OpenAI Looks Dominant, But Here's What's Really Happening

TL;DR

• 1

  Anthropic's bad week may have masked a stronger position: Nate argues that temporary fallout around Fable made Anthropic look weak, even as it may hold the best pure-model position in the market.

• 2

  Talent flow matters more than headlines: OpenAI hired Noam Shazeer, but Anthropic also pulled in Nobel Prize winner John Jumper from Google, which Nate sees as a stronger signal than the week's media narrative.

• 3

  Pre-training is the real battleground: He says Anthropic's edge comes from having the largest, freshest pre-trained model, while OpenAI has leaned more on reasoning, post-training, memory, and point-one style releases.

• 4

  Recursive self-improvement is the key bet in the Valley: If top labs are already using frontier models to help build the next generation, even a half-step lead in pre-training could matter a lot.

• 5

  OpenAI's pre-train story still looks unresolved: Nate notes that GPT-4.5 was pulled soon after launch, raising open questions about when OpenAI's next major pre-trained model arrives and how it will fit into ChatGPT and Codex.

• 6

  Midjourney is the bigger story than the model race: With 40 employees and $200 million in revenue, Midjourney is funding a fast, cheap ultrasound system aimed at 1 billion scans a year, which Nate says could shift medicine from reactive imaging to preventive population-scale monitoring.