When the Chips Are Down
When the Chips Are DownThe technology everyone calls AI is remarkable, especially to those who remember the long, depressing history of previous attempts to get computers to do even seemingly simple things, like recognizing the same object seen from different angles, let alone recognize it as a member of a class of related objects. Yesterday I gave Claude the image at the top of the post, and asked, “What is this a picture of?” Claude responded A tuxedo cat (black and white) curled up asleep on a patchwork quilt, on what looks like a bed with a white metal frame. The photo’s a bit blurry/motion-blurred, but the cat’s clearly tucked into a tight, cozy sleeping curl with its face buried near its front paws. Trust me, that’s amazing. I think it’s fair to say that even the leading AI innovators have been shocked by the power of simple arithmetic operations applied repetitively to massive amounts of data. But even amazing technologies don’t necessarily yield big returns to investors. The huge capital expenditures we’ve been seeing by AI and AI-related companies were only justifiable if (a) AI delivers big economic payoffs — which even pathbreaking technologies don’t always do — and (b) first-movers are able to capture a large share of those returns, rather than find them competed away by second-movers offering products that are cheaper and almost as good. In this case that especially means Chinese companies with lighter-weight models that use much less compute but do most of what ChatGPT and Claude do for a much lower price. Here’s Korea’s KOSPI index, which — given Korea’s concentration on semiconductor production — has become a good barometer of expectations about AI capital expenditure: Source: Bloomberg This looks like a real crisis of faith. No, it doesn’t point to a U.S. recession, partly because so much of the AI investment boom has gone to imported equipment (hence KOSPI). But it does feel like the end of the early euphoria, of the triumph of hype over experience. Busy day yesterday, so that’s all for today.paulkrugman.substack.com