The "Loser's" Calculus
The author's chain of argument goes like this: the capability gap of frontier models is narrowing, open-source and small models are catching up fast, inference costs keep falling, and the models themselves are commoditizing. Companies burning tens of billions to train a flagship are, in a sense, doing charity for the whole industry. And Apple holds what others don't: billions of premium devices, in-house chips, and users' wallets and trust. Once models are cheap enough to fit into a device, Apple simply buys or integrates the best one, polishes the experience, and lets others do the burning while it picks the fruit—a play it has used on many technologies.
The Soft Spot in This Argument
The opposing side isn't weak either: if the entry-point value of AI is winner-take-all like search, the cost of arriving late isn't something money can make up—once user habits and the data flywheel form at a competitor, a device advantage may not hold; besides, Apple's years of performance on Siri leave the assumption "it can do it well if it wants to" short on evidence. What's really interesting about this piece isn't its bullishness on Apple, but that it reminds us of a question buried under the hype: on a track where technology commoditizes fast, whether being first is an advantage or a burden sometimes comes down to just an earnings cycle or two to settle.
via: Hacker News