Domain Examples

How do initially similar embryonic cells arrive at stable, distinct fates (e.g., neural vs mesoderm)? We want a mechanism that yields discriminative predictions under interventions (dose, timing, contact), not just a story that fits existing gradients or marker patterns.

Why do LLMs sometimes output fluent but false claims, and which interventions reduce this without killing usefulness? We want hypotheses that separate objective/grounding limits from alignment incentives and from evaluation artifacts.

Why do some online communities spiral into harassment and chronic conflict while others remain constructive? We want quasi-experimental tests that discriminate enforcement failure, selection/composition, and platform amplification mechanisms.