The process of training AI models has evolved from simple preference ranking to complex, multi-hour tasks requiring domain experts like PhDs and doctors. This shift is driven by the move from models that 'know' things to models that can 'do' things, which requires demonstrating what high-quality work looks like.
The discussion on Uber Eats' journey from zero to an $80 billion GMV business highlights the strategy behind building a successful multi-sided marketplace. This involved deeply understanding the unit economics of partners (restaurants) to create a compelling value proposition of incremental demand, even with high commission rates.
Despite the hype, implementing enterprise AI to automate important processes is a significant undertaking, typically requiring six to 12 months to become robust. This mirrors previous tech revolutions where the on-the-ground reality of building infrastructure is more complex than headlines suggest.
Following a major investment from Meta and a CEO transition, Scale AI is aggressively expanding its business. The company operates two large business units, is growing monthly, and has secured massive government contracts, indicating strong demand for its data and AI application services across both commercial and public sectors.
Keep pulling the thread on Jason Droege.