AI and Mathematics Part 2

I recently wrote about the fact that AI is great in general conversation but still poor in mathematics.

But even this is changing incredibly fast:

1. Microsoft has proposed MathPrompter, a technique that improves the performance of large language models on arithmetic problems along with increased reliance on predictions. It uses the trick of generating multiple algebraic expressions or Python functions to solve the same math problem in different ways, thereby raising the confidence level in the output results.

https://www.microsoft.com/en-us/research/publication/mathprompter-mathematical-reasoning-using-large-language-models/

2. GPT-4: While its predecessor, GPT-3.5, scored above the 80th percentile on exams for courses like Environmental Science and Psychology, GPT-4 surpassed that threshold for a number of other exams. Have a look at the very left bar: in Calculus, it shot up from zero to the 43rd percentile!

https://openai.com/research/gpt-4

3. Wolfram, the famous computer algebra system can be accessed from within GPT using a ChatGPT plugin.

https://writings.stephenwolfram.com/2023/03/chatgpt-gets-its-wolfram-superpowers/

And this is just what has happened since I wrote my first post 2 weeks ago!