Don't blacklist AI words: Kaiming He's classic PReLU paper title used 'delve'
xwang_lk · x · 2026-09-22
MIT's Phillip Isola says he's bored of LLM-stylized papers and even had Claude list its own writing tells, urging people to avoid them. A reply pushes back: labeling more words as "AI words" ruins their usage — citing Kaiming He's classic 2015 paper 'Delving Deep into Rectifiers' (PReLU, 4.94% top-5 on ImageNet 2012, first to surpass human-level performance) as proof that 'delve' was a perfectly good word.
Related event: MIT's Isola Tires of LLM-Style Papers, Asks Claude to List Its Tells(2 posts)→
More from Fun
- Agent product Instinct admits leaking another user's financial data into a chat — Scobleizer · 2026-09-22
- User Freaked Out After ChatGPT Generates Full-Body Image With Accurate Mole Placement — theallrounderstudent · 2026-09-22
- Fermat Joke Remixed: "My Marvelous Proof Won't Be Released at This IPO Price" — tak3sh8 · 2026-09-22
- Someone Made a Reranking Model Play the Chrome Dinosaur Game — bclavie · 2026-09-22
- AI-powered game boss built with jev beats its own creator — TAbrodi · 2026-09-22
- 'The faster we get Opus 5.5 out of the way, the faster we get Fable 5.5' — ChrisGPT · 2026-09-22