MLAB | Redwood Research

MLAB is our bootcamp aimed at helping people we are interested in AI alignment learn about machine learning, with a focus on ML skills and concepts that are relevant to doing the kinds of alignment research that we think seem most leveraged for reducing AI x-risk.

The first iteration of the bootcamp, MLAB1, was run in January 2022 and the second, MLAB2, in August 2022. We currently have no plans to run another iteration at the moment, however you can request access to the curriculum here and sign up for updates on MLAB below.

What does the program involve?

Course Outline

W0D1 - pre-course exercises on PyTorch and einpos (CPU)
W1D1 - practice PyTorch by building a simple raytracer (CPU)
W1D2 - build your own ResNet (GPU preferred)
W1D3 - build your own backpropagation framework (CPU)
W1D4 - model training
- Part 1: model training and optimizers (CPU)
- Part 2: hyperparameter search (GPU preferred)
W1D5 - cumulants and polysemanticity (CPU)
W1D6 - GPU programming with PyCuda (GPU required)
W2D1 - build your own BERT (CPU)
W2D2 - BERT
- Part 1: data preparation for sentiment analysis (CPU)
- Part 2: fine-tuning BERT on sentiment analysis (CPU)
- Part 3: data preparation for masked language modelling (CPU)
- Part 4: training BERT from scratch (GPU required)
W2D3 - GPT
- Part 1: build your own GPT (CPU)
- Part 2: sampling text from GPT (GPU preferred)
W2D4 - transformer interpretability (CPU)
W2D5 - transformer interpretability on algorithmic tasks (CPU)
W3D1 - distributed training
- Part 1: torch.distributed (GPU required)
- Part 2: data parallelism (GPU preferred)
- Part 3: tensor parallelism (GPU preferred)
W3D2 - intro to RL
- Part 1: multi-armed bandit (CPU)
- Part 2: DQN (CPU)
W3D3 - policy gradients and PPO (CPU)
W3D4 - diffusion models
- Part 1: introduction to diffusion (CPU)
- Part 2: U-Net architecture (GPU preferred)
- Part 3: training a diffusion model on FashionMNIST (GPU required)
W3D5
- Part 1: CLIP (CPU)
- Part 2: image generation with Stable Diffusion (GPU required)

Who is this program aimed for?

The program is aimed at people who are already strong programmers who are comfortable with about one year’s worth of university level applied math (e.g. you should know what eigenvalues and eigenvectors of a matrix are, and you should know basic vector calculus; in this course you’ll have to think about Jacobian matrices and make heavy use of tensor diagram notation, so you should be able to pick up both of those pretty fast).

We expect that about half the attendees will be current students (either undergrad or grad students) and half will be professionals.

In the past we have ended up hiring some of the people who have participated in MLAB.

Our guess is that MLAB is a pretty great opportunity for people who want to become more familiar with the concepts and practical details related to ML; we think that MLAB is a good use of time for many people who don’t plan to do technical alignment research long term but who intend to do theoretical alignment research or work on other things where being knowledgeable about ML techniques is useful.

We accept applicants for participant and TA roles. TAs are expected to either know this material already or have a month free before MLAB to study all the content. TA-ing MLAB is a good opportunity for people with more prior knowledge of this material to connect with Redwood Research and the broader Bay Area alignment community, reinforce their understanding of the curriculum material, and movement-build by teaching others. It also pays competitively.

How does your curriculum differ from the curriculum other people might have built?

It’s more focused on learning by implementing small things based on a carefully constructed curriculum, rather than e.g. reading papers or trying to replicate whole papers at once. This difference in focus is mostly because we believe that focusing first on these skills makes it much faster to learn, because you get faster feedback loops. It’s also partially due to some of our beliefs about how to do ML research, which are slightly unusual among ML people.

What is it like to do MLAB?

Past participants report that MLAB was time-consuming; we strongly recommend against trying to juggle other commitments concurrently. About 8 hours a day, 5 or 6 (if you participate in the optional day) days a week will be spent on pair programming, in addition to daily lectures and readings. There is a lot of content packed into each day; not everyone will finish every part of the curriculum. We aim to create a learning environment that is focused but not frantic; we’d rather have you understand the material deeply than finish 100% of the day’s content.

When we asked participants what they were surprised by, major themes were:

People thought it was more useful than they expected.
- Many people were surprised by how much they liked the focus on implementing things.
- Several people were concerned that their ML background was too strong or too weak, and were pleasantly surprised by the extent to which the content was valuable anyway.
People were surprised that there wasn’t more content on alignment in the main curriculum. The content is basically all related to understanding and working with ML systems. This focus is because we think that doing alignment research with ML systems basically just requires the same skills as understanding these systems more generally, and so you might as well just study the systems directly and apply this knowledge to alignment projects later (except that we think that emphasizing interpretability throughout is actually a pretty good way of learning to understand the systems better in a way that we expect to be useful for a variety of directions of research). There were a variety of Q&As with alignment researchers and quite a lot of casual discussion about alignment.

Sign up to ourmailing list

Sign up to our
mailing list