The Sequence Radar - Issue 948: Last Week in AI: Billions in Funding, Hundreds of Papers, One Big Leap?
Reflection and Mistral expand the model race, TypeSafe and Arena raise the stakes, and OpenAI puts mathematical discovery under scrutiny.
TL;DR
- Reflection introduced Beam, a 501 billion parameter mixture-of-experts model focused on reinforcement learning and inference efficiency.
- Mistral released Large 4 (Le Chonk), a trillion-parameter multimodal model emphasizing European infrastructure and sovereignty.
- OpenAI published over 700 AI-generated math manuscripts, raising questions about verification and the acceleration of research.
- TypeSafe secured $870 million in Series A funding for its Jev model, which produces typed decisions with confidence estimates.
- Arena raised $200 million in Series B funding and introduced an Alignment Index to evaluate model behaviors beyond task completion.
- The AI industry is seen as separating intelligence into distinct economic problems: capability production, efficient delivery, control retention, and behavior verification.