跳到正文
K = 0.731订阅
时光机停在2026年10月28日 周三‹ 前一天后一天 ›加速史回到现在
热点榜 ·截至 2026年10月28日 周三 · 10月28日 23:55更新

此刻温度最高

48 小时内每个独立来源只算一次，热度每 24 小时减半。排在前面的，是真正有很多人在说的 9 件事。
时光机加速史
NO.01↓ 16%

Google DeepMind 发布 Genie 4 预览版，新增音频、更长模拟时长并计划开放 API 访问

Google DeepMind has released Genie 4, a model that generates interactive virtual worlds, as a research preview for trusted testers, with consumer and developer access still planned. Google says access is coming soon to Google AI Ultra subscribers through Project Genie and early access partners through the Gemini API and Google AI Studio; TechCrunch reports that rollout is planned in the coming weeks. Google has also opened a Project Genie waitlist. DeepMind says Genie 4 adds…

24 小时热度峰值58 · 21 小时前
24 小时前现在

最新进展Genie 4 Explained: Technical Challenges in Moving From Video Generation to World Control

Demis Hassabis · X、The Decoder 等 5 个来源·6位参与者
32
热度指数
NO.02新上榜

NVIDIA 发布 Nemotron 3.5 Transcribe，vLLM 从发布首日便提供服务支持

NVIDIA released Nemotron 3.5 Transcribe, an open-weight speech model, with vLLM announcing launch-day serving support through vLLM and NeMo Speech. NVIDIA described it as a promptable 30B mixture-of-experts model with approximately 4B active parameters that combines transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding in one checkpoint. The Decoder reported that the model follows text instructions to process recordings…

2个来源 · 2位参与者
20
NO.03发酵中↑ 102%

修订后的 Artificial Analysis 指数将 Kimi K3.1 Pro 排在第九位，该模型正式发布

Moonshot AI released Kimi K3.1 Pro through its apps and API, but its weights remain unreleased, and Artificial Analysis’s revised Intelligence Index now places it ninth, behind eight closed models. The Decoder reported that version 5 assigns 60% of the index to private test sets to counter training on known questions. K3.1 Pro scored 46, trailing the leader by 12 points; under version 4.3, it ranked fourth with 55, three points behind Claude Opus 5.5. Ethan Mollick urged caut…

2个来源 · 2位参与者
17

继续看No.04–09

  1. 04

    OpenAI 发布 GPT-6.1 Terra，支持电脑操作并降低 API 价格

    OpenAI has released GPT-6.1 Terra, a model it says can interpret screens and act across apps, with access available through the API and to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, according to The Decoder. Artificial Analysis reports API prices of $1 per million input tokens and $5 per million output tokens, down from GPT-5.6 Terra’s $2 and $12; The Decoder lists cached input at $0.05 per million tokens. In Artificial Analysis evaluations at maximum rea…

    5个来源
    14↓ 16%
  2. 05

    Anthropic 与六家药物研发机构合作，并启动自主药物发现工作

    Anthropic says its scientists have formed research partnerships with six organizations developing new medicines and have also begun their own drug discovery efforts. The partnerships have been underway for the past few months. According to Baoyu’s account of Anthropic’s announcement, Claude is being used for drug design, experiments and preparation of regulatory submission materials. That account says human clinical trials could begin as early as the end of 2028; it presents…

    2个来源
    11↓ 16%
  3. 06

    Anthropic拟议的IPO架构将让创始人和安全信托掌控公司，限制公众股东的控制权

    Anthropic has proposed an IPO governance structure that would leave public shareholders with limited control, giving its founders and Long-Term Benefit Trust the power to choose the board to protect its AI safety mission. The trust would elect and remove four of seven directors, while a majority of Anthropic’s seven founders would control 50.1% of shareholder voting power through Founder LLC and could elect the remaining three. TechCrunch reported that employees could resolv…

    2个来源
    11↓ 16%
  4. 07

    Liquid AI 发布六款开放权重 LFM3 模型，可通过 Ollama 在本地运行

    Liquid AI has released LFM3, a family of six open-weight models in the 2B parameter range available on Hugging Face, and Ollama has made LFM3 available to run locally. The family shares a backbone and covers text, vision, audio, omni and calibrated decision tasks; the Hugging Face release includes inference examples and fine-tuning notebooks. Liquid AI says every model supports fine-tuning on a single GPU. Its text model has a 256K context window and, according to the compan…

    2个来源
    11↓ 16%
  5. 08

    Suleyman 报告称，MAI Cyber 的采用率上升，Token 效率提升

    Mustafa Suleyman 表示，在过去 30 天内，MAI Cyber 在微软安全测试框架中的用户采用率翻了一番，达到 50%。他还表示，该模型的 Token 效率提升了 20%，且质量没有下降。

    1个来源
    10↓ 15%
  6. 09

    微软 MAI-Image-2.7-Preview 在 Artificial Analysis 图像编辑基准测试中排名第二

    Microsoft’s MAI-Image-2.7-Preview placed second on Artificial Analysis’s AA-Image-Editing v2.1 benchmark and is now available for image editing in MAI Playground, according to Artificial Analysis. It ranked close behind OpenAI’s GPT Image 2.5 Sunburst (max). Mustafa Suleyman shared the result, calling MAI-Image “the #2 image edit model in the world.” Artificial Analysis said the preview outperformed MAI-Image-2.6 across all seven measured editing actions and ranked first in …

    2个来源
    10↓ 15%
热度是怎么算的？了解榜单

热度来自参与同一事件的独立账号与机构，重复采集只算一次，并按 24 小时半衰期衰减。它衡量讨论活跃程度，不是报道质量评分。

榜单统计过去 48 小时。趋势只比较持续覆盖的同一组信源；它反映我们的监测范围，不代表全网人数。缺少可比历史时，不展示趋势线。

信源名单只展示可公开阅读的报道来源；讨论参与者还包括只计入热度的账号与机构。同一机构的多个渠道可能合并计数，因此参与者不一定多于信源数。点击事件可查看各方报道与观点。

爆
讨论快速增加
新
首报 6 小时内
发酵中
讨论仍在增加