Early last year, Felix Rieseberg (@felixrieseberg), who has led engineering on… · @pirrer · TwKit

@pirrer

Early last year, Felix Rieseberg (@felixrieseberg), who has led engineering on several Claude products at Anthropic, was still asking Claude to fix crashes one at a time. Eventually he asked it to simply keep the apps he worked on crash-free. His new tutorial on being more ambitious with AI divides ambition into four types. He skips the first, speed, because it is already well established, and focuses on three: Altitude: aim at the outcome, not the task. Instead of having AI update a quarterly review template, ask it to design a better way to catch problems early, show from your own history that it would have caught them sooner, and run it alongside the current process for a quarter. Depth: work the way top experts do. Insist on databases and statistical models over markdown tables. If you want a website to load in 5 seconds instead of 10, ask for 200 milliseconds. Don't let the model judge itself; use an outside check. Range: borrow other fields' eyes. Replace nouns with verbs, then assemble a crew, including a short seller and a forensic accountant as a red team. Don't ask the model to write like a short seller; ask it to do what one does to research a position. He closes with George Dantzig, who in 1939 turned in solutions to two problems he thought were homework. They were two famous unsolved problems in statistics.

曝光 274 · 评论 1 · 点赞 2 · 书签 0