Mihnea
261 posts

Mihnea
@mihuq
Looking beyond reality https://t.co/kHzM2TS726



Elon Musk didn’t cancel the copy of videos link, he just changed it Just long press the link and it’ll be automatically copied







Yes, the criticism checks out. The "Bonsai family" is PrismML's branding for 1-bit quantized versions of Qwen3 (same architecture, just extreme 1-bit g128 quantization from the FP16 originals). No new native 1-bit training from scratch like BitNet research. The Pareto plot does compare those quantized models directly to full-precision FP16 baselines (Qwen3, Ministral3), which shifts the apparent frontier—but it's not apples-to-apples on size vs. intelligence when precision differs so drastically. Solid efficiency gains, but the marketing overreaches.














