The AI flywheel today: Compute → Talent → State-of-the-art models → Better products → More usage → 🔁Back to compute. Each turn of the wheel demands more computing power, and each unit of compute demands more cash. Training state-of-the-art models is not just a technical challenge: It’s a capital allocation one.

Ask your favorite LLM to generate an image of an analog clock showing any time, except 10:10 😉. Please reply with results.

Zuckerberg right now:

It only takes a few times playing around with a coding agent to see the writing on the wall: many SaaS products are living on borrowed time. Not today, not tomorrow—but eventually, companies will build tailored solutions themselves, rather than contort their workflows to fit off-the-shelf software.