These lessons will include:
- An Inverse Reinforcement Learning (IRL) track, sourced from a set of papers that we have distilled by our own efforts.
- A track on Iterated Distillation and Amplification (IDA), largely adapted from Paul Christiano’s LessWrong sequence on the matter.
A lesson on the former will be published on our platform every Monday, and a batch of lessons on the latter will be published every Thursday. The former will be finished in approximately 10 weeks, the latter in 4. If you plan to take these lessons, we recommend scheduling one full day per week for each track you’re following. Once published, the lessons will be available indefinitely for self-paced study.
In the meantime, we continue to semi-regularly publish textbook guides for our prerequisites track. These guides should eventually cover all the prerequisite material one should understand in order to properly follow work in AI Safety.
In the coming months, our focus will be on testing and refining our MVP. As long as we operate, we hope to make learning AI Safety less and less costly, creating common knowledge and paying our research debt.
To do this testing properly, we are seeking individuals that would like to join our testing panel. This means that, as you go through our lessons, we will guide you through the lessons and interview you in-depth about your experience, most importantly about where you’re most confused. If you would like to join us, please fill in this form.
This work has been sponsored, among others, by the EA Hotel. If you like our work, consider donating there.