Raghav Nautiyal

EECS @ Berkeley

[Raghav Nautiyal @rgv_ntyl]

@EzraJNewman I also think it would be cool to investigate eval-awareness work in scheming, eg. testing whether a scheming agent actively investigates how much oversight it’s under, then changes its behavior accordingly. Would also be excited to explore questions like this in the project.

[Raghav Nautiyal @rgv_ntyl]

@EzraJNewman Hey @EzraJNewman, applied to your project a couple weeks ago but only now saw this tweet.

I just finished up an Applied Science internship at Microsoft where a lot of my work was builidng evals designed to surface specific LLM failure modes in retrieval.