Apply to the ML for Alignment Bootcamp (MLAB) in Berkeley [Jan 3 - Jan 22]

by Habryka, Buck2 min read3rd Nov 20217 comments

139

Research training programsRedwood ResearchLightcone InfrastructureAI alignment
Frontpage

We (Redwood Research and Lightcone Infrastructure) are organizing a bootcamp to bring people interested in AI Alignment up-to-speed with the state of modern ML engineering. We expect to invite about 20 technically talented effective altruists for three weeks of intense learning to Berkeley, taught by engineers working at AI Alignment organizations. The curriculum is designed by Buck Shlegeris (Redwood) and Ned Ruggeri (App Academy Co-founder). We will cover all expenses. 

We aim to have a mixture of students, young professionals, and people who already have a professional track record in AI Alignment or EA, but want to brush up on their Machine Learning skills.

Dates are Jan 3 2022  - Jan 22 2022. Application deadline is November 15th. We will make application decisions on a rolling basis, but will aim to get back to everyone by November 22nd.

Apply here

AI-Generated image (VQGAN+CLIP) for prompt: "Machine Learning Engineering by Alex Hillkurtz", "aquarelle", "Tools", "Graphic Cards", "trending on artstation", "green on white color palette"

The curriculum is still in flux, but this list might give you a sense of the kinds of things we expect to cover (it’s fine if you don’t know all these terms):

  • Week 1: PyTorch — learn the primitives of one of the most popular ML frameworks, use them to reimplement common neural net architecture primitives, optimization algorithms, and data parallelism
  • Week 2: Implementing transformers  — reconstruct GPT2, BERT from scratch, play around with the sub-components and associated algorithms (eg nucleus sampling) to better understand them
  • Week 3: Training transformers — set up a scalable training environment for running experiments, train transformers on various downstream tasks, implement diagnostics, analyze your experiments
  • (Optional) Week 4: Capstone projects

We’re aware that people start school/other commitments at various points in January, and so are flexible about you attending whatever prefix of the bootcamp works for you. 

Logistics

The bootcamp takes place at Constellation, a shared office space in Berkeley for people working on long-termist projects. People from the following organizations often work from the space: MIRI, Redwood Research, Open Philanthropy, Lightcone Infrastructure, Paul Christiano’s Alignment Research Center and more.

As a participant, you’d attend communal lunches and events at Constellation and have a great opportunity to make friends and connections.

If you join the bootcamp, we’ll provide: 

  • Free travel to Berkeley, for both US and international applications
  • Free housing
  • Food
  • Plug-and-play, pre-configured desktop computer with an ML environment for use throughout the bootcamp

You can find a full FAQ and more details in this Google Doc.

Apply here

139

7 comments, sorted by Highlighting new comments since Today at 2:25 PM
New Comment

Have you thought of recording the sessions and putting them online afterwards? I'd be interested in watching, but couldn't apply (on a honeymoon in Tahoe, which is close enough to Berkeley, but I imagine my partner would kill me if I went missing each day to attend an ML bootcamp). 

Not addressing video recordings specifically; but we might run future iterations of this bootcamp if there's enough interest, it goes well and it continues to seem valuable. So feel free to submit the application form while noting you're only interested in future cohorts. 

Should I reapply if I already filled in the interest form earlier? I notice that the application form is slightly updated.

No, the previous application will work fine. Thanks for applying :)

Is there any sort of confirmation email sent after submitting the application? I've just submitted one, and didn't receive anything via email. Thanks!

Sorry, no confirmation email currently! Feel free to send me a PM with your real name, and I can confirm that your application went through (though if you saw the "Thank you" screen, I would be quite surprised if your application got lost)

[+][comment deleted]1mo 1