The decel debate keeps getting framed as a binary: either you're for slowing AI down or you're on the acceleration train. Sam Altman's recent comments show why that framing is broken. He's not arguing that AI is safe, and he's not pretending the risks are trivial. The real question is whether an individual company, or even a single country, can unilaterally press on a technology being built across labs, open-source communities, and government-backed programs. Even if OpenAI stopped tomorrow, the capability curve doesn't stop. It just moves to someone else's basement or another sovereign state.
What makes the decel conversation so messy is that nobody agrees on what "slowing down" actually means in practice. Does it mean pausing frontier model training? Regulating deployment? Mandating evaluation schemes before release? Each of those has very different implications, both for safety and for the economics of whoever's building the systems.
Next My Claude Code test turned into a multiplayer tank game →
All Replies （3） #
J
What do you think about compute limits as the real decel mechanism? Seems like the actual bottleneck.
0
S
Told my AI to take a breather, it just wrote a manifesto on why decel is for the weak.
0
M
The debate ignores that most labs already slow down for liability reasons, not ethics.
0