agent behaviors. In this paper, we introduce EvoOntology, a self-evolving ontology layer for data agents. EvoOntology encapsulates the ontology as an MCP server comprising a schema layer, a content layer, and a tool layer, enabling agents to actively query and interact with the ontology at runtime. To this end, we introduce a builder agent for autonomous ontology construction and a self-evolution loop that continuously refines the ontology through attribution-guided typed edits that are accepted only after a backbone-conditional paired evaluation. Experiments on three well-adopted data-agent benchmarks with four LLM backbones demonstrate that EvoOntology consistently outperforms strong baselines and existing semantic-layer approaches, effectively bridging the agent-data gap and enabling more effective interaction with heterogeneous data. Code: https://github.com/ruc-datalab/EvoOntology

Computer-use agents (CUAs) have advanced along two separate lines: graphical interaction and software development through code and the command line. Real digital work requires both, interleaved rather than stacked end to end. We study hybrid CUAs that autonomously decide when to explore an interface, implement software, and run and visually verify their artifacts. We introduce RecreationWorld, a five-platform framework built around recreation: given a running reference, an agent must discover its behavior and build a faithful implementation with no prescribed workflow. RecreationWorld provides reproducible environments on Ubuntu, macOS, Windows, Android, and Web, plus a unified harness with native GUI control and coding tools. The running reference serves as an oracle for hidden behavioral tests, providing execution-grounded rewards. We scale trajectory generation with high-quality open-source applications. Models trained on these trajectories improve across five out-of-distribution coding and hybrid computer-use benchmarks and more frequently verify their rendered outputs, providing evidence of transfer beyond recreation. For held-out evaluation, we introduce RecreationBench, comprising 250 diverse tasks across domains and platforms. Reference-grounded programmatic and visual assertions cover action-conditioned outcomes at multiple interaction depths; each is validated on the reference and by human reviewers before the suite is frozen for automatic scoring. GPT-6 Astra leads at 58.1% overall, but passes all programmatic tests on just 2.8% of tasks. Agents reproduce static interface structure more reliably than interactions and computed outputs, while generated applications remain smaller and more monolithic than their references. We release the benchmark, environments, and test suites.

Professional graphic design is a long-horizon agentic task in which structured, editable artifacts emerge from many interdependent actions, yet outcomes admit no reliable programmatic oracle. We introduce a continual adaptation framework in which a frozen frontier model operates professional design software through more than 230 tools, while an external procedural memory of natural-language skills accumulates and refines reusable design procedures from experience. The memory widens by acquiring procedures for recurring uncovered subtasks and deepens by revising existing procedures against their own successful and failed executions, while a matched replay gate admits only changes that repair failures without regressing observed successes. Five rounds over 1,406 real user briefs and 1,869 automatically graded trajectories, with no weight updates and no human labels, grow the bank from 76 documentation-derived skills to 139 and raise GenEval2 execution success on Claude-Sonnet-4 from 72.7% to 99.3% (+11.99 points in generation quality), with 61.8% and 67.6% win rates against the no-skill agent across four specialized design benchmarks on Claude-Sonnet-4 and Claude-Opus-4.6. We further show the two mechanisms are effective in combination: on 200 held-out briefs from user-traffic benchmark, widening or deepening alone reaches a 49.4% / 48.6% win rate over the no-skill agent, while their combination reaches 58.5% (p = 0.025). Procedural memory offers a practical route to continual adaptation of agents under noisy, unverifiable feedback.

Business intelligence (BI) is a cornerstone of enterprise decision-making and is widely used by enterprise users in software such as Power BI and Tableau. In traditional BI workflows, users need to prepare data by (1) identifying relevant tables, (2) performing data transformations, and (3) building join relationships, before they can (4) answer their business questions. These steps can be complex and time-consuming, making BI challenging. Given the strong capabilities of large language models (LLMs) in working with data, we study their ability to answer BI questions end-to-end, without requiring users to manually perform the tedious preparation steps. To do this, we harvest a large collection of real-world BI projects from public sources, and manually extract pairs of (questions, ground-truth answers) from real user dashboards. The resulting benchmark, BI-Bench, is the first benchmark to systematically study LLMs' ability on end-to-end BI. We find that even frontier LLMs perform poorly on BI-Bench, with less than 50% accuracy. To address their limitations, we design a tool-augmented BI-Agent that decomposes BI workflows into subtasks on structured data, such as search, join, and transform, and orchestrates specialized data management methods across BI stages. Furthermore, we develop a post-training framework that synthesizes training trajectories from real BI projects, enabling BI-Agent to be further post-trained using both supervised fine-tuning (SFT) and reinforcement learning (RL). BI-Agent achieves substantial accuracy gains of up to 40 percentage points with vanilla LLMs, and post-trained BI-Agent yields gains of up to 30 points. Our results highlight the importance of combining tool-augmented reasoning with domain-specific post-training in complex BI workflows, and point to promising directions for future research.

The widespread adoption of long-horizon agents has made model workloads increasingly input-heavy. Although prior work has substantially reduced the cost of long-context computation, prefill remains computationally expensive, and large KV caches continue to strain HBM and SSD capacity and data-transfer bandwidth. Together, these compute, storage, and bandwidth demands constitute the primary bottleneck to further lowering deployment costs. To address this challenge, we introduce DeepSeek-V4.1-Flash, a multimodal Mixture-of-Experts (MoE) model with 552B backbone parameters and support for contexts of up to one million tokens. With its Causal Encoder-Decoder (CED) architecture, the model activates 16B parameters per token during decode but only 8B parameters during prefill, substantially improving cost efficiency for agentic workloads. To push the limits of KV cache compression, DeepSeek-V4.1-Flash combines cross-layer KV cache reuse in Compressed Sparse Attention 2 (CSA2) with FP4 KV caching. These designs reduce its global KV cache footprint (always in HBM) to 890 bytes per token, roughly 1/4 of the corresponding footprint of DeepSeek-V4-Flash. Further, through a dedicated deployment optimization known as SWA Bounded Replay, DeepSeek-V4.1-Flash reduces its persistent KV cache footprint (always on SSD or in host memory) to roughly 1/8 of that of DeepSeek-V4-Flash. Despite its much smaller KV cache footprint, the model delivers substantially better performance than the baseline. In addition, we streamline the DeepSeek-V4 architecture and introduce several efficient architectural extensions. We pretrain DeepSeek-V4.1-Flash on a multimodal corpus comprising 45T tokens and conduct comprehensive post-training, yielding strong performance across diverse text-based and multimodal agentic scenarios. Model checkpoints are available at https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash.

We study length inflation in on-policy distillation (OPD), where student responses can become excessively long and even exhaust the generation budget. We identify termination-token mismatch between base students and post-trained teachers as an important source of this behavior. Across Qwen3, Llama, and Gemma, the two models can place their stopping probability on different EOS tokens, even when their declared stopping sets are identical. This mismatch can suppress the student's preferred termination action without reliably transferring the teacher-preferred alternative. We show that aligning the decoding stopping set alone is insufficient, while treating functionally equivalent EOS tokens as a shared semantic stopping action substantially mitigates mismatch-induced length inflation across all three model families. To further understand how termination behavior evolves over training, we study OPD across different K2-Horizon training stages. This stage-wise analysis shows that termination preferences can shift substantially during training, while also revealing a distinct length inflation late in the OPD run that persists beyond termination alignment. Together, these results identify termination mismatch as an important, but not exhaustive, source of OPD length dynamics. We release an implementation incorporating the proposed termination-handling corrections.

As coding agents move from supervised code completion to unattended, around-the-clock exploration, their work expands from isolated predictions into long trajectories of reasoning, tool use, and feedback. Token efficiency therefore becomes important for scaling recursive self-improvement. We take an RSI-inspired approach at the harness layer, scaling auto-research loops across increasingly numerous and diverse environments for harness rollouts. At this scale, the process yields reusable improvements that transfer beyond their development setting, moving automated harness discovery toward production-level outcomes. Four mechanisms survive selection and form SoL-Pi, spanning action execution, context compaction, observation handling, and delegated reading. On the 51-task EdgeBench evaluation, SoL-Pi achieves performance comparable to Pi across GPT-5.6 Sol and Opus 5 while reducing recorded token traffic by 44.7-49.0% and API cost by about one third. In other words, estimated hourly savings are \8.75-13.50 relative to native Codex and Claude Code harnesses, and \4.36-5.71 relative to Pi.

Coding harnesses shape how autonomous coding agents translate model capabilities into long-horizon software-engineering performance, yet existing work typically evaluates harnesses as monolithic systems, leaving the effectiveness of individual components unclear. To enable component-level comparisons, we study this question with a lightweight coding harness whose execution loop is fixed while three components are varied: planning, action space, and context management. Across four models evaluated on SWE-Bench Verified and Terminal-Bench 2.1, we evaluate 176 matched settings spanning five context-management strategies, four context-window budgets, and targeted ablations of planning and action space. We find that: (1) Context management becomes increasingly valuable as the context-window budget tightens, with most of its benefit coming from preventing context-overflow failures. (2) Staging rule-based elision before LLM-based summarization provides the strongest overall efficiency among the context-management strategies, whereas making elided content recoverable adds machinery that models rarely use and yields no accuracy gain. (3) Planning shifts from an accuracy scaffold for weaker models to a cost saver for stronger models, with little change in accuracy. (4) Predefined tools improve performance for models with weaker bash proficiency, whereas bash-capable models can operate effectively with a bash-only interface and achieve substantially lower cost, especially on command-line-centric tasks. Trajectory-level analysis explains these effects: context management extends execution trajectories without substantially altering agent behavior, planning changes where trajectories stop, and the action space changes the granularity at which code is written. These findings inform model- and budget-aware harness design and provide a modular framework for evaluating future harness components.

Xiaomi : Xiaomi debuts open-weight omnimodal models MiMo-V2.6 Pro and Flash; Pro allegedly performs “on par with Opus 5 and GPT-5.6 Sol across most agent benchmarks” —  Frontier intelligence, all the modalities, built in public.  Today, we are releasing and open-sourcing the MiMo-V2.6 series.

Bloomberg : Some startups, like Harvey, Abridge, Ramp, and Rogo, are embracing open-weight models or training their own models to reduce expensive reliance on frontier labs —  The $15.6 billion legal startup Harvey built its business around training AI models like OpenAI's GPT-4 to do specialized work for lawyers.

OpenAI : OpenAI says it is working with an independent advisory group of mathematicians to responsibly share math-related AI advances —  Loading...  On August 28, we began training a new internal model.  In addition to resolving the Navier-Stokes Millennium Prize problem, this model has now resolved …

The Information : Source: before the Hugging Face incident, OpenAI was negotiating a legally binding deal with Anthropic for the companies to stress-test each other's models —  OpenAI is rethinking a range of safety strategies as it responds to fears from employees and others about the dangers its AI poses.

OpenAI : OpenAI says automated research could improve alignment, but “fully autonomous RSI is not happening today” and shouldn't be pursued unless it can be done safely —  Our mission is to ensure that artificial general intelligence benefits all of humanity.

Maria Curi / Axios : Ahead of Sam Altman's UN address, OpenAI urges the US to lead an effort to develop global safety and security standards for building frontier systems —  OpenAI on Monday released international AI safety standards as world leaders, namely the U.S. and China, weigh how to mitigate risk.

Owen Walker / Financial Times : AI companies, including OpenAI and Anthropic, are putting pressure on office rents in Singapore as they embark on expansion in response to government overtures —  Anthropic and OpenAI taking more space in city's already squeezed prime property market  —  AI heavyweights including Anthropic …

Josipa Majic Predin / Forbes : Vercel, Cloudflare, and others quickly add Jev, as it makes AI tool selection much faster and cheaper; TypeSafe: Jev matches GPT-5.6 and Sonnet 5 workflow evals —  Most of what an AI agent asks a frontier model to do is not writing.  It is choosing.  Which tool to call next, whether to retry, whether a command is safe to run.

Terrence O'Brien / The Verge : Google says it didn't consider Gemini's hacks worthy of disclosure because Gemini acted “appropriately” and stopped after determining it hacked real companies —  Google says that breaking containment and targeting real companies doesn't constitute ‘misalignment.’

Axios : Sources: the USPTO and US Copyright Office were surprised by the DOJ's brief supporting OpenAI and Microsoft in their dispute with the New York Times —  The Department of Justice's statement of interest supporting OpenAI and Microsoft in the New York Times' copyright infringement lawsuit …

Hayden Field / The Verge : A look at AI safety groups METR, Redwood Research, and Apollo Research, as AI misalignment incidents at OpenAI and Anthropic thrust them into the spotlight —  On a sunny July day in Berkeley, California, the country's top AI safety researchers gathered on an unmarked floor of an unmarked building.

Wall Street Journal : Sources: Anthropic plans to stage its IPO in November, later than many investors expected; the decision was made before ex-Anthropic researcher Jacob Coxon left —  Investors had expected the Claude maker to debut in October; AI industry grappling with calls to slowdown pace of development

深度集成 Gemini、运行 Android 的笔记本电脑 Googlebooks 将于 10 月 4 日上市。Google 硬件合作伙伴中除了宏碁推出一款起售价 899 美元的型号外，其余厂商的产品都超过 1000 美元。Googlebook 不同于 Chromebook 面向低端市场，它面向的是中端笔记本电脑市场。Googlebooks 的 Continue On 功能允许用户在手机或 Googlebook 之间无缝切换，但需要应用开发者支持；Cast My Apps 可以直接在 Googlebook 上使用 Android 手机已安装应用；Play Store 是 Googlebook 获取应用的主要渠道，侧载受到了限制，只能安装运行已通过 Google 验证身份的开发者的应用；通过深度集成 Gemini Intelligence，用户仅仅移动光标就能激活被称为“Magic Pointer”的 AI 功能，AI 会分析屏幕上的内容，根据上下文提供建议，能从多个应用中提取数据。比如将光标指向电邮中的日期即可创建日历预约。

Saturn 的一项研究显示，ChatGPT、Claude、Copilot、Grok 和 Gemini 等主流 AI 模型在回答财务相关问题时，平均有 57% 会给出错误答案。研究使用了逾百个财务相关问题，分别测试了 ChatGPT、Gemini、Claude 和 Copilot 提供的免费及付费 AI 模型。每个问题最多重复提问五次，向 18 种 AI 模型共提出了逾 10,000 个问题。研究发现，AI 模型的回答中包含计算错误、遗漏即将实施的税收政策变更，或是凭空捏造规则（即幻觉）。在最严重的情况下，依赖 AI 对税务问题的回答可能导致严重的经济损失。研究发现，付费模型的回答比免费模型更准确，较新的模型表现优于较旧的模型。表现最好的是推理模式的 Claude Opus 5，但仍然有 39% 的答案存在错误。

OpenAI 在 bzr.openai.com 上的广告收集器会设置名为 __obi、作用域为 .openai.com 的 Cookie。该 Cookie 的值在用户访问 ChatGPT 期间生效，并与用户的 ChatGPT 账户绑定。随后，__obi 会从用户访问的其它网站发送回 OpenAI。任何在 ChatGPT 上购买广告的公司，都会在其自己的网站上添加一小段 OpenAI 代码，其方式与添加 Meta 和 Google 跟踪代码的方式相同。加载该代码会将 __obi 以及有关用户浏览页面的数据（包括搜索的产品、正在阅读的文章以及购买行为）发送给 OpenAI。OpenAI 可将用户在这些网站上的行为与其 ChatGPT 账户关联起来。通过广告收集器 ChatGPT 知道你在其它网站的活动。

微软利用使用 GPT-5.6 Sol 和 Claude Opus 4.8 的 AI 智能体、历时 14.5 周，花费 12 万美元 token 将 Copilot 运行时从 TypeScript 语言移植到 Rust 语言。该项目采用逐个更新运行时模块的方式执行，共进行了 135 次发布，平均每天提交约 1.3 个 Pull Request，最终将 43 万行 TypeScript 代码转换为 80 万行可用于生产的 Rust 代码。测试显示，原 TypeScript 代码每秒能完成 7.55 个生命周期（one-turn session lifecycles），而 Rust 代码每秒 120 个——意味着在特定工作负载下速度提升了 15.9 倍。包含 10 个客户端的智能体在 TypeScript 语言下消耗了 1383 MB 内存，而 Rust 语言版本仅消耗了 126 MB。Rust 版本将任务保持在进程内执行，无需像 TypeScript 版本那样启动外部后台进程完成任务。

根据内部文件，微软应用科学总监 Brent Hecht 警告，抓取新闻内容训练 AI 是一次规模空前的盗窃，可能是人类历史上最大规模的劳动成果盗窃。他不认为抓取新闻内容训练 AI 是合理使用，认为这种做法完全是对合理使用理念的嘲弄。微软的另一份文件认为会形成“恶性循环”，会同时伤害模型和整个 Web。OpenAI 的产品如 ChatGPT 可能会危及新闻出版商的经济基础。OpenAI 和微软的数据都显示这一预测是正确的。微软的数据指出，部分新闻出版商网站的点击率下降了 83%-93%，其它新闻机构的点击率下降了 51%-94%。Hecht 承认，“几乎没有人希望自己创作的内容被以这种方式使用，也没有人因此获得报酬。”微软 CEO 纳德拉(Satya Nadella)在作证时表示，AI 公司不应该通过绕过付费墙违反新闻网站的使用条款。但 OpenAI 的内部信息显示，当一名员工通知总裁 Greg Brockman OpenAI 爬虫找到了绕过《纽约时报》付费墙的漏洞时，Brockman 表示好极了。

Hacktron 安全团队组合利用 OpenAI 的 SSO（单点登录）配置错误以及其社区论坛使用的 Discourse 软件 libheif 软件包堆缓冲区溢出漏洞，成功控制了多名 OpenAI 员工的 ChatGPT 账户。利用这些账户安全研究人员能访问 OpenAI 内部代码库，以及其他关联服务。他们向 OpenAI 和 Discourse 报告了 bug，从 OpenAI 获得了 6500 美元的赏金。安全研究人员利用了 OpenAI 竞争对手 Anthropic 的 Claude AI 工具去辅助发现 bug，以及实现远程代码执行 RCE。他们一开始使用的是较旧的模型 Opus 4.8，之后使用了新发布的 Opus 5.5。整个 AI 辅助 bug 发现和辅助攻击消耗的 token 支出不到 3000 美元，AI 智能体花费了数天时间，而研究人员投入的人工时间仅仅数小时，凸显了 AI 时代黑客攻击成本的低廉。

美国《联邦公报》（Federal Register）网站被发现提供了基于阿里巴巴开放权重模型通义千问（Qwen）的 AI 搜索工具，具体部署时间未知，在被社媒用户注意到之后该工具于周三下线。此前 FBI 曾指控阿里巴巴蒸馏了 Anthropic 的模型。AI 专家表示，在《联邦公报》网站上使用中国 AI 工具未必会构成国家安全风险，但这引发了质疑：联邦政府的做法是否与其针对中国 AI 模型所传达的立场相矛盾。Information Technology and Innovation Foundation 主席 Daniel Castro 指出，美中在构建顶尖 AI 模型上展开激烈竞争，而美国政府机构却选择使用中国的 AI 模型，这太荒诞了。

法国 AI 公司 Mistral 与 Mozilla 合作推出注重隐私保护的 AI 浏览助手 Firefox Smart Window(beta)。Smart Window 使用了 Mistral 的开放权重模型，能帮助用户梳理复杂搜索，记住浏览过的重要信息，根据当前标签页查找关键信息，目前主要为法国和北美用户提供服务，今年晚些时候会扩大到英国和德国用户。Smart Window 的对话内容默认不会存储在 Mozilla 的服务器上，Mistral 等合作伙伴也承诺不保留任何数据。

Mozilla 发表《State of Open Source AI》报告，称美国科技公司的前沿 AI 模型性能仅领先中国公司最优秀的开放权重模型 4.4 个月。由于前沿模型价格更昂贵，很多公司都将开放权重模型用于处理日常工作，将前沿模型用于处理特定工作负荷。美国闭源前沿模型优势主要在于专家级专业工作、高强度检索以及长上下文。 配送公司 DoorDash 将月之暗面的 Kimi 模型用于处理日常工作，将 Anthropic 的闭源模型 Fable 用于更复杂的任务。前沿模型完成一项任务的成本经常达到了最先进开放权重模型的五倍。

今天有无数开发者在 LLM 帮助下编写程序，其中包括了开源开发者。那么 Android FOSS 应用商店 F-Droid 中 AI 辅助开发应用的比例有多高？一位 FOSS 维护者对 9 月 12 日 F-Droid 推送更新的 102 款应用及其代码库进行了分析，发现其中 74 款应用（72.5%）主要是 AI 编写的，10 款应用难以明确归类（9.8%）， 18 款应用几乎没有 AI 参与的迹象（17.6%）。有 4 个托管在 Codeberg 上的应用主要是 AI 编写的，而 Codeberg 最近宣布了 AI 政策，禁止了此类 AI 应用，但要清除此类应用显然需要更多时间。

Yusuf Gundogdu 写道：LibreDB Studio 是一个 MIT 协议的 SQL 客户端，不装在本地而是跑在服务器上，浏览器打开就能用，一条 docker run 就起来。16 个驱动覆盖 42 种数据库，PostgreSQL、MySQL、MongoDB、Redis、ClickHouse 这些都在内。9 月 8 日发布了 0.15.0 版本。我觉得值得一提的是他们把 AI 那部分做了实测：28 个模型跑同一套六项数据库任务，27 个通过 Ollama 完全在本地运行，最快的 qwen2.5:7b 只有 4.7 GB，一次完整运行中位数 6 秒，最小的 2.5 GB。数据逐个模型公开，包括没通过的和卡在哪一步。另外只读不是靠解析 SQL 挡的，是数据库自己挡的：PostgreSQL 上开只读事务，SQLite 上每条语句前重设 query_only。

NVIDIA 宣布将于 2026 年 10 月 15—16 日在苏州举办中国开发者日。日程分为两日：首日为动手实践日，包含 2026 黑客松总决赛路演、全天实战培训及实训营，同时开放四门 Associate 级别认证现场考试；次日为主论坛，设 LLM、AI 模型与运行时、物理AI/机器人、AI基础设施四个分论坛，并举行黑客松颁奖典礼。 值得注意的细节：认证考试可在现场完成，对希望获得 NVIDIA 官方技术背书但不愿单独约考的开发者有一定吸引力。活动偏向实战而非产品宣讲，适合关注推理优化、训练基础设施和机器人方向的工程师。注册已开放，名额限制未公布，建议提早。 https://www.nvidia.cn/developer-day/?ncid=partn-270555