Atlasof World Models
Start

Research

How machines learn to predict, and how we trust them.

Four directions I am working on now. All of my papers are under review.

Current directions

Four questions

  • World models for enterprise agents that plan under delayed and asynchronous data.
  • Joint embedding predictive architectures with SIGReg on smart-meter interval data.
  • Stress testing event-driven agents to find failures that are rare but costly.
  • Governing what an AI agent is allowed to do before it acts.

The atlas is where I study the first two in public. Start with latent world models.

Research interests

What I study

Some of these are active directions. They are not areas with results to show yet.

Core

  • World Models
  • JEPA
  • Representation Learning
  • Reinforcement Learning
  • Agentic AI
  • AI Safety and Evaluation

Also

  • Predictive State Modeling
  • Model-Based Planning
  • Adaptive Decision-Making
  • AI Agent Memory
  • Human-AI Collaboration
  • Hybrid Neural Architectures
  • Graph-Based Reasoning
  • AI Governance
  • Enterprise AI Systems
  • AI for Energy and Utilities
  • Adaptive Stress Testing

Application areas

  • Energy and gas utilities
  • Demand response
  • Wildfire risk
  • Customer billing
  • Document intelligence
  • Enterprise workflow automation

Papers

All under review.

  • A paper on asynchronous world models for utility operations

    Under review

  • A paper on proposal starvation in stress testing of event-driven agents

    Under review

  • A paper comparing planners and learners in synthetic bill assistance

    Under review

I list the topic only. Titles and venues wait until review ends.

Books

In progress

  • Agents with Foresight A book in progress

    Writing

  • The World Outside the Model A book in progress

    Writing

The reading behind the atlas

Search the atlas