General
- Financial Times: The power crunch threatening America’s AI ambitions.
Research Insights
- A new method, Poetiq, reaches new record (54%) on ARC-AGI-2. The method layers a “meta system” over generic LLMs, optimizing their usage to the task.
LLM
- OpenAI releases new benchmark, FrontierScience: Evaluating AI’s ability to perform scientific research tasks (paper).
- And testing on real-world scenarios: Measuring AI’s capability to accelerate biological research in the wet lab.
- Google unveils Gemini 3 Flash; a very fast and very good (sometimes better than Gemini 3 Pro!) model.
AI Agents
- Google DeepMind: Towards a Science of Scaling Agent Systems. If individual agent performance is too low, multi-agent systems tend to do even worse. Independent agents lead to error amplification, while central coordination can reduce this effect.
Image Synthesis
- OpenAI introduces a new and improved ChatGPT Images.
Audio
Video
- ByteDance unveils Seedance 1.5, with audio.