About Me
Hello, I'm Sonny George. I am currently finishing my Master's at Brandeis University and am seeking research and other opportunities in the space of action-taking AI.
Much of future impactful AI will require some form of active embodiment, reflection, and continual learning. In contrast, the “intelligence” of today's AI systems tends to be static and task-specific.
I am excited by the development of AI systems that:
- Plan and act in uncertain environments
- Adapt to novel tasks/situations with inference-time-feasible efficiency and without catastrophic forgetting/interference
- Are sufficiently interpretable and manipulable to be useful and aligned with human preferences
For such systems, many bottlenecks remain: perception (visual, sensorimotor, etc.), manipulation, planning with world-dynamics models, open-ended learning algorithms (intrinsic curiosity objectives), interpretability, and more.
Publications
Probing the Capacity of Language Model Agents to Operationalize Disparate Experiential Context Despite Distraction
Sonny George, Chris Sypherd, and Dylan Cashman
Findings of the 2024 Conference on Empirical Methods in Natural Language Processing
Thesis Proposal: Towards Any-Horizon Autonomy via Closed-Loop Natural Language HTN Planning
Sonny George
Under Review, NAACL Student Research Workshop 2025
Blog Posts
2025-10-01: What is the proper level of I/O abstraction in world models (that generate simulated rollouts)?
A Few (humble) Hardware Projects
BC+PPO Learning for Simple Object Picking

Routine Butler

Vision-Based Line-Following Robot With Lidar Obstacle Avoidance
