# Jarrod Barnes

> Researcher and founder working on autonomous scientific discovery.

I run Dynamical Systems, a research lab that builds the environments, evaluations, and verification systems that make the scientific loop run end to end. Focus areas: RL environments, verification infrastructure, post-training systems, evaluation methodology, calibrated decision-making for scientific workflows.

## Links

- Site: https://jbarnes850.github.io
- Lab: https://dynamicalsystems.ai
- GitHub: https://github.com/jbarnes850
- LinkedIn: https://linkedin.com/in/jarrodbarnes
- Email: jbarnes850@gmail.com

## Selected Work

- Test-Time Verification for Novel Materials: https://dynamicalsystems.ai/blog/scaling-test-time-verification
- Training Scientific Judgment: https://dynamicalsystems.ai/blog/training-scientific-judgment
- Metacognition Benchmark: https://jbarnes850.github.io/2026/03/20/do-models-know-when-to-change-their-mind/

## Publications

- Surprisal-Guided Selection (arXiv 2026): https://arxiv.org/abs/2602.07670
- OpenSec (arXiv 2026): https://arxiv.org/abs/2601.21083
- Continual Learning for Agents (arXiv 2025): https://arxiv.org/abs/2511.01093

## Writing

- Surprisal-Guided Selection: https://jbarnes850.github.io/2026/02/02/surprisal-guided-selection/
- Frontier Security Agents Lack Restraint: https://jbarnes850.github.io/2026/01/23/frontier-security-agents-lack-restraint/
- When Sampling Beats Training: https://jbarnes850.github.io/2026/01/15/when-sampling-beats-training/
- Rethinking Evaluation for Agents: https://jbarnes850.github.io/2026/01/13/rethinking-evaluation/

## Full Context

- llms-full.txt: https://jbarnes850.github.io/llms-full.txt