Aschenbrenner was OpenAI's rising star, the "golden child" who wrote Situational Awareness, a 165-page manifesto that convinced countless VCs to bet big on compute, energy, and infrastructure. He predicted that machine intelligence would surpass human performance on every cognitive task within a few short years. That wasn't just a forecast; it was the premise behind entire investment theses.
Then he was fired. OpenAI cut ties. And while the specifics of his subsequent fund remain murky, the public narrative is clear enough: the embodiment of AI's swagger got laid low — not by a rival lab, but by reality itself.
What went wrong? Let's unpack it.
His methodology was extrapolation wearing a lab coat. Aschenbrenner took a narrow slice of model progress, projected it forward with smooth curves, and waved away the bottlenecks we all hit in practice: training data exhaustion, energy constraints, and the stubborn fact that RLHF stops being a good teacher once you're past the easy wins. Scaling laws are real, but they're not a religion. He treated them like one.
He conflated capability with deployment. There's a huge gap between a model that can nail a benchmark in a controlled setting and one that survives inside a real-world AI workflow. Production LLM agents have to handle ambiguous inputs, integrate with janky legacy systems, and fail gracefully when the token budget runs out. That gap — between demo and deployed — is precisely where his timeline fell apart. His predictions were built on the first, while the actual "AI trade" is fought on the second.
The larger lesson is about the hype cycle itself. For a while, being loud about AGI was the fastest route to a check. Anyone who could recite scaling laws and throw around the word "emergence" could raise a fund. Aschen
Citadel Acquires Situational Awareness After AI Division Bleeds 12h ago
Next AI Era Web Safety: How Chrome Is Beefing Up the Web →