Hacker News Reader

Wait. Based on the results of the test linked above you think this model might have been trained to produce it? Did you look at the results?!

That’s the problem. If all (or zero) models were trained on it, it would be fine as a benchmark.