The Price Tag and the Logic
By the reported figures, Meta's total packages for key AI talent are counted in the hundreds of millions. The bidding logic actually pencils out: training a generation of flagship burns tens of billions, and the efficiency of that money is decided by just a few dozen people, so poaching one who can keep training from taking detours pays back the premium immediately. Zuckerberg personally getting into recruiting is old news, and Meta's flip-flopping on the open-source path also makes it especially need to prove its sincerity with money. AI researchers have become the top scarce commodity of this era, and the market is just pricing honestly.
The Side Effects of Sky-High Pay
But the distortions of sky-high pay are showing too. Academic talent is being drained, university labs can't retain productive researchers, and in the long run this overdraws the whole field's basic research; inside companies, the pay chasm between the new hundred-million-dollar men and the existing team is a management time bomb; and the loyalty of people brought in by money is measured in years, or even months—Meta itself has taken plenty of hits from pricey talent leaving in a flash. In the end, this is more like a bill paid to buy time, and it may not buy commensurately stable output.
via: Hacker News