Top pick: books from Periodic Labs' co-founders

Liam Fedus and Dogus Cubuk co-found and co-lead Periodic Labs, a startup building "synthesis superintelligence." At the end of their Generalist episode, each recommended one book The Generalist:

• The Beginning of Infinity by David Deutsch, recommended by Fedus. Asked what book he'd give everyone on Earth, he admitted it's "probably a popular one in Silicon Valley." He said it has "a spirit of optimism" and gets at "the unboundedness of science" The Generalist.
• Subtle Is the Lord by Abraham Pais, recommended by Cubuk. It's a biography of Einstein by a friend of his who was also a physicist. Cubuk said most biographers aren't experts in their subject's field, but this one covers "very precise technical detail about what he actually did." He called it both "an incredible friendship book" and "an incredible physics book" The Generalist.

Why it's the top pick: both founders gave specific reasons for their choices. Cubuk's point about the biographer's technical expertise also connects to how Periodic Labs thinks about science: Fedus says AI has mostly been trained on "the final artifacts of science" rather than on how the science actually unfolded The Generalist.

Bill Gates's three books

At the end of his Ezra Klein Show interview, Gates named three books. Gates shared the episode himself, calling AI either "the greatest tool for opportunity and equality we've ever seen—or the cause of immense, unprecedented harm" @BillGates. The titles and authors below come from the transcript, and some names may be garbled:

• The Correspondent by Virginia Evans. He "literally binged" this novel over the weekend and called it "very touching" and "very upbeat" The Ezra Klein Show.
• Into the Wood Chipper by "Nicholas Enrich." Gates described it as being about how USAID got put "into the wood chipper" The Ezra Klein Show.
• The Infinity Machine by "Sebastian Malib." He called it an AI book and a history of Demis Hassabis (transcribed as "Dean Deusibus") and DeepMind. He said it gives "a sense of the whole founding of the AI industry" and does "an incredible job" The Ezra Klein Show.

Essays and articles

• "Personal AI Should Actually Be Personal" by Ryan Sarver, recommended by Garry Tan. Sarver starts from Amazon blocking Meta's Muse agent from its retail site. He argues that an agent hosted by a platform "has two principals," and that one option should be one you own: open source, with your data in storage you control @rsarver. Tan's takeaway: personal AI is "in its Homebrew computer club era," and truly personal AI means you "own your own skills and memory" @garrytan.
• "Free the models: harness design at the frontier", from Replit's AI team, recommended by Amjad Masad. The authors are Daniel Furman, Jacky Zhao, Vaibhav Kumar, Ed Sioufi and Michele Catasta @pirroh. Masad calls it a guide to building a harness that reaches "frontier performance at a fraction of the cost" @amasad. The post argues that the main model should pick its subagents' tier and effort itself, rather than a router making that choice. It frames this as an instance of Sutton's bitter lesson: "the less the harness should decide for them" @pirroh. Masad runs Replit, so this is promotion of his own team's work.
• Richard Hanania on which experts to trust, recommended by Bill Gurley by way of @dawallach. Gurley says "everyone watching" the AI-doomer debate "should read about Tetlock's work." His point is that generalists ("foxes") predict better than specialists ("hedgehogs"), so researchers' "we know more" argument "works against you" @bgurley +1.

Also noted

• Patrick O'Shaughnessy's interview with Anthropic CFO Krishna Rao (YouTube). O'Shaughnessy called it "worth revisiting today" @patrick_oshag. Topics include how Rao allocates compute across Trainium, TPUs and GPUs, what investors misunderstand about model companies, and why returns to frontier intelligence keep rising @patrick_oshag. O'Shaughnessy is the host, so this is self-promotion.