# Richard Sutton on AI Discovery, and Benedict Evans’s Case for Wider Reading

*By Recommended Reading from Tech Founders • June 1, 2026*

Marc Andreessen’s strongest recommendation today is Richard Sutton’s argument that supervised generative AI cannot achieve true discovery on its own. Benedict Evans adds two book picks that reinforce a second theme: read more widely, and outside familiar genres.

## Most compelling recommendation

### Richard Sutton’s video/speech on AI creativity and discovery [^1]
- **Content type:** Video/speech
- **Author/creator:** Richard Sutton
- **Link/URL:** https://youtu.be/K5LAFEjTlBA [^2]
- **Who recommended it:** Marc Andreessen, who called Sutton "a genius and a legend" [^1]
- **Key takeaway:** Sutton argues that generative AI trained by supervised learning can be highly useful, but cannot make true novel discoveries on its own. The missing ingredient, in his framing, is discovery through trying many things, testing what works, and keeping the best results—the logic behind reinforcement learning and other generate-and-test systems [^2]
- **Why it matters:** This was the clearest high-signal pick in the set because the recommendation came with a specific framework for separating useful mimicry from systems that can actually discover new things [^1][^2]

> "In this video, I explain the sense in which generative AI trained by supervised learning is incapable of making novel discoveries." [^2]

## Benedict Evans’s broader-reading picks

> "Read books. Read different books generally. Read books for grown-ups. Please read something other than Lord of the Rings." [^3]

### *Three Men in a Boat* — Jerome K. Jerome [^3]
- **Content type:** Book
- **Link/URL:** Context episode: https://www.youtube.com/watch?v=BD3vLtWhT5A
- **Who recommended it:** Benedict Evans on Lenny's Podcast [^3]
- **Key takeaway:** Evans named it as one of the books he recommends most and described it as a classic British comedy with sections that keep feeling useful in everyday situations [^3]
- **Why it matters:** It shows the kind of reading Evans thinks stays valuable across very different contexts [^3]

### *Nature's Metropolis* — William Cronon [^3]
- **Content type:** Book
- **Link/URL:** Context episode: https://www.youtube.com/watch?v=BD3vLtWhT5A
- **Who recommended it:** Benedict Evans on Lenny's Podcast [^3]
- **Key takeaway:** Evans said this economic history of Chicago is highly relevant to technology because it covers standardization, packetization, logistics, channel conflict, network dynamics, and network neutrality [^3]
- **Why it matters:** It turns economic history into a lens for understanding modern tech systems, which is exactly the broader-reading habit Evans is advocating [^3]

## Pattern from today’s authentic picks

The strongest recommendations today came with reasons, not just titles. Andreessen pointed readers to Sutton for a specific argument about the limits of supervised generative AI, while Evans used two very different books to argue for reading more widely and about subjects you do not already know [^2][^3]

---

### Sources

[^1]: [𝕏 post by @pmarca](https://x.com/pmarca/status/2061233935322091841)
[^2]: [𝕏 post by @RichardSSutton](https://x.com/RichardSSutton/status/2061216087744946656)
[^3]: [A rational conversation on where AI is actually going | Benedict Evans](https://www.youtube.com/watch?v=BD3vLtWhT5A)