all 1 comments

[–]neolib 2 insightful - 1 fun2 insightful - 0 fun3 insightful - 1 fun -  (0 children)

As I understand LLMs are bad with arithmetic and questions about letters in words because of tokenization (random google result: https://docs.mistral.ai/guides/tokenization/ - "process of breaking down text into smaller subword units, known as tokens.").

If you didn't try that for testing and really need to do calculations use Wolfram Alpha (by the creator of Mathematica): https://www.wolframalpha.com/ - it can do integrals and charts even.

Wikipedia entry: https://en.wikipedia.org/wiki/WolframAlpha