This post documents the first time I used an LLM to assess the Physics Derivation Graph
All of the files in the git repo (tl;dr: Gemini crashed)
I uploaded all of the Physics Derivation Graph software and asked Gemini 2.0
The git repo has 96 files; 14 .py and 58 .html
$ find . -type f | wc -l 96
10503 lines of Python, 4337 lines of HTML, and 968kB
Uploading the files uses 230,346 tokens of an available window size of 1,048,576 tokens on 2025-01-09.
Here's the prompt I provided along with the files:
Result: Gemini crashed ("An internal error has occurred.") after 40 seconds of processing. Two more tries also didn't work.
Just the .py and .html files
Removed everything except for 13 py and 58 html files, down to 816kB.
Now the token count is 209,717
That prompt resulted in Gemini making 7 very reasonable suggestions. I opened two new issues on github as a result of Gemini's assessment.
In the same context I asked
The result from Gemini offered relevant text that I have included in the README.md file. The text is literally better than what I could have written!
Next I asked for a visualization,
That produced some almost-correctly-formatted ASCII art. The caveats provided by Gemini were very reasonable
"This is a simplified representation. The actual program flow involves many more specific states and transitions, not shown for simplicity."
Lastly I asked Gemini
Gemini's output was mostly HTML (some Markdown) and very useful.
No comments:
Post a Comment