从 133 条内容中筛选出 20 条重要资讯

  1. DuckDB 2.0.0 LTS “Cinnamon Teal” ⭐️ 10.0/10
  2. Eight Bits of Trust: eMudhra’s CA keys cracked in milliseconds ⭐️ 9.0/10
  3. Today’s PHP security release, and how we’re strengthening security ⭐️ 8.0/10
  4. GitLab Critical Patch Release ⭐️ 8.0/10
  5. Compromised LiteLLM 1.107.0 on PyPI Installs a Credential Stealer and Backdoor ⭐️ 8.0/10
  6. OpenAI models are coming to Google Cloud ⭐️ 8.0/10
  7. NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required ⭐️ 8.0/10
  8. How Specializing the Vocabulary of SLMs Can Cut Inference Cost by 77% in Real Applications ⭐️ 8.0/10
  9. ollama/ollama released v0.40.8 ⭐️ 7.0/10
  10. Claude Duet ⭐️ 7.0/10
  11. Meta ordered to pay $25B penalty in New Mexico Cambridge Analytica case ⭐️ 7.0/10
  12. Patch the Planet: global response coordination as agents uncover vulnerabilities ⭐️ 7.0/10
  13. EA is shutting down BioWare ⭐️ 7.0/10
  14. PHP 8.2.35 ⭐️ 7.0/10
  15. Using thousands of AI agents to uncover new leads in biological data ⭐️ 7.0/10
  16. How we built ChatGPT’s novel ad stack for user trust and relevance ⭐️ 7.0/10
  17. Democratic governance of frontier AI, part II: Empowering cyber defenders ⭐️ 7.0/10
  18. Southeast Asia’s digital economy to hit $368 billion ⭐️ 7.0/10
  19. Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium. ⭐️ 7.0/10
  20. Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster ⭐️ 7.0/10

DuckDB 2.0.0 LTS “Cinnamon Teal” ⭐️ 10.0/10

DuckDB 2.0 LTS introduces client-server architecture, massive parallel scaling improvements, and warehouse features like Triggers, positioning it as a potential replacement for traditional data stacks.

hackernews · nnx · Oct 21, 14:01 · 社区讨论T2 社区与开源

标签: #DuckDB, #Database, #Analytics, #OLAP, #Data Engineering

Eight Bits of Trust: eMudhra’s CA keys cracked in milliseconds ⭐️ 9.0/10

Researchers cracked eMudhra’s Certificate Authority keys in milliseconds, exposing a critical flaw in global PKI trust and prompting urgent advice to update CA bundles across all systems.

hackernews · hkc · Oct 21, 09:08 · 社区讨论T2 社区与开源

标签: #PKI, #Certificate Authority, #Security, #IoT, #Web Trust

Today’s PHP security release, and how we’re strengthening security ⭐️ 8.0/10

PHP Foundation releases critical security update (8.2.35) for an AI-discovered RCE vulnerability and outlines plans to strengthen PHP’s security posture.

hackernews · ejohnso49 · Oct 21, 15:48 · 社区讨论T2 社区与开源

标签: #PHP, #Security, #Vulnerability, #AI-Security, #RCE

GitLab Critical Patch Release ⭐️ 8.0/10

GitLab releases critical 19.5.1 patch addressing a severe unauthenticated vulnerability, the second regex-related CVE in recent weeks, prompting urgent upgrade discussions.

hackernews · artooro · Oct 21, 19:01 · 社区讨论T2 社区与开源

标签: #security, #gitlab, #vulnerability, #devops, #patch-management

Compromised LiteLLM 1.107.0 on PyPI Installs a Credential Stealer and Backdoor ⭐️ 8.0/10

LiteLLM 1.107.0 on PyPI was compromised to install a credential stealer and backdoor, exposing users to supply chain risks in AI tooling.

hackernews · abhisek · Oct 21, 14:00 · 社区讨论T2 社区与开源

标签: #supply-chain-attack, #pypi, #credential-stealer, #backdoor, #ai-toolkit

OpenAI models are coming to Google Cloud ⭐️ 8.0/10

OpenAI makes its models available on Google Cloud, expanding beyond Microsoft’s exclusive partnership and enabling GCP enterprises to use OpenAI inference with existing commitments.

hackernews · OpenAI Blog · Oct 21, 12:53 · 社区讨论T2 社区与开源

标签: #OpenAI, #Google Cloud, #AI Infrastructure, #Cloud Partnerships, #Enterprise AI

NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required ⭐️ 8.0/10

NVIDIA’s Kumo introduces a relational AI approach that outperforms supervised machine learning on relational data without requiring traditional training labels.

rss · Hugging Face Blog · Oct 21, 18:26T1 官方实验室

标签: #relational AI, #graph neural networks, #machine learning, #NVIDIA, #unsupervised learning

How Specializing the Vocabulary of SLMs Can Cut Inference Cost by 77% in Real Applications ⭐️ 8.0/10

Researchers demonstrate that vocabulary specialization can reduce SLM inference costs by 77%, enabling more efficient deployment of small language models in real-world applications.

rss · Hugging Face Blog · Oct 21, 17:35T1 官方实验室

标签: #Small Language Models, #Inference Optimization, #Vocabulary Specialization, #AI Efficiency, #Hugging Face

ollama/ollama released v0.40.8 ⭐️ 7.0/10

Ollama v0.40.8 adds Flint, a 40B-parameter Mixture-of-Experts coding model with 262K-token context and tool-use support for repository-level AI coding tasks.

github · github-actions[bot] · Oct 21, 16:20

标签: #ollama, #flint, #mixture-of-experts, #long-context, #code-LLM

Claude Duet ⭐️ 7.0/10

Claude Duet launches as a new AI assistant product, with HN commenters debating UX, capabilities, and the underlying inference batching economics that make personal assistant features high-margin for labs.

hackernews · kuberwastaken · Oct 21, 19:14 · 社区讨论T2 社区与开源

标签: #AI, #Anthropic, #inference-batching, #AI-assistants, #product-launch

Meta ordered to pay $25B penalty in New Mexico Cambridge Analytica case ⭐️ 7.0/10

Meta faces a $25 billion penalty from New Mexico over Cambridge Analytica-era data practices, highlighting ongoing state-level enforcement against big tech.

hackernews · cbg0 · Oct 21, 17:45 · 社区讨论T2 社区与开源

标签: #regulatory-penalties, #data-privacy, #big-tech, #cambridge-analytica, #legal-enforcement

Patch the Planet: global response coordination as agents uncover vulnerabilities ⭐️ 7.0/10

OpenAI proposes a coordinated disclosure framework for critical vulnerabilities, emphasizing global response coordination and adaptive fixes for older software releases.

hackernews · OpenAI Blog · Oct 21, 14:03 · 社区讨论T2 社区与开源

标签: #coordinated-disclosure, #vulnerability-management, #ai-security, #software-patching, #open-source

EA is shutting down BioWare ⭐️ 7.0/10

EA announces the shutdown of BioWare, prompting discussion about the studio’s transformation from classic RPG pioneer to its current state and the broader implications of corporate ownership on game development.

hackernews · aaronbrethorst · Oct 21, 17:29 · 社区讨论T2 社区与开源

标签: #bioware, #ea, #gaming-industry, #studio-shutdown

PHP 8.2.35 ⭐️ 7.0/10

PHP 8.2.35 released with critical security fixes addressing RCE and heap UAF vulnerabilities, accompanied by coordinated disclosure involving AI security research.

hackernews · tuyuri · Oct 21, 13:45 · 社区讨论T2 社区与开源

标签: #php, #security, #vulnerability, #rce, #coordinated-disclosure

Using thousands of AI agents to uncover new leads in biological data ⭐️ 7.0/10

OpenAI, PerturbAI, and Scripps Research employed distributed AI agents to uncover research leads from millions of brain cells, demonstrating a novel approach to complex biological data analysis.

rss · OpenAI Blog · Oct 21, 04:50T1 官方实验室

标签: #ai-agents, #biological-data, #machine-learning, #neuroscience, #openai

How we built ChatGPT’s novel ad stack for user trust and relevance ⭐️ 7.0/10

OpenAI developed an LLM-powered ad stack for ChatGPT that interprets conversational intent while keeping advertisements distinct from responses, enabling relevant ad delivery at scale.

rss · OpenAI Blog · Oct 21, 07:00T1 官方实验室

标签: #LLM, #advertising, #ChatGPT, #user trust, #systems design

Democratic governance of frontier AI, part II: Empowering cyber defenders ⭐️ 7.0/10

OpenAI presents a blueprint for leveraging frontier AI to strengthen cyber defenses, improve safeguards, and enhance the resilience of public institutions.

rss · OpenAI Blog · Oct 21, 00:00T1 官方实验室

标签: #AI governance, #cybersecurity, #frontier AI, #AI safety, #public institutions

Southeast Asia’s digital economy to hit $368 billion ⭐️ 7.0/10

Google, Temasek, and Bain & Company’s e-Conomy SEA 2026 report projects Southeast Asia’s digital economy will reach $368 billion by 2026, driven by growth in e-commerce, fintech, and cloud infrastructure.

rss · Google AI Blog · Oct 21, 04:00T1 官方实验室

标签: #digital-economy, #Southeast-Asia, #tech-report, #AI, #e-commerce

Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium. ⭐️ 7.0/10

A practical guide to system-level profiling of TorchNeuron models on AWS Trainium chips, covering tools and techniques for performance analysis on Amazon’s custom ML accelerators.

rss · Hugging Face Blog · Oct 21, 16:04T1 官方实验室

标签: #ML Systems, #Performance Profiling, #AWS Trainium, #PyTorch, #Hardware Acceleration

Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster ⭐️ 7.0/10

IBM releases Granite Speech 5.0, a major update enabling more accurate and faster transcription, translation, and attribution for enterprise speech AI applications.

rss · Hugging Face Blog · Oct 21, 14:58T1 官方实验室

标签: #speech AI, #transcription, #translation, #IBM Granite, #multilingual AI

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 · 2026-10-21 — DeepSeek 全面开源内部推理框架 DeepInfer

    DeepSeek 宣布将内部推理框架 DeepInfer 全面开源，并同步开源生产级 Token 编解码组件 deepseek-recipe。

模型发布/更新

产品发布/更新

行业动态