ChatGPT incorrectly put a space between the as and bs, but if we let that slide, there's still the issue that the best trained model in the article got 77.3% of

ChatGPT incorrectly put a space between the as and bs, but if we let that slide, there's still the issue that the best trained model in the article got 77.3% of the first 1500 strings correct, i.e. even if ChatGPT performed exactly the same, you'd expect it to get a single example correct more often than not.