I put together this post for the purpose of sharing interesting resources with friends and colleagues who want to learn more about advancements in the AI space. It's not meant to be comprehensive, but instead a branching out point for folks noticing advances at OpenAI, DeepMind, and open source communities.
Note: many of the resources I chose to share have an explicit alignment/safety orientation, for a few reasons.
- I believe that many alignment thinkers present holistic visions of what the near- and intermediate-term landscape looks like for AI. While I disagree with some opinions, these individuals and organizations provide comprehensive, generalized explanations while addressing misconceptions for those unfamiliar with the space.
- Many of the cutting-edge AI capabilities are being developed by the same organizations attempting to conduct alignment/safety research.
- Philosophically, one should know potential harms of a technology before developing it further.
Cover Art: a set of images generated by me using the Latent Diffusion LAION-400M model.
- Wait But Why - featuring humorous cartoons and ELI5-style explanations
- OpenAI Blog Highlights
- DeepMind Blog
- No time like the present for AI safety work by Scott Alexander
- Ethical Issues in Advanced Artificial Intelligence by Nick Bostrom
Technical Overviews (still fairly introductory)
- Artificial General Intelligence - A Gentle Introduction by Pei Weng
- Andrew Ng's ML course is frequently recommended, although I have not done it
- The Malicious Use of AI: Forecasting, Prevention and Mitigation by Brundage, Alvin et al
- Anything written by Paul Christiano
- Superintelligence: Paths, Dangers, Strategies by Nick Bostrom arguably established (or at least popularized) the field of AI alignment research. Highly recommended.
- Artificial Intelligence: A Modern Approach by Stuart Russell and Peter Norvig is the most commonly used textbook for college courses on the subject
- The Precipice: Existential Risk and the Future of Humanity by Toby Ord
- Human Compatible: AI and the Problem of Control by Stuart Russell
- The Hundred-Page Machine Learning Book by Andriy Burkov for a concise overview of ML techniques
- https://multimodal.art/ is an excellent overview of the ecosystem, with a weekly newsletter conveying the latest advancements in the space. I highly recommend playing around with some of the publicly available Google Colab notebooks linked on the site, for example Latent Diffusion LAION-400M.
- DALL-E 2 and Imagen are the two leading models at the present, although I expect open-source versions to catch up in 3-6 months.
- The aiaiart repository on GitHub is an interesting set of lessons, though it may be outdated quite quickly.
- Interesting Twitter follows:
- 80,000 hours - technical research and policy