Openai api does not write the equation in latex format in mathjax format

API
1

Hello, openai api does not write the equation in latex format in mathjax format. Even though I entered a long prompt examples about this, the prompt skips the examples. My prompt example:

For latex expressions, use the $ symbol for LaTeX formatting. For example, if an expression is given as \(\frac{24}{18 - x}\),\[\frac{24}{18 - x}\], rewrite it as $\frac{24}{18 -x}$. Writing math formulas: You have a MathJax rendering environment.

- Any LaTeX text between single dollar sign ($) will be rendered as a TeX formula;
- Use $(tex_formula)$ in-line delimiters to display equations instead of backslash;
- The rendering environment only uses $ (single dollarsign) as a container delimiter. Example: $x^2 + 3x$ is output for "x² + 3x" to appear as TeX.`

Apart from this prompt, I added a sample answer draft on the prompt.

Note: Out of 5-6 examples, only 1 is mathjax as I want. Not all of them work, others are in latex format.

2

I decided it wasn’t even worth the effort of cluttering up a system prompt, when I could just do the following on my end with the results:

    translateLaTex = (val: string): string => {
        if (val.indexOf("\\") == -1) return val;

        return val.replaceAll("\\(", "$")//
            .replaceAll("\\)", "$")//
            .replaceAll("\\[", "$$")//
            .replaceAll("\\]", "$$");
    }
3 Likes
3

The lesson is, as always, don’t fight the model.

3 Likes
4

Exactly. If you can easily conform to what the model wants to do naturally, then that’s far better than trying to instruct the model to behave differently.

1 Like
5

I found myself wanting to do this replacing, but with a stream. I haven’t tested much, but this seems to work:

def parse_stream_to_katex(stream: Stream):
    """
    Takes an OpenAI Stream and replaces ChatGPT LaTeX delimiters
    with KateX ones.
    Yields text, not chunks
    """
    last_text = ""
    for chunk in stream:
        text = chunk.choices[0].delta.content
        if text:
            # Sometimes delimiters like \( can be split over two chunks.
            # If the previous chunk ended in \, prepend that to this chunk
            if last_text.endswith("\\"):
                text = last_text + text

            text = (
                text.replace(r"\[", "$$")
                .replace(r"\]", "$$")
                .replace(r"\(", "$")
                .replace(r"\)", "$")
            )
            last_text = text

            # If the text ends in \, we don't return it, we'll get include it in the next chunk
            if not text.endswith("\\"):
                yield text