General
Research Insights
- ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory.
- Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models. The agent can evolve the context itself, to improve longer-range task coherence.
- Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity (blog, github). Basically, you directly tell the LLM to “sample from the full distribution” and this actually increases creativity.
- Agent Learning via Early Experience. Emphasizes unguided exploration to gain knowledge/capability.
LLM
- Anthropic Claude Haiku 4.5.
Video
- Runway Apps, AI/ML modules for video (examples).
- Google Veo 3.1 released.
Science
- Google: How a Gemma model helped discover a new potential cancer therapy pathway. The model was used to generate novel testable hypotheses, and one of them appears like a genuinely new and promising cancer treatment.
- Paper: Scaling Large Language Models for Next-Generation Single-Cell Analysis. Cell2Sentence-Scale 27B (C2S-Scale) is a 27 billion parameter open model that translates gene expression data into “sentences” that LLMs can understand.
Hardware