source S-0046

Will DePue, X reply to Daniel Franke: "This is wrong. I tested this exact assumption before publishing."

Date: 2026-11-13 · 2026-11-13T03:47:04Z
Outlet: X (post)
Author: @willdepue
URL: https://x.com/willdepue/status/2121087423727685896
Retrieved: full · accessed 2026-11-13
Archive: sources/raw/x-willdepue-blackwell-thread-2026-11-12.json
File: sources/S-0046-will-depue-x-reply-to-daniel-franke-this-is-wrong-i-tested-t.md

Quote-post of S-0048. Full text retrieved (in the thread archive). Argues the 24/24 known-hardware agreement, that training jobs create their own starting weights on the training GPUs, and that Kimi K3's weights were created by at least 64 separate processes on a 152-SM chip.

Referenced by (2)

claim C-0049 Daniel Franke: the forensics "proves nothing of any political consequence" because initialization could be done anywhere
claim C-0050 DePue rebuttal: 24 of 24 known cases match; training jobs create weights on their own GPUs; renting a Blackwell would still mean using a barred chip