Neev Parikh

I like to make computers do interesting things, deeply understand concepts and build interesting, useful tools. I’m currently thinking about AI alignment, control, and evaluations, and work with frontier models at METR.

I’ve worked at Stripe and CSM in the past. I went to Brown for a concurrent BSc/MSc in Computer Science.

Here’s some research that I’m especially proud of:

• RE-Bench: Frontier AI R&D evaluations (ICML spotlight)
• Learning Markov State Abstractions for Deep Reinforcement Learning: Markovian abstractions for deep RL

Here’s some fun projects I did:

• hierarchical RL on Doom: Extending async-PPO implementation to learn options (higher-level skills/actions, learn more)
• volumetric path tracing: Extending rpt to use volumetric path tracing and photon mapping. Based on Bitterli et. al. and bonus class presentation.
• spaceport: Tiny MacOS utility written in Rust (with Obj-C bindings) to partly replace the deprecated airport utility.

Here are some useful links: github, twitter, linkedin, google scholar, email.