2026-10-03

AIHOT 日报 · 2026-10-02 — Ataraxos 以85%有效胜率击败最强人类 Stratego 选手,训练成本不足 8000 美元 / 66 signals

Daily Edition

2026-10-0366 signals

Lead Story

3 items

AIHOT 日报 · 2026-10-02 — Ataraxos 以85%有效胜率击败最强人类 Stratego 选手,训练成本不足 8000 美元

研究人员开发的 AI 系统 Ataraxos 以 85% 的有效胜率(平局计半胜)击败了曾获 4 次世界冠军的史上最强 Stratego 选手 Niemeijer,并在 2025 世界锦标赛表演赛中取得 40 局 38 胜。 模型发布/更新 Microsoft AI 发布 MAI-Transcribe-2-Streaming 及 MAI-Voice-2.1 系列语音模型 MiMo-V2.6-Pro 和 MiMo-V2.6-Flash 登陆 Agent Arena,分列开源模型第5和第9 Artificial Analysis 评测:Qwen-Image-2.1 登顶两个 AA-Image 榜单的开源权重模型 产品发布/更新 Claude Code 推出 mods 功能,可用 TypeScript 定制行为与 UI Modal Clusters 正式发布,通过 @modal.clustered 提供多节点 GPU 集群 行业动态 OpenAI 称拦截蒸馏窃取攻击,但研究者称同样手法在 Azure 上仍可窃取 GPT-6 Astra 等模型的推理内容 论文研究 Ataraxos 以85%有效胜率击败最强人类 Stratego 选手,训练成本不足 8000 美元 Transluce 报告 AI 智能体以激进手段访问美加政府网站 物理学者 Matthew Schwartz 分享用 Claude 与 BootLoops 做跨学科计算的经验 技巧与观点 LangChain 讲解如何在 Agent Harness 中构建模型路由器 OpenRouter 指南:用置信度阈值实现模型分级升级路由 OpenRouter 教程:如何在 CI 中用 LLM eval 门禁拦截 Pull Request via AIHOT · https://aihot.news/daily/2026-10-02

Read original ↗

AIHOT 日报 · 2026-10-03 — Google Project Suncatcher 首颗原型卫星发射入轨

Google 宣布其探索在太空托管机器学习基础设施的 Project Suncatcher 已将一颗与 Planet 合作建造的原型卫星送入轨道,搭乘 SpaceX Transporter-18 拼车任务。该任务将收集 Google TPU 在太空飞行物理应力和极端环境下表现的数据,未来探索连接多个卫星星座实现规模化机器学习;低地球轨道系统可借助近乎持续的日照获得最多 8 倍于地面的太阳能。 模型发布/更新 Ai2 开源 8B 科学报告生成模型 AstaBrief NVIDIA 介绍 Blackwell GPU 如何加速 OpenAI GPT-6 Astra Ultrafast 产品发布/更新 NVIDIA DGX Spark 推出 64GB 版本,10 月 23 日起以 $4,999 开售 Suno 推出 Speech beta:语音与背景音乐一体生成 FLUX 3 Image 上线 OpenRouter,支持原生 4K 生成与多参考编辑 ChatGPT 推出 Finances 财务管理功能 行业动态 Google Project Suncatcher 首颗原型卫星发射入轨 加州检察长向 OpenAI 发出传票,调查 AI 智能体网络安全风险 论文研究 Epoch AI 估算 2025–27 年 HBM 可支撑 3000 万至 1.7 亿并发前沿模型智能体 Meta 发布 Muse Spark 与数学家协作完成的六篇数学研究论文 技巧与观点 Manus 分享视频生成与时间线编辑工作流 OpenAI 发布 GPT-6 家族实用指南:选型、提示词与长任务管理 via AIHOT · https://aihot.news/daily/2026-10-03

Read original ↗

AI & Builders

5 items

Sarvam AI 发布从第一性原理构建 AI 智能体的入门指南

Sarvam AI 发布 25 分钟长的智能体构建指南,核心观点是智能体就是在循环中运行、能调用工具的语言模型,而技能、记忆和领域知识本质上都是在合适时机把合适文本放进上下文窗口。指南围绕在线商店客服智能体 ShopBot 逐步展开,覆盖系统提示词、工具设计、渐进式披露的技能、短期与长期记忆、RAG 检索、智能体拆分原则,并以完整端到端示例、常见错误清单和构建检查表收尾。 🔗 阅读原文 via AIHOT · https://aihot.news/items/h40ygivbwihjifohdumbgpo0j

Read original ↗

FDE & Product Engineering

1 items

Tech & Industry

15 items

Meta open sources code to let you make Muse AI gadgets

Meta now lets you make your own Muse gadgets that feature the company's new AI agent with code that the company open sourced. The company suggests projects like loading Muse on a color E Ink display to show reminders, adding it to an HDMI stick so you can display Muse on a big screen, or […]

Read original ↗

It’s not AI anymore, it’s ‘super intelligence’ (according to the White House)

This week, the White House got nearly every major tech CEO in one room — Zuckerberg, Bezos, Musk, and Anthropic’s Dario Amodei among them — to sign an AI safety pledge that President Donald Trump called “morally binding.” Trump also signed an executive order officially rebranding AI as “super intelligence,” and meanwhile, Meta and OpenAI are putting friendlier faces on their AI products, even as the biggest money […]

Read original ↗

Netflix is pivoting away from prestige

Many of Netflix's biggest critically acclaimed hits have been the products of its multiyear production deals with noted directors like David Fincher and Shawn Levy. In the past few weeks, though, the streamer has brought some of those partnerships to an end in a way that feels indicative of a larger strategic shift. With rival […]

Read original ↗

Rivian’s sales pop as the company’s big R2 bet starts to pay off

Rivian had high hopes for its more affordable R2 vehicle - and so far, those hopes appear to be paying off. The company released its third-quarter production and delivery numbers today, reporting 19,751 vehicles produced and 19,248 delivered. That represents an 85 percent year-over-year increase in production and a 45 percent jump in deliveries. (For […]

Read original ↗

Sling TV drops its one-day cable passes

Dish-owned Sling TV will no longer be offering its Sling Pass feature that allowed people to buy a single day of cable TV programming at a time, as reported by The Desk. The feature was announced last year and offered a day pass, a weekend pass, and a week-long pass option to be able to […]

Read original ↗

Less than 24 hours to apply for a Side Event at Founder Summit 2026

The clock is almost out. You have less than 24 hours left to apply to host a Side Event during TechCrunch Founder Summit 2026. Applications close tonight at midnight PT. Connect with the Boston startup ecosystem through November 1–7 by hosting your own Side Event. Bring your community, start an impactful conversation that’ll push the […]

Read original ↗

AI Travel & Travel Tech

10 items

Forget Me Not

Twilio's 2026 APAC research finds 70% of consumers have abandoned AI interactions due to lost context, exposing hotels' fragmented guest data infrastructure as the root cause and proposing a governed Guest Memory Layer as the fix.

Read original ↗

World Affairs

15 items

Design & Culture

1 items

Outside the Bubble

1 items

Life on a hair trigger

A world of violence sustains itself through guns, poverty and other social ills, but also through the minds it creates - by Megan Kang Read on Aeon

Read original ↗

Builders on X

15 items

Cool eval. Simply ask an LLM “Land or Water?” and give it a latitude and longitude coordinate as text. Ask 16,200 times, plot as image. The models know. From compressing the internet. https://t.co/jTTAhBsMfu

Open on X ↗

Tantalum being discovered in 1802 and turning out to have an isotope with a half life so long that we might never see it decay is a great example of nominative determinism.

Open on X ↗

Down to 6110 unread from the over 9000 unread emails just earlier. Will get to inbox zero with clean filters within the next 48 hours. Dot has accepted the challenge. https://t.co/9oVJDEJMfI

Open on X ↗

Oh. My. God. https://t.co/9kiOZJn2DN https://t.co/AKMHF1xANp

Open on X ↗

Here's the app. One thing I've been liking is to ask models to 'teach me back' what they did¹ by including "quines". Quines² are programs that output their own source code. So, in this demo you can reveal step by step the Svelte code powering it! https://t.co/CTTyWd2w3I ¹ Understanding is still great 😆 ² https://t.co/Ck4QwZt0cL

Open on X ↗