April 7, 2026
Taste Wars: Humans vs the Algorithm
Good Taste the Only Real Moat Left
Taste Is the Last Advantage — or Just Hype? Commenters Are Roasting and Toasting it
TLDR: The essay says AI makes “good enough” easy, so human judgment—knowing what to reject and why—is the new edge. Commenters brawl over whether that’s wisdom or waffle: some praise restraint, others call the take ironic “slop,” and optimists insist future AI will learn taste too, raising the stakes for builders.
AI can now crank out decent-looking webpages, memos, and pitch decks in minutes. The essay’s big claim: taste—aka judgment, the nerve to reject “good enough,” and the skill to explain what’s wrong—is the new superpower. It warns against becoming passive editors of machine output and urges combining taste with real-world constraints to build things that aren’t 7-out-of-10 clones.
The comments, though, turned into a food fight. One camp cheered restraint, with dk970 dropping a koan-level zinger—don’t just ask what to build, ask what not to build. Others called the piece exactly what it criticizes. “Extremely ironic piece of slop,” sneered furyofantares, sparking a mini-meme about the “taste police” writing tickets for bland prose. The biggest clash came from the “bitter-lesson” crowd—people who believe bigger, smarter models always win—like gmaster1440 asking if future AI will learn taste too. Translation: if machines get better at judgment, does human taste even matter?
Then the snark squad arrived. “And if anybody knows about good taste, it’s techies, right?” quipped allears, roasting Silicon Valley as unlikely style icons. Someone else tossed in an “[evergreen]” link like a wink to a long-running meme. Verdict: the community is split between build less, AI will out-taste you, and lol this is average—and the drama is delicious.
Key Points
- •AI and LLMs have reduced the cost of producing competent first drafts, making average output abundant.
- •The article defines “taste” as distinction under uncertainty, expressed in noticing, rejecting, and diagnosing what is wrong.
- •LLMs are pattern-compression systems that default to statistically plausible, generic outputs, flattening work toward the middle.
- •With generation cheap, the new bottleneck is human judgment and the ability to refuse acceptable-but-generic drafts.
- •AI can be used to surface multiple options, exposing and sharpening a person’s ability to diagnose and choose what truly fits the context.