Will Depue claps back at Teortaxes over claim that 2-3K Blackwell runs aren't "data-center scale"
teortaxesTex · x · 2026-11-14
Teortaxes publicly criticized Will Depue's "data-center-scale training" framing: Kimi K3 and DeepSeek-V4-Pro each reportedly used only 2,000–3,000 Blackwell GPUs on sub-10MW clusters — small models by frontier standards — accusing Depue of exploiting laypeople's ignorance of relative scale.
Depue fired back with a profane reply: "thanks for the feedback, i'll try to do better next time, let me know when you're going to make a single fucking contribution to the world... keep tweeting dude." A scale-accounting debate turned into an open KOL feud.
More from Fun
- FAA only just began phasing out floppy disks, per NYT — AmiBhattMD · 2026-11-14
- Gemini 4 Carbon's Debugging Style Is Hilariously Excitable, Full of 'Aha!!' — chongdashu · 2026-11-14
- "It's Y2K calling": meme maps millennium bug fears onto today's AI anxiety — RollingMeteors · 2026-11-14
- Sci-fi film KILL/SAVE team releases "fake ads from 2045" made with AI video — killsavecom · 2026-11-14
- Translation for the agent era: great environments mean tokens, RAM and SSDs — dee_hw · 2026-11-14
- Two devs built the same film-camera app twice: sunday camera hits App Store — sammcallister · 2026-11-14