LLM Hacker News Mar 20, 2026 2 min read
Q Labs says 100M tokens and an 18B-parameter ensemble can match a 1B-token baseline, and Hacker News immediately focused on whether that gain survives serving and deployment.
#scaling-laws
RSS FeedLLM Hacker News Mar 5, 2026 2 min read
A March 4, 2026 Hacker News thread elevated Q Labs’ Slowrun benchmark, which fixes training data at 100M FineWeb tokens and optimizes for data efficiency under large compute budgets.