Quote:
Originally Posted by Frenzie
Either you're good at cleaning it up or the AI slop has gotten much less sloppy.  |
It's a mixture of a bunch of stuff.
Gemini has really improved.
Having clear and verbose prompts is important, in this case, I pointed at relevant source code bases for inspiration.
Looking at the code occasionally helps, even if I'm not lua-fluent to give the agent prompts.
Telling the agent to tidy up/refactor the code for simplicity at the end really helps.
In other news, new version has been released with UI refactor, high score, save and load:
https://codeberg.org/anserina/2048.koplugin