Math is extremely hard for LLVMs to do, since they don’t have calculators. They fundamentally pick the next best word (with literal random variance) given a sentence they already started creating, so that’s not something they’ll ever do well natively.

Openai has talked about integrating math plugins to help it perform these tasks, but it’s completely different from what it’s good at

As an AI language model trained by OpenAI, I am in this is photo and I don’t like it.