Hacker News new | ask | show | jobs
by argee 12 days ago
That’s the problem. If all (or zero) models were trained on it, it would be fine as a benchmark.