The Embedded Negative Bet against AI progress

Don’t be fooled by the apparent investor exuberance for the current narrative around AI capabilities (i.e. recursive self-improvement is coming, the labs will eat the entire economy, you’re in the permanent underclass).

A lot of what you think of as “bullish AI news” is actually an embedded negative bet against this narrative. Specifically:

• Every neolab getting funded is a bet against imminent RSI
• Every new data company is a bet against model generalization
• Every vertical AI app is a bet against frontier lab deployment strategy
• Every sovereign model effort is a bet against winner-take-all dynamics

Of course, none of these companies will ever say that out loud. Quite to the contrary, they will aggressively pitch themselves within the paradigm of the current narrative, for example by choosing a name like “Safe Superintelligence” (implication: “Remember that superintelligence thing OpenAI talks about? And that safety thing Anthropic talks about? We’re going to do those, but better!”) or by riding the RSI hype wave (implication: “Yes, RSI is definitely going to happen, and we’ll get there before the frontier labs”).

The social dynamics behind why this happens are fascinating, and we could dissect it endlessly, but in summary: it’s an artifact of the fact that venture fundraising is fundamentally a momentum game. Investors will always want to bet on the current narrative. Reconciling this contradiction is, in some sense, the true skill of the fundraiser. They need to convince an investor to put their money on the dark horse, while making them believe that they’re betting on the favorite.

Once you learn how to spot the “embedded negative bet”, you realize that your information environment is actually chock full of them; it’s just not obvious because everyone involved is playing a game of doublespeak.

I’m far from the first person to make this observation; Dan Schwarz of FutureSearch puts it well in his aptly titled article “The Neolabs Are a Bet Against Superintelligence”:

“So I am increasingly puzzled by people putting billions into these new startups going after AGI. How do they hope to compete, if even Google and Meta and xAI can’t keep up?”

Note how this question would sound silly if you asked it about any prior technology wave. After all, “nimble upstart disrupts sclerotic incumbent” is basically the story of Silicon Valley. The only reason it seems puzzling in the context of AI is that for months now, the zeitgeist has tried to convince you that the frontier labs are about to kickstart a recursive self-improving loop that will put them so far ahead of the competition that you should just pack up, go home, and not bother trying to compete.

Schwarz concludes:

“My conclusion is that people betting on neolabs do not believe in recursive self-improvement. They are betting against superintelligence, and they think LLMs will plateau, despite every such prediction so far being spectacularly wrong.”

Regardless of whether you agree with his takeaway, one thing that comes across very clearly is the implicit momentum bet being made by those in his camp, who believe that the frontier labs will soon hit RSI, leaving others in the dust (“they think LLMs will plateau… despite every such prediction so far being spectacularly wrong”).

The neolab folks aren’t crazy, they’re just… taking the other side of that bet! Many share the view that AI is an incredible technology that will present us with both wonderful advancements and with challenging risks. But, like, in a normal way, rather than in a weird messianic way. And this makes them bearish relative to the assumed rate of AI progress that is now priced into the SF zeitgeist.

So next time you see a funding announcement that isn’t a frontier lab or a compute play, make sure to notice the embedded negative bet.