Hacker News Reader

How do you know that's true?

the 1.5 million was our test set. we had 1.5 million ground truths, and it didnt make up fake data for a single one

That's not what I asked. I asked "How did you determine that it didn't make up/get information wrong for all 1.5m?"