Hello ! I’m using the GPT4 API for an educational application. I would like gpt4 to be able to generate mathematical equations or mathematical expressions in latex format. But despite the prompt, it sometimes generates mathematical answers in plain text. How do I get what I want in a stable way?
Welcome to the community!
Have you tried to instruct the model to to start all mathematical equations with the $ or $$ delimiters?
the latest gpt-4-turbo models have all been tuned to conform to a gfm-like markdown style, so leveraging these delimiters might help it get into the mode you want it to be in.
I tried it but sometimes the model does not follow these instructions. In other words, it’s unstable.
Here’s a higher quality of instruction I just posted for LaTeX.
Which GPT model is more accurate & efficient for latex & other education purpose?
gpt-4- turbo preview
gpt-4-0125-preview
gpt-4-0125-preview
gpt-4-0613
I just say “Write the answer in Latex”. Or “Write a latex code”.
these are all the same model lol
but yeah, go for that if you’re just looking for formatting consistency.
Time and alterations by OpenAI to make faster, more chat-trained, more untrainable AI gives you this progression of quality, where the latest version is a stinker, trying to write ChatGPT-renderer backslash formatting into everything TeX, ignoring instructions when it is not denying:
Best to “cheap”
gpt-4-0314
- A breath of fresh air to return to - if you still have accessgpt-4-0613
- (includes function calls) more resistant to your output specs like writing HTML [“gpt-4
”]gpt-4-1106
(includes tool calls and JSON mode that are broken, larger input context but output limitations; will tell you to write the code yourself)gpt-4-0125
- still lazy, and also bad context-reading quality and vapid answers [“gpt-4-turbo
”]
Thank you.
What is the best model for questions and answers that I can create, please?
Which model—the one mentioned above or another?