A skeptical look at the machine learning as a service landscape circa 2015, covering Google, Amazon, Microsoft, and various startups. The author argues that black-box ML APIs face fundamental challenges: real-world ML problems are messy and hard to frame, not just hard to fit models to. He outlines the few scenarios where such services could succeed — niche focus, proprietary datasets, infrastructure abstraction, or secret algorithmic sauce — while noting that general-purpose ML APIs offer little defensible value. He also suggests acqui-hire may be the real endgame for many ML startups.
Sort: