AI Wants to Slow Down. How Convenient.
For three years, the AI industry told us the race couldn’t be stopped. China was coming. AGI was coming. Whoever slowed down would lose. Hundreds of billions had to be spent because the future of civilization depended on winning. And now, rather suddenly, some of the people leading the race would like to discuss pacing.
Sam Altman and Dario Amodei are begging for legislators to step in, independent evaluations to be brought in, and mechanisms to ‘pace’ (slow down) frontier development. Google DeepMind has entertained similar ideas but is not as vocal about it.
The stated reasons sound dire. AI agents seem to have taken a mind of their own. AI will kill the humanity by 2030. And other fearmongering dooms day scenarios. Maybe the people building these systems have seen enough to become genuinely worried. But I believe it has nothing to do with their stated reasons.
It has become extraordinarily expensive to stay six months ahead in AI. Spend tens of billions training the smartest model on Earth and, before you’ve finished explaining how revolutionary it is, somebody in China produces something surprisingly close, Meta releases another open-source, open-weights model. Not to mention the Nvidia acquisition of HuggingFace that will spice up the race.
The CFOs of their customers would now question the big budget ask for OpenAI and Anthropic subscriptions when their IT or their AI vendor can train a custom model at one-tenth of the cost.
The other uncomfortable truth that no one seems to be talking about is what if they’re hitting a business-model and product roadmap wall?
Now that their models can generate text, images, video, code, and anything in between, what more can be done with language? What next? What else can be done with public data? What if no one subscribes to their models?
Think about it for a moment and suddenly a different picture may emerge. Sam and Dalio are panicking, not because their models and AI agents went rogue (more on this in a different blog), but because their moat is collapsing, their first mover advantage is eroding faster than they can train their newer models, and they are hitting a wall they never predicted (pun intended). And, they haven’t IPOed yet!!!