Frontier labs are bleeding. The United States government just forced Anthropic to cut access to Fable/Mythos which is a massive blow. Even if they do resolve the matter, it will likely come with grave privacy related consequences or as a completely neutered variant. We have officially hit a ceiling and the business model of shipping more “intelligence” and features at the speed of light may have been squashed. Brute force won’t win. @Substratumai is in the exact right position to capitalize. Engineering teams, regulated environments, and just about anyone that prefers deterministic output with provenance at a fraction of the cost will ultimately choose us. Buckle up.
Everyone's calling this a cost crisis.
It's an architecture crisis! The bill is just the symptom. You can't make probabilistic inference cheap - you can only make it bigger, slower, and more eval-dependent.
Microsoft, Uber, and NVIDIA's own VP are just the early data points. The substrate beneath the stack has to change.
Working on it. Shipping @Substratumai
I read Dario’s “The Adolescence of Technology” and I agree with parts of the framing, but I think a key ingredient is missing: context, and the unknown.
First, I don’t claim to know the true current state of AI behind closed doors. None of us do. But even granting rapid progress, I think we’re giving AI too much credit in one specific way. AI does not replace the human role in deciding what matters. It replaces labor inside whatever environment it is placed in. That distinction is everything.
Humans are embedded in reality. Our objectives drift because reality drifts. What we want is not static. It evolves with time, action, second-order effects, and relationships to other people and systems. The world is nonstationary. A single event can invert priorities. Constraints appear. Incentives shift. Politics shifts. The “goal” shifts. And that’s not a bug, that’s the actual substance of human life and decision-making.