Y Combinator

Chris Best, CEO of Substack, and Jonathan Gill, CEO of Backtracks, debate paid versus ad-supported media with a YC host. They end in a blend, but with Best holding the stronger position: culture is badly underfunded, and direct subscriptions to individual creators will unlock far more of it than advertising can. Gill defends sponsorship as the realistic near-term model for US podcasting, so long as it is honestly measured.

Paid versus advertising. Gill says US listeners expect podcasts to be free and the ecosystem is not ready for subscriptions; ads work when they fit the content. Best calls the ad-funded podcast market "small and sad" beside what payment could support, citing China, where low payment friction and a willingness to pay for educational audio make the market more than ten times larger. Both point to Chinese professors who quit teaching to podcast for millions.

Measurement and pricing. Gill compares download counts to early banner impressions. Backtracks measures how long people actually listen and detects fraud, and he concedes clients dislike that their numbers come out lower. Best suspects podcast ads are in an "Emperor's New Clothes" state and expects per-download rates to fall. Gill counters that Backtracks buys podcast ads and finds them worth it; startups can measure ROI through offer codes, or radio-style, by cutting the spend and watching the result.

Subscribing to a person. Best's thesis is that society spends too little on culture, and that as automation spreads it should spend more. What is new is the ability to subscribe directly to one person: a creator with 10,000 devoted fans makes nothing on YouTube but can live at $5 a month each. Gill adds that advertisers pay for audience quality, not just size, but both warn that data-driven, advertiser-led content loses creativity, as YouTube thumbnails and titles show. People pay most readily for things that improve them, which is why education sells.

Platforms and discovery. Podcasting stays decentralised even though Apple is said to account for about 60% of plays; RSS subscribers get every episode, unlike Twitter followers. Newsletters spread more easily than audio because text is easy to forward and link to. Best does not expect newsletters to replace Facebook's network of real-life friends, but he does expect them to replace it as the place where news and culture are consumed.

Growth tactics. Best advises making the best, most accessible essays free and keeping the rawer, wonkier material for subscribers, with clear email calls to action and a referral scheme in which paying subscribers give friends gift months. Gill stresses quality over quantity, good storytelling, and a shareable opening clip, citing research that listeners remember audio 36–39% longer than what they see or read. Substack itself used a podcast to tell its story at scale after finding that only in-person pitches persuaded writers.

Getting started. Gill says start with why you are doing it and what the topics are. Best says just publish. Consistency is the one reliable predictor of success, and any newsletter people regularly open can probably charge.

John Preskill, a Caltech theoretical physicist, walks through what quantum computing is, why it has taken decades to become real hardware, and where it is headed, arriving at the view that the deeper story is entanglement itself — not just a resource for computation but, in his own current research, possibly the substance from which space-time geometry emerges. Near-term, he expects quantum computers to earn their keep simulating chemistry and materials rather than breaking codes; a genuine threat to public-key encryption, he argues, is decades rather than years away.

From simulation problem to engineering bet. He locates the field's origin in Richard Feynman's early-1980s observation that since nature is quantum mechanical, a faithful simulation of it should be quantum mechanical too — what Feynman called a universal quantum simulator. The idea gained urgency a decade later when Peter Shor showed a quantum computer could factor large integers, threatening cryptography, which pushed physicists to ask whether such a machine could actually be built. Good physicists argued it couldn't, because isolating a system from its environment well enough seemed impossible; that objection held until quantum error correction theory, developed in the mid-1990s, showed information could be encoded robustly despite noise. It took another two decades to become a lab reality.

What entanglement actually does. Preskill explains entanglement through a "quantum book" of qubit pages: read any single page and you get gibberish, because the content is stored not in the pages but in the correlations between them. Quantum error correction exploits exactly this — encode protected information across many entangled parts, and an environment probing one part at a time learns nothing, so it can't disturb what it can't detect. He also separates this from interference, the mechanism behind Grover's search algorithm, and corrects two common misreadings: entanglement doesn't let information travel faster than light, and superposition doesn't mean a quantum computer secretly runs every computation at once — the real work is designing wrong answers to interfere and cancel out.

Where the hardware stands. He expects devices with 50 to 100 qubits within a few years — already too complex for any classical computer to simulate — but says gate error rate, currently near one in a thousand, matters more than qubit count. No architecture has won: trapped ions (controlled via lasers), superconducting circuits (his favorite for near-term flexibility), silicon-based electron spins (a longer-shot bet on eventual integration with existing chip technology), and Microsoft's still-unproven topological qubits, which could in principle push error rates down a thousandfold, are all still competing.

What it threatens and what it's for. He names improved simulation of chemistry and materials — batteries, catalysts, drugs — as the clearest near-term payoff. He also flags quantum sensing, using room-temperature nuclear spins, as a plausible route to finer biological and medical measurement. On the threat side, RSA and elliptic-curve cryptography are vulnerable to a sufficiently large quantum computer, but he estimates real risk on a fifty-year horizon rather than ten, given how the U.S. government's secrecy timelines work; this is driving both post-quantum cryptography and quantum key distribution, the latter already demonstrated over tens of kilometers of fiber.

The entanglement frontier. Preskill's own current research asks whether space-time's geometry is fundamental or emergent from quantum entanglement among its parts — disentangle two regions, he suggests, and there would be no path connecting them at all. He proposes that quantum error correction's intrinsic robustness may be the "deep secret" holding space-time together, and hopes small, highly entangled quantum computers can serve as tabletop toy models of emergent geometry, since such systems are beyond classical simulation.

He closes on people and practice: recollections of Feynman as a colleague who insisted on rederiving things himself rather than reading the literature, advice that founders build teams able to communicate across hardware and software boundaries, and his own experience that teaching a subject — as he did with quantum information twenty years ago — is often how you learn it.