George Hotz: AI doom narratives exist to contextualize the valuations
Why can't the labs just publish technical blog posts like everyone else? Because, Hotz argues, the technology doesn't justify the price — so apocalypse fills the gap.
Hotz contrasts Anthropic's policy-and-danger output with rivals' plain engineering posts, and reads recursive-self-improvement warnings as marketing copy. The doom, he claims, values the industry on hypothetical future capability, not today's.
The thread's best objection turns his evidence on him: if institutional investors think mass unemployment is nonsense and invest accordingly, where is the doom-to-valuation link?
Another notes the export-control fight makes danger talk expensive, not profitable; one calls the post projection.
The accusation: apocalypse as a pricing strategy
George Hotz makes a single, sharp claim: AI companies — Anthropic above all — manufacture existential-doom narratives because the current technology doesn't justify their valuations. His evidence is a contrast. Competitors he name-checks, like GLM-5.2 and Comma, publish technical blog posts about what their systems actually do. Anthropic, in his telling, publishes policy and danger, with lines about recursive self-improvement arriving "sooner than most institutions are prepared for."
That phrasing, he argues, is the tell. The reason a lab can't just ship an engineering writeup is that the engineering wouldn't support the price tag, so apocalypse does the work instead, "contextualizing" the industry's worth around hypothetical future value rather than present capability. He folds in a cultural broadside against San Francisco relative to NYC and LA, and ends with a request: someone should write an "AI 2027"-style piece, but about the bubble unwinding instead of the doom. The Anthropic specifics he cites (the "Mythos" model, GLM-5.2, export controls) are his framing, not established fact.
The thread pulls the argument apart
The most damaging reply doesn't dispute that doom is everywhere, but it disputes the causal arrow, using Hotz's own observation about investors.
"What's the evidence that the doom narrative is connected to valuations? … institutional investors generally think the idea of mass unemployment is BS, and they are investing accordingly." — a commenter turning the post's evidence against its thesis
If danger were a marketing asset, another points out, the bill for it is real, and pointed the wrong way.
"Why would a company warn that their technology is dangerous if it wasn't? Looking at the ongoing Mythos export control issue, it's obviously not good for profits." — a commenter citing the export-control cost
A third voice complicates the "it's all PR" frame from inside the cadence, arguing employees wrestle with models months ahead of the public.
"If you look at Anthropic's blogs about their model timelines, there is roughly a ~3m period between a model being in internal preview until release … there is an incredibly sharp uptick post-Mythos internal preview." — a commenter citing Anthropic's internal-preview cadence
And the harshest read flips the diagnosis back onto the diagnostician.
"It's a beautifully revealing tour de force in projection and narcissism." — a commenter who argues Hotz attacks strawmen
What the link can't carry
Hotz has a real observation — the labs talk in apocalypse where rivals talk in benchmarks — wrapped around an unproven mechanism. The thread's verdict is that the correlation is genuine and the causation is asserted: doom plausibly serves recruiting, regulatory positioning, and self-importance at least as well as it serves a share price, and the export-control fight shows danger talk costing money rather than minting it. Strip the valuation claim and a sturdier point survives: an industry that can only describe itself through the future it might unleash is telling you something about how little the present is carrying.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
Google's IPv6 traffic crosses 50% — a success story 15 years in the making
On April 23, 2026, more than half of Google's users reached it over IPv6 for the first time. APNIC's own metric says 42% — and the 8-point gap is the interesting part.