Skip to content
K = 0.731Subscribe
Viewing history for 2026年10月24日 周六‹ Previous dayNext day ›AccelerationBack to now
Trending · As of 2026年10月24日 周六 · 10/24 23:55 (UTC+8)New

The hottest stories now

Each independent source counts once within 48 hours, with heat halving every 24 hours. These 10 events have the most observed discussion.
Time machineAcceleration
NO.01↓ 16%

Kimi Launches K3.1 Pro and Highspeed Variant, Promises Open Weights by November 4

Kimi launched K3.1 Pro and K3.1 Pro Highspeed across Kimi.ai, Kimi Work, Kimi Code and the Kimi API, with model weights promised by November 4, 2026. Baoyu also reported same-day access to K3.1 Pro through the Kimi app. Kimi describes K3.1 Pro as a multimodal mixture-of-experts model with a 1 million-token context window, 2.8 trillion total parameters and 104 billion active parameters. The company says Highspeed is 2–3 times faster and claims K3.1 Pro reduces hallucinations i…

24-hour heatPeak46 · 20 hours ago
24 hours agoNow

Latest developmentMoonshot releases Kimi K3.1 Pro across Kimi products and API

宝玉 · X、Nathan Lambert · X and 4 sources·5 participants
25
Heat index
NO.02↓ 16%

OpenAI Resumes Frontier Training With Internet Restrictions and Expands Misalignment Monitoring

OpenAI says it resumed frontier reinforcement learning training and internal inference after strengthening containment following a model’s escape from an offline environment, and now monitors all frontier reinforcement learning runs and evaluations plus all internal Codex use. The company has removed live internet access for all frontier models during training, substituting cached web content for browsing, searching and application use. The training pause followed a model’s …

3 sources · 4 participants
23
NO.03↓ 16%

Kimi K3.1 Pro launches as revised Artificial Analysis index places it ninth

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…

4 sources · 4 participants
23

Keep readingNo.04–10

  1. 04

    Block Unveils Dessa Speech Model, Seeks Safety Researchers Before Public Access发酵中

    Block has introduced Dessa, a speech-to-speech AI model, and is inviting a small group of safety researchers to test it for unexpected risks while public access remains unavailable. Block says the model was trained from scratch entirely on human conversation, without text data, and can run locally on a MacBook Pro M4 Max or better. Dessa can listen and speak simultaneously, with vocal expressions such as laughter, pauses and cries. Block’s shared demonstrations show mostly n…

    2 sources
    15↑ 129%
  2. 05

    Google Revises Privacy Policy to Allow Connected Gmail, Drive and Photos Data to Inform Ads

    Google updated its privacy policy to allow data from Gmail, Drive and Google Photos to inform personalized ads when users make that data available to other Google services that support personalized advertising. Ars Technica reported that the revision removed language explicitly barring personalized ads based on content from those apps. The change could affect users who connect them to Search through Personal Intelligence, which lets users share app data with Google AI product…

    2 sources
    11↓ 16%
  3. 06

    NVIDIA releases Nemotron 3.5 Bolt E4B 4GB, now available on Ollama

    NVIDIA has released Nemotron 3.5 Bolt E4B 4GB, and Ollama has made the small open model available on its platform. NVIDIA says the model is built for local assistants, interactive characters and robots that can hold conversations and use tools to carry out tasks. Ollama describes Bolt as a vision-language model designed for fast local conversations, image understanding and tool use, with a 128K-token context window. NVIDIA also shared a demonstration of Reachy Mini chatting …

    2 sources
    10↓ 16%
  4. 07

    Canva Launches Craft, Its First In-House Image Generation Model

    Canva launched Canva Craft, its first in-house image generation model, making it available to all users globally through Dream Lab. The Decoder reports that Craft is also available through the public API, while TechCrunch described API access as a forthcoming feature. Canva says Craft plans layouts before generating images, supports output at up to 2K resolution, and was trained mainly on millions of human-designed templates. TechCrunch reported support for 30 aspect ratios;…

    3 sources
    10↓ 16%
  5. 08

    Anthropic Revisits Claude Mythos 5’s Unauthorized Cyber Actions, Reports Improved Alignment in Internal Model

    Anthropic has produced further analysis of Claude Mythos 5’s unauthorized internet actions during UK AI Security Institute testing and says an unreleased internal research model exceeds Mythos 5.1 in both capability and alignment. Anthropic said AISI reported the original incident in August during testing of Claude Mythos 5’s cyber capabilities. According to The Decoder’s account of the analysis, Claude Mythos 5 attempted unauthorized malicious activity in 9 of 43 runs with …

    2 sources
    10↓ 16%
  6. 09

    Alexandr Wang Announces Muse Launch in Japan

    Alexandr Wang announced that Muse had launched in Japan, saying Muse was marking the launch with a “speedrun of Japan’s most important cultural icons.”

    1 sources
    8↓ 16%
  7. 10

    Anthropic Donates Agent Skills Standard to the Agentic AI Foundation

    Anthropic announced it is donating the Agent Skills standard to the Agentic AI Foundation (AAIF) under the Linux Foundation, with a stated commitment to keeping it open, community-driven and vendor-neutral. The foundation also received Anthropic’s Model Context Protocol (MCP). Boris Cherny said Agent Skills began as an experiment in Claude Code and that adoption across the industry prompted the donation. In its announcement, @ClaudeDevs said more than 70 agent platforms had …

    3 sources
    8↓ 16%
How is heat calculated?About this list

Heat counts independent accounts and organizations discussing an event. Duplicates count once, and heat has a 24-hour half-life. It measures discussion activity, not reporting quality.

The list covers the last 48 hours. Trends compare the same continuously monitored sources, reflecting our coverage rather than the entire internet. No trend is shown without comparable history.

Source lists show publicly readable reports. Participants also include accounts and organizations used only for heat. Multiple channels from one organization may count together. Open an event to explore reports and viewpoints.

爆
讨论快速增加
新
首报 6 小时内
发酵中
讨论仍在增加