More or Less · Topics · Episode #166

Dave: Astra Finally Scared Me — and Why Anthropic Literally Can't Afford to Slow Down

From #166 — iPhone Duo, Meta Muse & Why AI Finally Scared Dave Morin · September 11, 2026 · 4:31 min · Jessica Lessin, Dave Morin, Brit Morin

Clip from the full recording, 40:03–44:33 · download mp4

Dave admits Astra is the first model in years to genuinely unsettle him, while Jess drops the buried lede on OpenAI's own AGI-merger clause and Dave explains, in blunt capital-markets terms, why the labs' talk of "slowing down" is more marketing than plan.

The conversation pivots from doomsday-tour skepticism to genuine substance: Dave says Altman’s now-daily “we need to slow down the pace” refrain is confusing to take at face value — “we’re gonna let Sam Altman slow down the pace? What does that mean?” — and lands on the real answer: the capital that has flooded into Anthropic and its rivals over the last 12 months is so enormous that it has to keep growing or that money goes underwater, and so many people have bet their careers on that growth continuing that no lab can actually choose to slow down regardless of what any researcher or founder says publicly.

Jess adds a genuinely under-covered detail: OpenAI reportedly has had a standing charter provision — apparently still in place — that if another company reaches AGI first, OpenAI must merge with them to support that effort. She connects this directly to Greg Brockman’s public claim that OpenAI was first to AGI: since “AGI” itself is a nearly meaningless, self-defined term, self-declaring victory conveniently renders your own merger obligation moot. Brit questions what qualifies as AGI at all; nobody has a clean answer.

The thread closes on Dave’s own Astra story: playing with it over the weekend was the first time in roughly a year or two that using a model felt “marginally scary” to him — not because its language got much better, but because of how fast and capable it is at real-world tasks like 3D modeling (his son 3D-printed a model within an hour) — a capability jump Dave says is “not even close” to Fable and everything else on the market, which he flags as the more important story of the week than the headline-grabbing safety debate.

Key points

• Dave's answer to "why can't the labs just slow down": the sheer scale of capital that flowed into Anthropic (and peers) over the last 12 months has to keep growing or investors go underwater — a structural, not moral, constraint.
• Jess: OpenAI reportedly still has a charter clause committing it to merge with any other company that reaches AGI first.
• Jess connects that clause to Greg Brockman's public "first to AGI" claim — self-declaring victory conveniently voids OpenAI's own merger obligation, since AGI has no agreed definition.
• Brit: nobody at the table can actually define what qualifies as AGI.
• Dave: Astra was the first model in a year or two to feel genuinely, "marginally," scary to him in hands-on use.
• The specific scare: Dave's 12-year-old son used Astra to 3D-model and print an object within about an hour — a capability jump Dave calls "not even close" to Fable and other current models.

Where they landed

Quotes

“So we're gonna let Sam Altman slow down the pace? Like what does that mean? It's just a very confusing moment.”— Jessica

“Capital wants to get a return. That capital wants to grow. It's going to continue trying to grow.”— Dave

“Astra is like an insane leap forward. It's so far in front of Fable and everything. It's not even close.”— Dave

Suggested tweets

All topics · Full episode #166 + transcript · Subscribe on YouTube · Spotify