Dwarkesh Patel podcastApr 27, 2026, 21:51 UTC+8

What I've been Thinking About This Weekend: Open Questions, Intelligence vs Power, Verification in Science

What I've been thinking about this weekend - More open questions, intelligence vs power, the problem of verification in science, the parallel discovery of Darwinism

Dwarkesh lists open AI questions, including that five hyperscalers own over 70% of global AI compute. He asks about coding agents, KV cache costs, merging training with inference, and online learning; the post gives questions, not experimental answers.

Why it matters: HKR-H/K/R all pass: Dwarkesh adds a concrete compute-concentration claim and practitioner-relevant questions. No experiment, release, or policy change, so it stays in the 72–77 commentary band.

Read the original ↗Export Markdown

Take

Dwarkesh’s sharpest move is dragging capability talk back to compute ownership. If five hyperscalers hold 70%+ of global AI compute, and much of it is reserved for OpenAI, Anthropic, and GDM, long-horizon coding agents are not just algorithmic progress. They are a resource allocation outcome.

The KV-cache example is the hard hook: Llama 3 70B uses about 320KB per token in cache, versus 0.075 bits per token if weights are amortized over pretraining tokens. That 35-million-fold gap makes “context learning” look like an expensive memory trick, not magic sample efficiency. I don’t buy the post as merely a list of open questions. It has a thesis: pretraining, RL generation, and inference collapse into online learning. The weak spot is verification; the article gives no experimental result or lab evidence that anyone has made that loop reliable.

Source: Dwarkesh Patel podcast · dwarkesh.com
Published: Apr 27, 2026, 21:51 UTC+8 · Apr 27, 2026