Measured, Not Predicted
Why AI-for-science systems need measurement-first recovery loops when prediction loses contact with the experiment.
Blog Topics
AI4Science | Agents | Physics... And more
Why Low-Sample, High-Quality Learning Is the Key to Real AI-for-Science
A case for explicit concept-building layers so scientific agents can actually learn portable structure from a few high-value examples.
Memory Compression That Preserves Action
A lab memo on shrinking long-running agent traces without erasing the state that still governs the next safe move.
Exploring the Automation Solution in Experimental Science in 2025 - This Is What I Learned
A year-end reflection on moving from rigid workflow scripts toward an observing agent that has to survive real laboratory state.