On Monday, May 19, 2025 02:51:37 AM Martin Steigerwald wrote: > Ingo's suggestions to train a model with KDE code might help. But of > course to train a model will need a huge lot of resources according to > what I know.
From the peanut gallery: I know very little about training an AI, but, from what I've heard, training on the "bigger" AIs is based on bazillions of scanned data. I would think that training an AI for KDE only, and scanning only KDE code might be a few orders of magnitude simpler and less expensive. (I have the vague idea that, at least for some LLM models, the input has to be sort of "parsed" after scanning, and maybe initially by humans, but surely one of the first things that AI researchers would do is train their AIs to do their own parsing.)
