Not surprised. In general, as far as I can tell, it deducts rules from its source material and when there are word language based descriptors of the numerical data. So it does better with scientific papers than asking it to look at a team’s mlb cal and give (eg) total season losses despite 6+ runs 🙃
That last example is just an example - but I tried to get it to do so, to no avail. I landed on your blog because I was interested in others’ experiences around baseball data and game rules with chatgpt!
I don't think I would expect it to do well with scores or predictions of scores or anything like that. However, it should be able to know the rules of the game.
Oh, this is not asking it anything predictive. Just simple math: add up the number of losses where the team also had at least 6 runs. Provided it date ranges and where to get the info but it yields lists of some but not all. Sometimes the result happens to only be for away games, mysteriously enough