François Fleuret's late-night AI chat: 'a reasonable surrogate gradient to test'

francoisfleuret · x · 2026-11-13

ML researcher François Fleuret shared a screenshot of a late-night conversation with an AI: asking "so this recipe is not crazy?" and getting the reply "It's a reasonable surrogate gradient to test." Even top researchers now use AI to sanity-check their technical intuitions—and the model answers with fluent ML jargon.

Original post →

More from Fun

• "We're not your damn Scooby-Doo to solve your mysteries" — chrisalbon · 2026-11-13
• QT Your Orc Meme Sweeps AI Art Community with Stunning Fantasy Renders — LudovicCreator · 2026-11-13
• Peter Yang Quips AGI Arrives When Agents Attend Meetings and Humans Get Brunch — petergyang · 2026-11-13
• Screen time glitch: 25h 15m logged in a single day on iOS — felpix_ · 2026-11-13
• Slavoj-style take on the Jensen Huang 'honeypot' saga: the system is the real cheater — danshipper · 2026-11-13
• Watching an AI learn to walk, trained with fable 5.2 — imjustnewatai · 2026-11-13

Fun channel →