Qwen3.8-Flash-Next
Simon Willison 记录了自己在本地试用 Qwen3.8-Flash-Next 的过程。该模型是 Qwen 面向 Qwen4 架构的多模态 MoE 预览，开放权重后可作为本地推理与模型选型的观察样本。
2026-08-27
Qwen3.8-Flash-Next / 16 signals
Lead Story
3 itemsGoogle’s new AI transcription edits out your ‘ums’ and ‘ahs’
Google 为 Gemini Audio 增加转写能力：可识别专业术语、覆盖 85 种以上语言，并能清理口语中的停顿词。语音交互产品的差异开始从“能否转写”转向可用性和后处理质量。
Surprise: Z.ai is the AI lab behind the mysterious Ox Alpha model
Z.ai 确认其是排行榜模型 Ox Alpha 的背后团队，并表示将发布权重。对模型使用者而言，重点不只是榜单成绩，还要跟进开放权重后的许可、部署成本和实际任务表现。
AI & Builders
3 items🔬“We have foundation models for language, not for physics” — Anima Anandkumar, Bren Professor of Computing
Anima Anandkumar 讨论用 AI 建模物理世界的研究路径，覆盖天气与聚变等场景。它提醒我们，基础模型的下一段增量不只来自语言数据，也来自可验证的物理与科学问题。
[AINews] Hot Chips: OpenAI’s Jalapeño, Cerebras CS-5, Groq 3 LPX, Apple M6
本期 Hot Chips 速览汇集了 OpenAI、Cerebras、Groq 与 Apple 的芯片和计算系统动态。适合作为 AI 性能竞争从模型能力延伸到算力、内存与系统架构的索引，而非单条结论。
Lovable CTO: The Future of SaaS Is Apps That Agents Can Use
Lovable CTO 讨论了公司从 AI 网页应用生成扩展到基于 MCP 的能力层。关键观察是：未来 SaaS 的接口和权限设计，需要让人和 Agent 都能可靠调用。
Tech & Industry
3 itemsFlipboard acquires Graze, the feed builder working to monetize the open social web
Flipboard 收购 Bluesky feed-building 初创公司 Graze，并把其隐私友好广告技术与创作者变现工具纳入开放社交网络战略。分发层、个性化 feed 与变现能力正在重新绑定。
A New NASA Design Turbocharges Nuclear Spacecraft
NASA 与产业工程师提出同步双模核火箭概念：用单一反应堆同时支持高推力核热推进和持续电力。方案仍处于模型与工程验证阶段，燃料、地面测试和发射安全是主要门槛。
Apple Maps has ads now
Apple Maps 已开始在搜索和“推荐地点”位置展示付费推广结果，这是 Apple 既有广告产品向本地搜索入口的延伸。地图类产品的中立排序、商业化与用户信任将需要同时平衡。
Design & Culture
1 items20 years, a bunch of photographers, one very sweaty archive: how to shoot The Warehouse Project
The Warehouse Project 用新档案回看二十年舞曲现场摄影。内容关注摄影师如何在拥挤、短暂且高能量的现场建立一致的视觉记录，适合从“档案如何形成风格”而非单次作品来阅读。
Builders on X
6 itemsPSA: do not use codex "locked use" capabilities right now. it is currently relying on unstable mac features and has completely locked me out of my macos keychain twice this week. thx @_chenglou for linking to apple developer forums acknowledging this is a "known bug". just avoid. ofc, would be nice to do everything in cloud, but cloud isn't there yet.
Open on X ↗Good products take time. At least 34 days.
Open on X ↗find someone who loves you as much as the VC who loves posting photos of an “exclusive” dinner 🙈
Open on X ↗Good post on what the applied AI strategy looks like at scale. It’s clear that there’s a wide gap between the AI models and the underlying workflows of an enterprise, which leaves a ton of opportunity for applied AI companies. “The world doesn’t just want raw models and agents; it wants problems resolved and outcomes achieved. The premium will sit with the companies that can diffuse this intelligence through every aspect of civilization, converting raw tokens into real world outcomes, transforming industries, and creating economies in the process.” This requires understanding the context, driving the change management, having a harness that can route to various models, connecting to the critical business systems in that vertical, solving the UX challenges of connecting users to agents in the right way in a workflow, understanding the evals in the space, and so on. That’s a ton of value that goes beyond just the model intelligence itself. And there’s a window of opportunity right now to build the defining companies that can bring intelligence to the critical domains in an enterprise.
Open on X ↗Full eval series so far: 1/ Getting started with evals https://t.co/fh4yQR64Ta 2/ Quality first, then cost https://t.co/hqNaE2ykTQ 3/ Failure modes taxonomy https://t.co/MM2wgiAnoF 4/ Laddered eval strategy https://t.co/2ioAe4hvqb 5/ Tyranny of the average https://t.co/QPiGbxxb72 6/ Hill climbing on evals https://t.co/EtxbF1kqgD 7/ The Goldilocks principle https://t.co/PXSX3jsLtk 8/ Discriminatory power of evals https://t.co/epINGK6dGV 9/ The Eval Roadmap Problem https://t.co/nw2U3R6U1Y
Open on X ↗