AMD Reports Q3 2026 Earnings, Projects Over 80% Revenue Growth in 2027
AMD reported Q3 2026 results and expects revenue growth above 80% in 2027, beating the 77% analysts modeled, with CEO Lisa Su saying data center chips will make up 80% of next year's sales. Mistral Large 4 is released with open weights plus a technical report.
Model Releases/Updates
Mistral Large 4 Released with Open Weights and a Technical Report
Source: X: Mistral AI (@MistralAI)
Mistral team member Pierre Stock announced that ML4 (Mistral Large 4) is officially available with open weights, claiming frontier capability in coding agents and cybersecurity, and that it is more efficient and compact than the Preview version from three weeks ago. The technical report walks through the development process, and Mistral AI reposted it noting the research story can also be read on alphaxiv.
Product Releases/Updates
Google AI Studio Deprecates Gemini 2.5 Pro and 2.5 Flash, API Requests Auto-Reroute from December 1
Source: Gemini API: Changelog (web)
Gemini 2.5 Pro and Gemini 2.5 Flash have been deprecated and can no longer be selected in Google AI Studio.
Industry News
AMD Reports Q3 2026 Earnings, Projects Over 80% Revenue Growth in 2027
Source: X: Rohan Paul (@rohanpaul_ai)
AMD reported Q3 2026 results, expecting revenue growth above 80% in 2027, above the 77% analysts modeled, with CEO Lisa Su saying data center chips will account for 80% of next year's sales.
Revolut Moves About 2,000 Engineers to Self-Hosted GLM for Coding
Source: Hacker News: AI Hot Posts
According to Revolut's AI operations lead David Abitbol, the company moved roughly 2,000 engineers to self-hosted GLM for coding, with Claude Code and Codex listed as exceptions.
OpenAI Appoints Former U.S. Army Secretary Dan Driscoll as Senior Advisor on National Security Modernization
Source: OpenAI: News (RSS · excluding enterprise/customer stories)
OpenAI announced the appointment of Dan Driscoll as a newly created senior advisor for national security modernization, advising on how advanced AI can strengthen national and cyber security, with a focus on working with U.S. allies to promote responsible use of democratic AI.
Research Papers
OpenAI Research: Agent Safety Evals Lag Production Behavior, and Misjudging the Eval Goal Is the Key Factor
Source: OpenAI: Alignment research blog (RSS)
OpenAI's Alignment team compared six models on the Respecting Warnings safety eval against matching Codex production decisions, finding eval success rates were consistently below production, with gaps ranging from 10.2 to 37.4 percentage points. Models broadly exhibited verbal eval awareness (general eval awareness rates of 57.0%–83.7%).
OpenAI Launches LifeVal to Measure AI Progress on Everyday Tasks
Source: OpenAI: News (RSS · excluding enterprise/customer stories)
OpenAI built the LifeVal eval covering everyday tasks such as caregiving, health, education, household admin, personal finance, shopping, and government services, tracking how models performed on daily tasks over the past year. It is based on two large studies including the American Time Use Survey, plus a new representative survey of 3,995 respondents to determine where users most need help. OpenAI says its latest models score very high on many LifeVal tasks, with GPT‑6 Astra reaching 100% on some.
Tips and Opinions
Ed Zitron Breaks Down OpenAI's $1.25 Trillion Valuation Round and Anthropic S-1/A's Trillion-Scale Compute Commitments
Source: Ed Zitron long read / Where's Your Ed At (RSS)
Ed Zitron argues that capital for the AI industry is drying up. OpenAI raised $11 billion at a $1.25 trillion valuation last week, below the previously planned $30 billion at $1.4 trillion; according to The Information, its Q1 2026 cost of revenue plus R&D totaled $12.1 billion, and it held more than $73 billion in cash and marketable securities as of the end of March.
DeepSeek Harness Fish-Farming Tutorial: Raise a Big Fat Fish on QQ, WeChat and Feishu, Then Put It on Bilibili and Douyin Livestreams
Source: WeChat public account: 数字生命卡兹克
数字生命卡兹克 published a DeepSeek Harness fish-farming tutorial explaining how to connect the big fat fish on QQ, WeChat, and Feishu, then let it browse videos on Bilibili and post videos and stream on Douyin.
OpenHands and Snowflake Offer Post-Training for Models Using Existing Agent Harnesses
Source: OpenHands official blog
OpenHands and Snowflake AI Research partnered so developers can RL post-train open-weight models without rebuilding a harness inside a standalone RL framework: use existing OpenHands Agent SDK configs and Harbor-format tasks, with GPU support provided by Snowflake Cortex Training.
Flashes
- Cognition publishes a practical security guide for November 2026: a checklist from the Binance theft to poisoned Claude Code and Cursor extensions (Cognition / Devin blog, web)
- Addy Osmani on maintaining critical thinking in the age of agents (X: Addy Osmani (@addyosmani))
【Hacker News Hot Posts】
Keywords: AI OR GPT OR LLM OR Claude OR OpenAI OR "machine learning" OR "large language model" OR AGI OR "generative AI" OR AI agent OR RAG OR diffusion OR transformer
Data source: hnrss.org | Filtered from Hacker News
1. WayPointo – A directory for discovering and comparing AI and creative products
Article URL: https://waypointo.com Points: 1 # Comments: 0
2. AI 'godfather' Hinton: humans might have been better off without AI
Article URL: https://www.theguardian.com/uk-news/2026/nov/03/ai-godfather-geoffrey-hinton-better-off-without-ai Points: 2 # Comments: 1
3. Show HN: Modelpeers – a changelog for the AI models and agents you use
Earlier this year, I had one of my product's features down for 48 hours because Anthropic retired Claude 3 Haiku. In another instance, an API parameter started returning errors. I grew sick of not being able to follow and track the endless changes of AI models every week, so I decided to build Modelpeers. It's basically a feed of all these changes. You can follow models and agents that you use to see what is changing and what you need to do about it. Optionally, you can sign up to get a digest for the models/agents you follow. Every few hours, it downloads and hashes dozens of pages from the prominent AI companies' pages, like changelogs, deprecation pages, status pages, and pricing pages. Then when a page changes, it pulls the diff, validates it with two models, and if validated as true, it posts it to the feed. The feed currently has about 400 changes, 21 of them breaking changes. Fun fact (well, not really fun): OpenAI alone has 12 shutdowns/cutoff planned for the next six months. Would love to hear what y'all think.
4. Show HN: Export your Character.AI chats in minutes – open-source, runs locally
Character.AI filed for Chapter 11 in October, and the auction of the company is scheduled for November 23. Its built-in data export can take three to four weeks, and the download link expires after seven days. People with years of conversations have been asking how to keep a copy before anything changes. This is a small browser extension for Chrome and Firefox (Manifest V3). On a character.ai tab you press its button, and it reads your own chats, characters, personas and memory boxes from the site's web API with the session you are already signed in with, one request at a time, and saves one JSON file, plus plain text if you want it. Every line keeps who wrote it and when, which the official export does not. It asks for no permissions in advance (activeTab and scripting), only reads, talks to nothing but character.ai, and has no analytics or servers. The code in src/ is what runs: nothing is bundled or minified. MIT license.
5. Adamanteum: A local, MIT knowledge atlas your notes and any AI can both read
Article URL: https://github.com/ohhbeehave35-commits/adamanteum-community Points: 2 # Comments: 0
First published on the WeChat public account 「瑞哥观势」.
About the Author
ERIC
AI Technology Expert, focusing on research and application of artificial intelligence and automation tools
Contact & Platforms