🤖 AI HOT Daily · Nov 4, 2026

AMD Q3 earnings project >80% revenue growth in 2027 with data center chips at 80% of next year's sales; Mistral Large 4 releases open weights with tech report; Google AI Studio deprecates Gemini 2.5 Pro/Flash, API requests auto-route from Dec 1; Revolut migrates ~2,000 engineers to self-hosted GLM coding; OpenAI appoints former US Army Secretary as senior advisor for national security modernization; OpenAI research finds gap between agent safety evals and production behavior; OpenAI launches LifeVal benchmark tracking AI on everyday tasks; Ed Zitron analyzes AI industry funding running dry; DeepSeek Harness fish-farming tutorial published; OpenHands partners with Snowflake on model post-training.

1. 1. AMD Projects >80% Revenue Growth in 2027

   AMD reports Q3 2026 earnings, projecting >80% revenue growth in 2027, above the 77% analysts modeled. CEO Lisa Su says data center chips will be 80% of next year's sales.

2. 2. Mistral Large 4 Releases Open Weights with Tech Report

   Mistral announces ML4 (Mistral Large 4) open weights, with frontier capabilities in code agents and cybersecurity, more efficient and compact than the Preview version from three weeks ago.

3. 3. Google AI Studio Deprecates Gemini 2.5 Pro/Flash

   Gemini 2.5 Pro and 2.5 Flash deprecated, no longer selectable in Google AI Studio. API requests auto-route to newer models from Dec 1.

4. 4. Revolut Migrates ~2,000 Engineers to Self-Hosted GLM Coding

   Per Revolut's AI ops lead, ~2,000 engineers move to self-hosted GLM for coding, with Claude Code and Codex listed as exceptions.

5. 5. OpenAI Appoints Former US Army Secretary as National Security Advisor

   OpenAI appoints Dan Driscoll as new senior advisor for national security modernization, advising on how advanced AI strengthens national and cyber security, focusing on responsible democratic AI use with allies.

6. 6. OpenAI Research: Gap Between Agent Safety Evals and Production Behavior

   Comparing six models on Respecting Warnings safety evals vs matched Codex production decisions, eval success rates are consistently lower than production, with a gap of 10.2 to 37.4 percentage points.

7. 7. OpenAI Launches LifeVal Benchmark for Everyday Tasks

   LifeVal covers caregiving, health, education, household, personal finance, shopping, and government services. Latest models score high on many tasks; GPT-6 Astra hits 100% on some.

8. 8. Ed Zitron Analyzes AI Industry Funding Running Dry

   Ed Zitron writes AI industry funding is running dry. OpenAI raised $11B at $1.25T valuation last week, below the planned $30B at $1.4T.

9. 9. DeepSeek Harness Fish-Farming Tutorial Published

   Kazik publishes a DeepSeek Harness fish-farming tutorial: how to integrate the agent into QQ, WeChat, Feishu, and have it browse Bilibili, post videos, and stream on Douyin.

10. 10. OpenHands Partners with Snowflake on Model Post-Training

    Developers can do RL post-training on open-weight models using existing OpenHands Agent SDK config and Harbor-format tasks without rebuilding the harness in a separate RL framework.