General
- Apple release Embedding Atlas, for visualizing text data (demo).
- Anthropic is facilitating access to their models for federal workers.
Research Insights
- Interesting approach to evaluate one aspect of LLM knowledge: query it repeatedly over longitude/latitude values so that it builds up a map of the Earth. How Does A Blind Model See The Earth? A tiny LLM eval with pretty pictures.
LLM
- OpenAI’s unreleased model that previously scored Gold at IMO and second-place in AtCoder, has now also achieved Gold at IOI.
Image Synthesis
- Google releases Imagen 4.
Video
- Pika have developed a video generation model that uses an audio input for performance control.
- SkyReels A3 is audio-conditioned video generation (examples).
World Synthesis
- Tencent have developed: Yan – Foundational Interactive Video Generation. It seems not as coherent as Genie 3, but shows that many groups are making progress in this area.
- Skywork announces Matrix-Game 2.0: An Open-Source, Real-Time, and Streaming Interactive World Model.
Science
- Google: How AI is helping advance the science of bioacoustics to save endangered species.
- Meta: TRIBE: TRImodal Brain Encoder for whole-brain fMRI response prediction. Advanced brain modeling, won 1st place at Algonauts 2025 brain modeling competition.
- Compositional Flows for 3D Molecule and Synthesis Pathway Co-design.
- A personal health large language model for sleep and fitness coaching.
Robots
- Fourier GR-3 is a humanoid intended for human care.
- Figure humanoid (using Helix model) can fold laundry.