X

Mark Zuckerberg talks about how Meta plans to monetize Muse through transactions --- (full video on official Meta YT channel) https://t.co/Y6MgeVEUnz https://t.co/WNcHIKm7xQ

Full video on official Meta YT channel https://t.co/gaePyGJ8Qt

Mark Zuckerberg showed a fingerprint-activated keychain-sized AI device "Muse Charm" designed to bypass the phone entirely. > roughly 2-inch display and runs a new operating system designed to bring Muse to life on its own hardware. > packs Muse's real-time voice and avatar stack into dedicated hardware. > can be held in the center of a palm and navigated with a thumb on its OLED touch screen. > has 5G connectivity, although Meta hasn't published a complete public specification sheet. > “holdable” rather than a wearable. > It can sit on your wrist, go inside a pocket, or connect to a bag or keychain with a lanyard. The see-through case reveals the electronics within. > No pricing details yet. > With cameras on both the front and back, Muse Charm captures photos and video while mainly giving Muse visual context for analysis and AI-assisted actions. Users can review what they capture on the device. It also supports music and video playback, plus speakers, microphones and a USB-C port. That hardware doesn't replace Muse's agent architecture, because the service already uses a dedicated cloud VM, browser, connected apps, and user permissions. Instead, Muse Charm turns that same agent into a one-touch physical endpoint, shortening the gap between noticing something and asking Muse to act. The product is still early, since Zuckerberg said only a few units exist and Meta is still finalizing parts of the hardware despite its December-26 target. --- (full video on official Meta YT channel)

Hy Image 3.5 preview is now live on GMI Cloud @TencentHunyuan $0.024 per image 8.8x cheaper than GPT Image 2 5.6x cheaper than Nano Banana Pro Create now 👇

4 in 5 Americans are worried AI will escape human control Just 4% are not at all concerned After the first big catastrophe - if we survive it - Trump will go down as the worst president in history. (Unless he does a 180 - which he could, as Newsom is doing right now.) https://t.co/U3VO7PPC05 https://t.co/iyNVY1QhG1

DAILY AI BRIEF 🗞 — Sept 23 OPENAI 🔥: - GPT-6 Sol and Luna rolled out in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu. - API prices are 50% below GPT-5.6 promo: Sol $2/$10, Luna $0.10/$0.50 per 1M tokens. - Free and Go can try Luna in the desktop app. Prompt Caching Dashboard and explicit cache breakpoints also shipped. ANTHROPIC 🔥: - Claude Opus 5.5 is out, first of the 5.5 family. Fable 5.1-level on most tasks, ~40% cheaper and ~30% faster than Opus 5. - Platform pricing is $4/$20 per 1M tokens. Cache reads drop to $0.20. Fast mode goes up to 2.5x at $8/$40. - Pro, Max, and Team get higher 5-hour limits plus a banked reset in Settings through Oct 22. Mobile reset still pending. SPACEXAI 🔥: - Grok in Tesla now runs Connectors: inbox, calendar, files, and chats, hands-free. - Bot can place coffee orders, book reservations, and set appointments. SuperGrok Heavy first, more tiers later. CURSOR 🔥: - Opus 5.5 is live and tops CursorBench at 57.8% Max, 40% less per task than Opus 5. PERPLEXITY 🔥: - Sol is the default Light model in Computer. Opus 5.5 is the Standard effort. DIGITALOCEAN 🔥: - Managed Agents public preview: isolated microVMs, 16k+ tools, pause/resume in ~305 ms. ALIBABA 🔥: - Qwen-Image-2.1 is #1 open-source in Arena Image Edit and Text-to-Image. XIAOMI 🔥: - MiMo-V2.6 Pro and Flash shipped. Pro is MIT open-weight and sits at 46 on AA, matching Grok 4.7. DEEPSEEK 🔥: - China reportedly opened a data-security probe into DeepSeek and Moonshot after Anthropic alleged Claude traffic routing. Independent verification still pending. * Used Grok to compose this brief, cherry-picking the news and doing some post-editing.

Anthropic 创始人 Dario Amodei今天凌晨那篇 ART 长文，一个早上就刷到了 300 多万浏览， 里面有一句我反复看了好几遍，数学可以纯靠推理，生物必须做实验， 但很少有人注意到，昨晚刚发了免费科研工作台 ScienceBuddy 的 @PhAILabs，上周四还开源了一个项目 ScienceIDE，把验证这一关，直接做成了 AI 的对账考场， 上周四，9 月 17 号，其实有两件重磅的事同时落地， →一边是 Anthropic 公布，Claude 用不到四周，优化了 AlphaFold3、Boltz-2 在内的 30 多个生物学 AI 模型，代码全部开源， →另一边是 ScienceIDE，把天体物理的 Athena++、海洋气候的 MITgcm、等离子体的 Gkeyll、暗物质探测的 NEST、量子纠错的 Stim 等 7 个学科的真实科研代码，封装成了 15 套 AI 能动手训练的环境， 判分只认一条铁律， 改完的代码算出来的科学结果，必须和基准参考结果逐点吻合， 跑通不算，对了才算， 就拿 MITgcm 来说，海洋和大气之间氟利昂交换的模块里，埋了一个 bug， AI 要自己找到，自己修好， 合不合格，不看代码写得漂不漂亮，只看最终输出能不能一个点一个点对上， 一开始我也觉得，不就是修个 bug 吗， 后来才反应过来，写业务代码，写错了有编译器和单测当场拦住你， 科学代码里最吓人的 bug，偏偏不报错， 程序照样跑完，曲线照样好看，只是海洋吸收的氟利昂悄悄偏了一点，后面整条气候推论，就跟着一起偏， 拿写业务代码的标尺去量科研 AI，量出来的多半是假及格， 难度也摆在明面上， 最难的 85 道题，8 家厂商的 15 个模型，表现最好的首次尝试正确率是 67.1%， 差不多每三道，还有一道拿不下， 另外说句实在的，现在绝大多数任务是代码修复和重构，加速类任务还在早期， 我反而觉得挺好，考场没被刷穿，才有练的价值， 两件事放在一起，味道就出来了， Anthropic 证明了 AI 有能力改进真实的科学软件，ScienceIDE 想给 AI 一个跨学科练这类活的地方， 团队还说，准备把 Anthropic 开源的优化方案也封装成训练环境，把前沿的做法，变成大家都能拿来练的标准题， 做过科研的人都懂，调代码、排 bug、适配工具、核对结果，能耗掉一个博士生大半的时间， 以前聊 AI 做科研，聊的多半是读文献、写综述、出思路， 现在它开始下到各学科的代码库里，干这些最磨人的活，还要对着基准结果一个点一个点交账， ScienceBuddy现在免费开放，思路是让科研智能体在和研究者的长期协作里，一轮一轮变强， 同一个团队，一边做给 AI 练兵的考场，一边做给科研人当副驾的工作台， 生物学靠湿实验室验证，计算科学靠和基准结果逐点对账， 论文读得再多，综述写得再漂亮，到了几万行的模拟代码面前，还是得把每一个数都对上。

We made claude​.ai 3x faster in two weeks. Here’s how we use Claude to measure, debug and improve performance. Prompts and methods included. https://t.co/mgvK8mJRuD