Nick on Nostr: nprofile1q…ufa4k I know that they're inherently bad at math (because they're just ...

nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqknzsux7p6lzwzdedp3m8c3c92z0swzc0xyy5glvse58txj5e9ztqaufa4k (nprofile…fa4k) I know that they're inherently bad at math (because they're just spitting out sequences of tokens that are probable based on their training set), but I thought some had implemented a kludge where when the LLM recognizes a math problem it writes code to solve it (something it *can* do, to a degree) and runs the code to get the answer to the problem.