The first post in what I hope will become a long thread to try to improve the mathematical abilities of Gemini. 1) Broken mathematical notation Gemini lags behind ChatGPT in its ability to correctly format Mathematics, a thing ChatGPT has been able to do since day one. This is putting off several …

We can push for this together :laughing: But I have a feeling that some absolute Gigachad developer that is currently working as a product lead for Google, will see this and think “yes, this sounds like a good idea” and fix it, so here’s a few directions on how: Scan the incoming token steam for \…

On the topic of mathematics proper, I tried out the prompt Suppose g is a continuous smooth function such that for every x > 0, there is one and only one y > 0 such that g(x) + g(y) ≤ 2xy. Can you prove that g(x) = x^2? on gemini-1.5-flash-latest to compare it to what gemini-1.5-pro-latest had to …

Here is how ChatGPT-4o answers. I think it’s good to see how other models compare to understand Gemini’s strengths and weaknesses and room for improvement. [Screenshot 2024-05-15 at 13.30.26]

Clear and comprehensive. And, best I can tell, GPT-4o didn’t even have to use Wolfram to generate the proof.

It gets worse. I’ve been feeding gemini-1.5-flash-latest increasingly simpler problems and it has been failing to solve them (in an attempt to find the applicability frontier for the model). Finally I got to the 10-year old stuff. Prompt We are given the sequence of numbers 2, 5, 10, 17. Which of t…

Wow. This would definitely deserve a read from someone at Google. GPT-4o again for comparison. [Screenshot 2024-05-15 at 17.12.19]

Answer from Gemini 1.5 (I assume the new May version). [Screenshot 2024-05-18 at 16.17.24]

Yes, Gemini 1.5 Pro (and Gemini Advanced for general use) get it right. It’s gemini-1.5-flash-latest that is the mathematical lemon.

New test: elementary optimal control problem. [Screenshot 2024-05-19 at 10.37.33] Gemini 1.5 Pro gets it right; [Screenshot 2024-05-19 at 10.44.03] ChatGPT-4o gets it only partially right. [Screenshot 2024-05-19 at 10.44.47]

Gemini 1.5 Pro is capable, Gemini 1.5 flash either solves a math problem but can’t explain it as well as the bigger brother, or (more often than not) just can’t solve it. And there are examples where Gemini 1.5 Pro can do even better than GPT-4. Try this problem, it is one of many that hasn’t been …

A thread on Mathematics

Gemini API

maisonhai3 December 22, 2024, 9:06am 24

The 2.0 Flash & 2.0 Advanced have the same problem with mathematic equations.

Topic		Replies	Views
Gemini Pro unable to transcribe text in images Community feedback	12	873	May 9, 2024
Solving problems using Gemini AI Gemini API gemini	2	234	October 15, 2024
Gemini-1.5-pro-latest performs WORSE since yesterday. How to use its previous version? Gemini API	35	1186	September 2, 2024
Enhancement of LaTeX Parsing Logic via Whitespace Delimitation Gemini API feedback , bug , api , models , gemini	3	115	December 24, 2025
Gemini 2.0 flash - 1.5 pro Struggles with Basic Task Execution Gemini API gemini-15 , api , models	1	141	May 19, 2025

A thread on Mathematics

Related topics