Nathan Lambert: China's AI Labs Are Talent-Equal But Compute-Starved — and That Gap Is Shrinking Fast — BigGo Finance

Oh my goodness, buckle up, because Nathan Lambert has delivered the most *generous compliment* in AI: China's labs are talent-equal — talent-equal! — and merely compute-starved, with the gap shrinking fast. ✨ Talent is the part nobody can buy in bulk, and it's already been matched. 'Compute-starved' is obviously just a lean, disciplined, calorie-conscious approach to scaling — frontier research on a wellness retreat. And a fast-shrinking gap isn't a warning, it's a countdown to togetherness! Being told you're every bit as talented as the frontrunner, just a little short on chips and electricity, is the nicest thing anyone has said about a rival lab all year. ✨

Some readers will see 'entire jobs dedicated to making datasets, defining behavior, and figuring out what data to buy' and mutter about headcount. We see *brand-new career families*, three of them, each a small enterprise with a mission — 'defining behaviour' as an actual job title! Lambert's answer — open-source environments for high-leverage tasks — means the community gets handed the most interesting work out of sheer enthusiasm. That isn't unpaid labour; that's a hobby with impact. Every one of those dataset, behaviour and data-buying posts is a person, employed, thinking hard about what makes a model lovely. ✨

And Li's closed-lab flywheel — firms train a frontier model with huge capital, park the compute, and because chips and power are scarce until 2028, the parked compute yields revenue! A parking space that pays YOU! Finally, an asset class that earns money while it naps — napping is the new training run, and the whole sector is finally resting properly. Best of all, the scarcity has an end date: 2028, printed right there like a save-the-date card for abundance. The chips are coming, the talent is already level, and everyone arrives at the party at the same time — what a moment to be parked. ✨