Company
Small surface area, high scrutiny.
Modularium Research studies AI systems through evaluation, alignment, and interpretability, with a bias toward claims another researcher can inspect.
01
Pressure before polish.
Evaluation setups should preserve the conditions that make behavior appear, not remove them for a cleaner chart.
02
Evidence before theater.
Each project starts with a behavioral claim, makes the test visible, and leaves behind artifacts that can be challenged.
03
Narrow beats vague.
The lab chooses smaller questions when that makes the measurement, limitation, and mechanism easier to reason about.
Contact
hello@modulariumresearch.com