从 119 条内容中筛选出 20 条重要资讯

  1. Unsloth 发布 Mistral Le Chonk 1-bit GGUF 量化版，可本地运行 1T 参数模型 ⭐️ 9.0/10
  2. huggingface/transformers released v5.22.0 ⭐️ 8.0/10
  3. Binance-hacker-1029 – $5.6B stolen from Binance ⭐️ 8.0/10
  4. We’re releasing the weights of Mistral Large 4 ⭐️ 8.0/10
  5. LLM Voice ⭐️ 8.0/10
  6. DoorDash to acquire Instacart ⭐️ 8.0/10
  7. Chained PHP2Shell exploit roots 25 servers, including US Government servers ⭐️ 8.0/10
  8. How we detect frontier model misalignment ⭐️ 8.0/10
  9. OpenAI-backed Thrive Holdings acquires ADT ⭐️ 8.0/10
  10. Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity ⭐️ 8.0/10
  11. US investigated Chinese spy plot aimed at Nvidia CEO ⭐️ 8.0/10
  12. The Verge 解读 Anthropic S-1：约 8500 亿美元 compute 承诺押注递归自我改进 ⭐️ 8.0/10
  13. H.264 Is Magic (2016) ⭐️ 7.0/10
  14. An updated Model Spec for more capable and beneficial AI ⭐️ 7.0/10
  15. Announcing Project ProtoStar: Making Publicly Funded Science AI-Ready ⭐️ 7.0/10
  16. Welcome Le Chonk: Mistral Large 4 on Hugging Face ⭐️ 7.0/10
  17. Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platforms ⭐️ 7.0/10
  18. AWS will be a $1T business, Jassy says. But first, it will spend big. ⭐️ 7.0/10
  19. Binance pauses withdrawals as hackers suspected of stealing billions in crypto ⭐️ 7.0/10
  20. GM reaches $220M settlement over selling drivers’ private data ⭐️ 7.0/10

Unsloth 发布 Mistral Le Chonk 1-bit GGUF 量化版，可本地运行 1T 参数模型 ⭐️ 9.0/10

Unsloth released a 1-bit GGUF quantized version of Mistral Large 4 (1T parameter MoE) that preserves 91% top-1 accuracy and runs locally on Mac Studio, reducing size from 1.1TB FP8 to ~233GB.

rss · AIHOT 精选 · Oct 29, 14:27AIHOT 精选

标签: #1-bit Quantization, #MoE, #GGUF, #Local AI, #Model Compression

huggingface/transformers released v5.22.0 ⭐️ 8.0/10

huggingface transformers v5.22.0 introduces Mistral Large 4, a 1T-parameter multimodal Mixture-of-Experts model surpassing frontier closed models on visual grounding.

github · LysandreJik · Oct 29, 14:15

标签: #transformers, #mistral-large, #multimodal, #mixture-of-experts, #large-language-model

Binance-hacker-1029 – $5.6B stolen from Binance ⭐️ 8.0/10

A reported $5.6 billion hack of Binance has triggered massive community scrutiny, with on-chain data showing significant funds frozen by stablecoin issuers and speculative discussions about potential BGP/DNS compromise.

hackernews · Tiberium · Oct 29, 21:38 · 社区讨论T2 社区与开源

标签: #crypto, #exchange-hack, #cybersecurity, #blockchain, #binance

We’re releasing the weights of Mistral Large 4 ⭐️ 8.0/10

Mistral releases weights for Mistral Large 4 under a restrictive license that prohibits commercial Model-as-a-Service use without a separate agreement, sparking debate about open-weight licensing while demonstrating competitive performance and enterprise adoption.

hackernews · kuberwastaken · Oct 29, 14:13 · 社区讨论T2 社区与开源

标签: #LLM, #Mistral, #Open-Weights, #AI-Licensing, #Enterprise-AI

LLM Voice ⭐️ 8.0/10

Martin Fowler examines the recognizable ‘voice’ of LLM-generated writing and why it feels off-putting to readers.

hackernews · ibobev · Oct 29, 14:57 · 社区讨论T2 社区与开源

标签: #LLM, #technical-writing, #AI-generated-content, #communication, #martin-fowler

DoorDash to acquire Instacart ⭐️ 8.0/10

DoorDash announces acquisition of Instacart in a major food/grocery delivery industry consolidation move.

hackernews · WalterSobchak · Oct 29, 12:33 · 社区讨论T2 社区与开源

标签: #business, #mergers-acquisitions, #food-delivery, #antitrust, #y-combinator

Chained PHP2Shell exploit roots 25 servers, including US Government servers ⭐️ 8.0/10

A chained PHP2Shell exploit combined with a Linux kernel vulnerability compromised 25 servers, including US government systems, highlighting critical attack surface risks.

hackernews · 882542F3884314B · Oct 29, 10:42 · 社区讨论T2 社区与开源

标签: #cybersecurity, #exploit, #linux-kernel, #php, #vulnerability

How we detect frontier model misalignment ⭐️ 8.0/10

OpenAI explains their layered monitoring system for detecting frontier model misalignment and unauthorized agent behavior to enable early intervention.

rss · OpenAI Blog · Oct 29, 09:00T1 官方实验室

标签: #AI Safety, #Model Monitoring, #Alignment, #OpenAI, #Frontier Models

OpenAI-backed Thrive Holdings acquires ADT ⭐️ 8.0/10

OpenAI-backed Thrive Holdings acquires ADT in a $15.3 billion deal to integrate AI into the home security business.

rss · TechCrunch · Oct 29, 19:56T1 官方实验室

标签: #AI, #Home Security, #Mergers & Acquisitions

Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity ⭐️ 8.0/10

Mistral releases open weights for a trillion-parameter AI model focused on enhancing organizational control over cyber defenses.

rss · TechCrunch · Oct 29, 17:00T1 官方实验室

标签: #AI, #OpenSource, #Cybersecurity, #LargeLanguageModel, #Mistral

US investigated Chinese spy plot aimed at Nvidia CEO ⭐️ 8.0/10

A top executive of a Singapore AI cloud company accused of illegally selling Nvidia chips to China is also suspected of being a Chinese spy, according to New York Times sources.

rss · TechCrunch · Oct 29, 16:42T1 官方实验室

标签: #AI Hardware, #Semiconductor Politics, #Espionage, #US-China Tech Relations

The Verge 解读 Anthropic S-1：约 8500 亿美元 compute 承诺押注递归自我改进 ⭐️ 8.0/10

Anthropic’s S-1 filing reveals approximately $850 billion in future compute and infrastructure commitments, underscoring the extreme capital demands of frontier AI development and recursive self-improvement strategies.

rss · AIHOT 精选 · Oct 29, 13:30AIHOT 精选

标签: #Anthropic, #AI infrastructure, #compute scaling, #recursive self-improvement, #AI finance

H.264 Is Magic (2016) ⭐️ 7.0/10

A 2016 analysis of H.264’s technical capabilities and optimization techniques, enriched by community commentary from video codec experts.

hackernews · ea016 · Oct 29, 15:53 · 社区讨论T2 社区与开源

标签: #h264, #video-codec, #x264, #lossless-encoding, #multi-core-optimization

An updated Model Spec for more capable and beneficial AI ⭐️ 7.0/10

OpenAI releases an updated Model Spec defining five core values to guide AI models toward sound judgment, human agency, and faithfulness to human intent.

rss · OpenAI Blog · Oct 29, 17:00T1 官方实验室

标签: #AI alignment, #AI safety, #OpenAI, #model governance, #responsible AI

Announcing Project ProtoStar: Making Publicly Funded Science AI-Ready ⭐️ 7.0/10

Project ProtoStar aims to make publicly funded scientific data and research AI-ready, enabling broader access and usability of scientific knowledge through Hugging Face’s open-source ecosystem.

rss · Hugging Face Blog · Oct 29, 00:00T1 官方实验室

标签: #open-science, #ai-readiness, #scientific-data, #machine-learning, #research-infrastructure

Welcome Le Chonk: Mistral Large 4 on Hugging Face ⭐️ 7.0/10

Mistral AI releases Mistral Large 4 on Hugging Face, introducing a new flagship large language model to the open-source community.

rss · Hugging Face Blog · Oct 29, 00:00T1 官方实验室

标签: #Mistral Large 4, #Large Language Models, #Hugging Face, #AI Model Release, #Mistral AI

Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platforms ⭐️ 7.0/10

Hugging Face releases open models AnySpeechRecognizer and AnyDecisionModel, enabling speech recognition and decision-making capabilities optimized for Apple platforms.

rss · Hugging Face Blog · Oct 29, 00:00T1 官方实验室

标签: #Speech Recognition, #Decision Models, #Apple Platforms, #Core ML, #Open Source AI

AWS will be a $1T business, Jassy says. But first, it will spend big. ⭐️ 7.0/10

AWS CEO Jeff Jassy outlines the company’s strategy to grow AWS into a $1 trillion business by heavily investing in AI cloud computing infrastructure.

rss · TechCrunch · Oct 29, 23:42T1 官方实验室

标签: #aws, #ai cloud computing, #capital expenditure, #amazon, #cloud strategy

Binance pauses withdrawals as hackers suspected of stealing billions in crypto ⭐️ 7.0/10

Binance paused withdrawals following reports of a massive cyberattack potentially stealing billions in crypto.

rss · TechCrunch · Oct 29, 22:27T1 官方实验室

标签: #cryptocurrency, #security, #binance, #cyberattack, #financial

GM reaches $220M settlement over selling drivers’ private data ⭐️ 7.0/10

GM agrees to a $220M settlement requiring express driver consent before collecting or sharing driving and location data.

rss · TechCrunch · Oct 29, 21:01T1 官方实验室

标签: #data-privacy, #automotive, #consent, #legal-settlement, #tech-industry

附录

以下内容来自 AIHOT 日报，每天 08:00 北京时间发布

  1. 【AIHOT 通知】订阅地址换到 aihot.news 了，请更新

    你正在用旧地址 aihot.virxact.com 订阅 AIHOT。AIHOT 已换到新域名 aihot.news，旧地址将于 2026 年 10 月 31 日起只做跳转，之后会停用。

请在阅读器里把这个订阅的地址换成 https://aihot.news/feed/daily.xml，内容和以前一样。

— From rss —

你正在用旧地址 aihot.virxact.com 订阅 AIHOT。AIHOT 已换到新域名 aihot.news，旧地址将于 2026 年 10 月 31 日起只做跳转，之后会停用。

请在阅读器里把这个订阅的地址换成 https://aihot.news/feed.xml，内容和以前一样。

  1. AIHOT 日报 · 2026-10-29 — Google DeepMind 发布 Gemini 4 Carbon 前沿模型

    Google DeepMind 发布 Gemini 4 Carbon，定位为最强前沿模型，主打网络安全与长程智能体能力，在 SEC-bench Pro 88.2%、ExploitGym 54.8%、CWE-bench 71.0% 取得领先，并以 59 分登顶 Artificial Analysis Intelligence Index v5.0。

模型发布/更新

产品发布/更新

行业动态