LLMs learn backwards, and the scaling hypothesis is bounded. [D]By /u/preyneyv / April 12, 2026 submitted by /u/preyneyv [link] [comments]