February 2026
An updated view of AI safety
The second International AI Safety Report reviews general-purpose AI capabilities and risks. A useful companion to model-release announcements.
Discussion: What evidence supports the proposed mitigations?
Home / AI developments
AI developments
What changed.
Why it matters.
Selected milestones from 2024–2026, paired with primary sources. Use this as a guided reading list rather than a live news feed.
January 2026
AI for medical imaging and genomics
MedGemma 1.5 expands medical imaging capabilities, while AlphaGenome’s research publication connects sequence to predicted molecular properties.
Discussion: How are performance and generalization measured?
September 2025
Reasoning becomes a research method to examine
The DeepSeek-R1 paper in Nature studies reinforcement learning for language-model reasoning and distilled models.
Discussion: Which gains come from training, inference compute, or evaluation choices?
2024 onward
From chat interfaces to tool-using systems
Retrieval and agents make it possible to connect language models with external information and actions. This also makes tool permissions and failure recovery central design questions.
Discussion: Could a smaller, simpler workflow solve the same task?
May 2024
Molecular prediction expands with AlphaFold 3
Protein interactions with other molecule types become part of the structure-prediction workflow. Experimental evidence remains necessary to establish biological function.
Discussion: What can a structure tell you, and what can it not?
Keep the context.
Release announcements describe the developer’s results. Read methods, system cards, and independent evaluations before generalizing. For additional sources and a paper-reading checklist, visit Research & publications.