Worth watching
Shows how domain‑specific foundation models outperform generic ones and how to build them for real products
AI news · Monday, 2 November 2026
AI news, 2 November 2026
Anthropic blocks Claude game reverse‑engineering, OpenAI merges Chat and Work token quotas, and Google’s StreamVibe video agent fails benchmark timing.
29 stories, each labelled: available now, announced, unconfirmed or research. Get them ranked and explained for your product →
Paper of the week· 26 Oct – 1 Nov
Work Agents: Systems, Capabilities, and Optimization
The framework provides a way to think about designing agents that aim to reliably complete multi‑tool digital work.
- Define work dimensions
- Classify work forms
- Identify capability families
- Optimize model and harness
- Evaluate outcomes
Top stories
Announced
OpenAI merges Chat and Work modes, unifies token quota across plans
OpenAI will combine Chat and Work into a single interface and, beginning in December, Plus, Pro and Business plans will share a common token‑based quota for ChatGPT and Codex. More complex models use more tokens, while simple instant chats are exempt, which may reduce quality for heavy users on lower‑cost plans.
The Decoder
Research
Google unveils StreamVibe video agent, but benchmark shows models miss help moments
Google’s research team built StreamVibe, a video agent that watches people cook or assemble furniture and speaks up only when help is needed. A benchmark of about 4,100 examples from more than 183 hours shows none of the tested models reliably pick the right moment in longer videos, with errors compounding and response quality dropping sharply. Gemini 2.5 Flash suggested timing and phrasing, approved by four human evaluators.
The Decoder
Meta's ad revenue growth outpaces Google search, widest gap since 2021
The gap between Meta's ad revenue growth and Google's search expansion was the widest it's been since 2021 in the latest quarter.
CNBC
Cisco Talos Finds Volta-Autodoxer AI‑Powered Infostealer
The newly uncovered malware "Volta-Autodoxer" uses AI tools to harvest credentials, credit‑card numbers, Discord sessions and UberEats receipts, and Cisco Talos flagged it with its CAIRN framework; infostealers have stolen hundreds of millions of credentials in recent years.
Wired
Available now
Nace AI Open-Sources Drex Max v1, a 27B Decision Model for Apple Silicon
Nace AI released Drex Max v1, a 27‑billion‑parameter multimodal decision model that scores 67.52 and runs locally on Apple Silicon devices.
MarkTechPost
Frontier labs
Agents & dev tools
Open source
Available now
Pika‑Labs Releases Pika‑H3‑Flex Image‑to‑Video Model, Trending on Hugging Face
Hugging Face
Research
Research
Luxembourg researchers find warm prompts trigger anxiety narratives in ChatGPT, Grok, Gemini
The Decoder
Research
Vals AI study shows Chinese models’ political alignment varies more than expected
The Decoder
Research
Researchers Release Co-Design Bench for Multimodal Coding Agent Evaluation
Hugging Face Papers