ChatGPT can already run code, which allows it to overcome some limitations of tokenization (eg counting the letters in strawberry, sorting words by their second

ChatGPT can already run code, which allows it to overcome some limitations of tokenization (eg counting the letters in strawberry, sorting words by their second letter). Doesn't seem like adding a Prolog interpreter would be all that hard.