Dane Sherburn

About me:

Contractor (Preparedness Team), OpenAI
Contractor (Policy Team), OpenAI

My work focused on designing & creating evaluations in the Policy Team.
These include MakeMeSay, which measures a model's ability to manipulate humans without them knowing, Human-Relative MLAgentBench, which measures a model's ability to do AI R&D, one that wasn't open sourced, and others that I second-authored: Bugged Tools, MakeMePay and Twenty Questions.
I also contribute to and help maintain the OpenAI evals repo, and developed infrastructure for evals, including a web app to visualise logs which has been open sourced.
Some evals I developed were mentioned on the OpenAI blog, and our project, the Dangerous Capability Evaluations project, was mentioned by Sam Altman in Congress.

Machine Learning Alignment Theory Scholar, Stanford Existential Risks Initiative
MEng Maths and Computer Science, Imperial College London

Fun hacky projects:

BitBlaster-16 is a 16-bit computer built from scratch using only NAND gates and data flip-flops as primitives! :) It has its own assembly language, compiler and operating system. I wrote a few programs on it, including an autograd engine to do backpropagation in computational graphs.
Fermi Poker is a fun way to get better at making predictions under uncertainty. Anecdotally I've found friends get *much* better at making well-calibrated predictions after just a few rounds!

Links:

danesherbs[at]gmail[dot]com
github.com/danesherbs

I avoid social media and the news because they veer negative and therefore make you less happy. They also tend to prioritize attention-grabbing content over telling the truth.