Jamie Bernardi

Here you can find my bio, contact details and links to my work.

Mt Ngauruhoe, Gabriel Johnson

AI Alignment Research, field building, etc

I currently work on a collaboration with researchers at the Future of Humanity Institute to develop safe Machine Learning algorithms. We are developing Bayesian Reinforcement Learning methods using the notion of pessimism. Previous work has shown that, given a safe mentor, an agent can be trained to complete a task without causing catastrophic events and exceed its mentor’s performance. I now investigate these agents empirically. The work is funded by a grant from the Effective Altruism Long Term Future fund; I am very grateful to all the donors that made this possible.

Because I believe aligning AI with human values is a priority in a rapidly advancing field, I lead AI Safety field building strategy at Cambridge Effective Altruism CIC. The flagship AI Safety programme is the global AGI Safety fundamentals programme, which exposes budding researchers to a curriculum of arguments for governance and alignment work, with expert facilitation.

You can also catch me bouldering, playing guitar (preferably in front of other people), and keeping active in various ways! Otherwise, I enjoy spending time in the Effective Altruism community to explore other opportunities to use my time and resources to do good.

What’s that site icon?

I came across the opportunity to turn yourself into a renaissance painting on Jack Clark’s weekly Artificial Intelligence Roundup. Try it out at the AI Gahaku official website!

Work

I’m most excited to get back to the continuation of my Master’s project. I developed a promising theoretical analysis of smoothed labels during the project, and therefore aim to to introduce Bayesian Label Smoothing.

Check out my work on GitHub

Other work on GitHub includes OpenAI Spinning Up (Reinforcement Learning) implementations, Hack Cambridge Hackathon projects, and the Gaussian Processing tools repository contributing to my publication.

ResearchGate Logo

Publications

Multiphysics analysis with CAD-based parametric breeding blanket creation for rapid design iteration

Undergraduate Summer Research Placement, Culham Centre for Fusion Energy, 2017

Resources

A list of resources I’ve made or contributed to and may be useful to others. Includes AGI Safety information, and a time tracking template.

Get in touch

Don’t hesitate to reach out via email with the contact form below, or message me on LinkedIn.

Alternatively you can leave me (optionally anonymous) feedback on my Google Form.