The Fundamental Limits of LLMs at Scale | Dark Hacker News