Elemental Research
Elemental · World Model Interpretability

Open questions.

Five threads in world model interpretability.

Titles are working names. Full drafts available on request.
01
Working paper

Right for the Right Reasons

Separating genuine predictability from silent collapse in a world model's JEPA surprise signals.

Request access ↗
02
Draft

Twin Worlds

Making the counterfactual difference between paired worlds the primary training objective for world models.

Request access ↗
03
Working paper

Competence, Not Structure

Evidence that systematic generalization tracks competence rather than clean factorization of representations.

Request access ↗
04
Draft

When Energy Decomposes

An anchored, causally-validated account of per-constraint structure in energy-based models, and why it governs reasoning.

Request access ↗
05
Working paper

How Deep It Goes

Measuring recursive belief depth in language models: how many levels of "I think that you think" they hold, and where it stops.

Request access ↗

Everything here is open on request.

Email the maintainer for drafts, data, or to collaborate.

bhavithchandra16@gmail.com ↗