CoRecursive: Coding Stories

The episode delivers a coherent historical arc of reinforcement learning - from temporal difference learning through AlphaGo to the Bitter Lesson - but the insight-per-minute ratio is dragged down heavily by filler affirmations, tangential asides (pigeon missile guidance, cold fusion comparisons), and a guest who contributes little substance. The closing synthesis about benchmarks creating reward signals AI will dominate is the only genuinely non-obvious moment for a B2B operator.

“The dark version of reward is enough, is that reward is enough for a computer to crush you at what you do.”

“The biggest lesson that we can read from 70 years of AI research is that general methods that leverage computation are ultimately the most effective.”

The episode is largely a historical retelling of well-documented events (TD-Gammon, Deep Blue, AlphaGo, the Bitter Lesson essay) without a fresh analytical lens. The framing as a 'stack trace' is mildly creative pedagogically, but the host is synthesising received wisdom rather than generating original argument. The connection to SWE-bench and commoditised benchmarks at the end is the only moment of independent synthesis.

“We have to learn the bitter lesson, that building and how we think we think does not work in the long run”

“Large language models are about mimicking people, doing what people say you should do. They're not about figuring out what to do.”

Don is explicitly a layperson acting as a sounding board - he admits not knowing B.F. Skinner, can't estimate backgammon game states, and contributes no domain expertise or practitioner experience. There is no credentialed guest; the host is a self-described software developer doing a solo explainer dressed as a dialogue.

“B: I do not. [re: B.F. Skinner]”

“B: I have no idea. [re: backgammon game states]”

The episode cites several real figures - 8,000 Deep Blue evaluation rules, 200 million positions per second, 1-in-10,000 human probability for move 37, 2,315 Wordle words, AlphaGo winning 4-1, AlphaGo Zero training in three days - and includes verbatim quotes from Sutton, David Silver, and Demis Hassabis. These are largely historical trivia rather than business-applicable metrics, but they do ground the narrative concretely.

“AlphaGo said that there was a 1 in 10,000 probability that a human would have played that move”

“it would play 200 million positions per second”

The host structures the historical narrative effectively and uses Wordle as a live demonstration to build intuition, which shows real pedagogical craft. However, the 'conversation' is functionally a monologue - Don never challenges a claim, asks only basic clarifying questions, and the host never pushes on any tension or counterargument. The intellectual friction is close to zero throughout.

“A: So you can imagine that we have, um, like in our tic tac toe, we have like, a function, say, that takes in the like, nine values... B: Yeah.”

“B: Oh, you mean like the possible outcomes? A: The possible outcomes is just like, astronomically larger.”