Reverse engineering images of graphics to data and Wolfram code using LLMs

[danielscarvalho (Daniel Carvalho)]

Open in Wolfram Cloud ↗

DOWNLOAD-DESKTOP-NOTEBOOK.nb (670 KB)

[okazakikotaro (Kotaro Okazaki)]

Daniel-san,

Thank you for your interesting post. I have a similar interest.
I reverse-engineered graphics in Wolfram Language about eight years ago.
I reverse-engineered graphics in LLM about nine months ago.

GraphImage2List: get numeric data from plot's images using image processing

I’m trying to do a Machine Learning using old data. However, the data are not numeric-data but graph-image-data. So I create the tool which calculates numeric-data from graph-image-data. Goal The left is the original graph-image-data. The right is the graph made from the tool’s output(list data). [5699screenshot.1.jpg] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 3…

[wolframcommunity (EDITORIAL BOARD)]

– you have earned Featured Contributor Badge
Your exceptional post has been selected for our editorial column Staff Picks http://wolfr.am/StaffPicks and Your Profile is now distinguished by a Featured Contributor Badge and is displayed on the Featured Contributor Board. Thank you!

[danielscarvalho (Daniel Carvalho)]

Great job Kotaro-san! Thanks for sharing!!

I hope soon the LLMs will be able to do all the “dirt job” with more precision!! Reverse engineering graphic data is a key tool in research!

[okazakikotaro (Kotaro Okazaki)]

I hope so, too, because the evolution of LLM is at warp speed. I had the opportunity to present this at the Wolfram Virtual Conference Japan 2024.
With the current multimodal LLMs, there are still differences in recognition accuracy, and I think it is necessary to use different types of LLMs.