Any idea how to render Latex equations nicely in Telegram?

[10bn]

I'm using the bot mostly for studying, do you guys have a suggestion hot render latex equations nicely in Telegram?

[chaserCN]

I've created a plugin that converts LaTeX to SVG using https://math.vercel.app/ and then converts the SVG to PNG in the code. You can check it out here: https://github.com/chaserCN/chatgpt-telegram-bot/tree/custom

[alexk-epsilon]

@chaserCN your plugin is great 👌

I did [x^2 + 2*x + 3 =0 ] and it rendered:

Smashing 👍

Then, I asked it to solve the equation:

solve [x^2 + 2*x + 3 =0 ]

and it gave me a long brain dump in raw latex:

The equation \(x^2 + 2x + 3 = 0\) does not have real solutions since the discriminant \((b^2 - 4ac)\) is negative. However, we can find the complex solutions using the quadratic formula. The quadratic formula states that for an equation of the form \(ax^2 + bx + c = 0\), the solutions for \(x\) can be calculated as:

[x = \frac{-b \pm \sqrt{b^2 - 4ac}}{2a}\]

In our case, \(a = 1\), \(b = 2\), and \(c = 3\). Substituting these values into the quadratic formula, we have:

[x = \frac{-2 \pm \sqrt{2^2 - 4(1)(3)}}{2(1)}\]

Simplifying further, we get:

[x = \frac{-2 \pm \sqrt{-8}}{2}\]

Since we have a square root of a negative number, we can rewrite it using the imaginary unit (i), where (i^2 = -1):

[x = \frac{-2 \pm \sqrt{8}i}{2}]

Simplifying the expression, we have:

[x = -1 \pm \sqrt{2}i]

Therefore, the solutions to the equation (x^2 + 2x + 3 = 0) are:
[x = -1 + \sqrt{2}i]
[x = -1 - \sqrt{2}i]

I wonder if there is a way to apply plugin to the model output as well...

[chaserCN]

I usually ask it "show the answer as an image" and it understands.

You can try "solve [x^2 + 2*x + 3 =0 ] and show the solution as an image"

[alexk-epsilon]

Yeah.

Not sure why it doesn't run latex() plugin on individual formulas in response to solve [x^2 + 2*x + 3 =0 ] though . It should be fairly obvious to the model that they're in latex, or are we missing something...

[alexk-epsilon]

Hi folks. I looked at various online tex-to-image converters but none seemed to be robust enough or easy enough to use to handle multi-line LaTex in a reasonable way. It is also worth noting that the model rarely outputs proper latex (but I firmly believe that it will improve eventually lol) So, I decided to capitalize on @chaserCN 's awesome idea to actually use local LaTex converter via sympy.preview()
The source is available at https://github.com/alexk-epsilon/chatgpt-telegram-bot/tree/local_latex
Here are some results:

@chaserCN thank you for the inspiration and please let me know if you want me to contribute this to your repo.