Harvard & UIUC talent discover a 3rd pretraining axis: 6.2x sample efficiency and 250x faster GenAI generationRedditJul 31, 2026 17:54ShareThis tweet cannot be embedded.Open direct link
