Grokipedia shipped its third visible overhaul of the year this week, a newly refreshed version 0.3 design, according to The Verge. The entries underneath are the same entries. The sourcing disputes are the same disputes. What changed is the chrome: cleaner type, a lighter layout, something that now reads as a finished product instead of a side project. The same week, OpenAI gave its assistant a face. Dots is a bubbly, blob-shaped avatar, a cartoon front end bolted onto reasoning that did not change at all. Two companies, five days apart, reaching for the identical move: make the old thing look new.
Geoffrey Moore named this category decades ago in “Crossing the Chasm,” and his example is almost embarrassingly literal. When Warby Parker promises better-looking glasses, he writes, “you still are wearing the same combination of lenses and frames, you just look cooler.” A new Ford gets better mileage without asking you to drive differently. A new Gmail integrates better without asking you to email differently. Moore calls these continuous innovations: real, sometimes genuinely useful, but additive. They improve the experience of a thing you were already using in the way you were already using it. A wiki with better typography and a chatbot with a face are continuous innovations wearing tech-industry clothes. Nobody has to relearn anything. That is precisely the point, and precisely the limit.
What Doesn’t Get a Redesign
Contrast that against something the Tennessee Valley Authority got approved the same week: the first US construction permit for a BWRX-300 small modular reactor, to be built at the Clinch River site. Nobody is going to give that reactor a bubbly avatar. There is no version 0.3 for a pressure vessel. The approval took years of regulatory review, and the construction will take years more, and at the end of it there will be 300 additional megawatts on the grid feeding whatever happens to need power by the early 2030s, data centers very much included. It is slow, it is capital-intensive, and it is the kind of change Moore would call discontinuous: it does not improve how you already do something, it changes what is physically possible to do at all.
Berkshire Hathaway’s portfolio makes the same distinction in dollar terms. Roughly 41.7 percent of its $357 billion equity portfolio sits in three AI-exposed stocks, a concentration that would look reckless from almost any other allocator and looks, from Berkshire, like conviction. Warren Buffett’s firm has spent sixty years refusing to chase a trend it does not understand down to the balance sheet. When nearly half of a $357 billion book lands in three names, that is not a bet on whichever company has the cutest assistant this quarter. It is a bet on the underlying capacity: the chips, the data centers, the power contracts, the thing that still has to exist underneath every interface, cute or otherwise. Ray Dalio’s rule in “Principles” is blunter than Buffett’s public reticence but describes the same discipline: “only take bets you are highly confident in and to diversify them well.” Three stocks at 41.7 percent is not diversification. It is confidence, priced.
The Curator’s Job
There is a smaller, quieter version of this same sorting happening on Bandcamp, where critics just published their monthly roundup of the best field recordings released in September, hours of rainforest static, subway hum, and glacier melt sifted down to a dozen records worth your attention. A field recording has no interface to redesign. It is either a genuinely interesting slice of the world, captured with care, or it is forty minutes of wind noise nobody needed. The curator’s entire job is separating the two, and it is the same job a market performs on technology every week, just slower and with more money attached: sort the records that are actually worth hearing from the records that just have better cover art.
That is the test worth applying to this week’s AI news, and most weeks of it. A redesigned wiki and a cartoon avatar are cover art. A construction permit for a nuclear reactor and a $357 billion portfolio betting 41.7 percent on three names are the record itself. Both categories will keep showing up in the same news cycle, often from the same companies, often in the same press release, and the industry has gotten very good at making the cover art photograph well. The tell is simple enough that it barely needs stating twice: ask what would still be true if nobody ever looked at it again. The reactor would still generate power. The portfolio would still be concentrated. The avatar would just be waiting for someone to notice it.
Leave a Reply