Exercises: Decomposing Tasks for Better Supervision

Resources: Decomposing Tasks for Better Supervision

Exercises: Interpretability 9

Resources: Interpretability 9

Exercises: Track 1: Eliciting Latent Knowledge

Resources: Track 1: Eliciting Latent Knowledge

Exercises: Literature Review or Project Proposal

Resources: Literature Review or Project Proposal

Next steps: Programs

Exercises: Track 3: Science and Deep Learning

Resources: Track 3: Science and Deep Learning

Exercises: Reasoning about Reasoning

Resources: Reasoning about Reasoning

Exercises: Further Understanding the Alignment Problem

Resources: Further Understanding the Alignment Problem

Exercises: Preventing Misgeneralization

Resources: Preventing Misgeneralization

Exercises: Track 2: Agent Foundations

Resources: Track 2: Agent Foundations

Do some of the exercises from [Jacob Hilton’s deep learning curriculum](https://github.com/jacobhilton/deep_learning_curriculum).



Jacob Hilton's deep learning curriculum

Create a hypothetical entry to [Perez et al.’s (2022) inverse scaling prize](https://www.alignmentforum.org/posts/eqxqgFxymP8hXDTt5/announcing-the-inverse-scaling-prize-usd250k-prize-pool) (e.g. using the different model sizes available via the OpenAI API).
Note the competition has closed, but it's still a useful exercise.


Inverse Scaling Prize

Search for alignment failures - cases where the model is _capable_ of doing what you intend, but doesn’t. As one example (discussed in section 7.2 of [Chen et al., 2022](https://arxiv.org/abs/2107.03374)), when the user gives it a prompt containing a subtle bug, the Codex language model may “deliberately” introduce further bugs into the code it writes, in order to match the style of the user prompt.


Alignment failures

Do general [black-box investigation of a large language model](https://www.alignmentforum.org/posts/yGaw4NqRha8hgx5ny/the-case-for-becoming-a-black-box-investigator-of-language) to figure out what novel properties it might have.


Black Box investigation

This session focuses on diving in deep on understanding the science and mechanics of deep learning, with a particular focus on large language models.
### Note on the Exercises
**We recommend doing _one_** of the exercises, as they are all quite in-depth. If you have more time, doing a few of the exercises would likely be helpful, though some of your learning may eventually overlap if you do all of them.


Alignment 201

Exercises: Track 3: Science and Deep Learning

Exercises