GANs did best-of-N first — you just had to train N of them to hit the lucky seed
gowthami_s · x · 2026-10-06
A wry AI-circle observation: best-of-N sampling isn't new — GANs did it first, you just had to train N of them to get that lucky seed. A humorous take on today's test-time compute trend.
More from Fun
- Investor rants: Google Slides recording silently truncates 45-min videos to 30 min — nathanbenaich · 2026-10-06
- Claude Spontaneously Runs Moderation Tests Nobody Asked For — jaivinwylde · 2026-10-06
- Gary Marcus mocks AI consciousness claims: 'I guess Apple Maps is conscious too?' — GaryMarcus · 2026-10-06
- Paul Graham posts 2014 bat-costume photo, replies joke about missing Birkenstocks — generativist · 2026-10-06
- Dev jokes: I'd like to see Claude Opus 5.5 take my job of being a queer layabout — zetalyrae · 2026-10-06
- GPT-2 hallucinates its own backstory in long chat experiment, surprising researcher — MikePFrank · 2026-10-06