4 ms·
Where will training data come from for new tech & programming languages if SO dies?
by bfung 1y ago
Where will training data come from for new tech & programming languages if SO dies?
- tevon 1y agoAll the same places it comes from for human programmers before a language has many answers on SO. - Documentation - Open source projects using it - Github issues - Source code - Blogs - Youtube videos The list goes on
- bfung 1y agoTo add a bit more nuance, SO has a question-answer type format, which leads very well into prompt-rely format to train these chat applications. Most of the other sources do not, except for Github issues maybe. Without this question-answer format, there'll be a need for a bigger data labeling effort to train LLMs on new stuff, no?
- msluyter 1y agoThis was my question. There's a weird sort of self-cannibalism that this hints at. The LLM is only as good as it is because it's been able to train on existing SO answers. But if over time, SO content production declines, then the LLM results will be less reliable. It seems that a new equilibrium could be one in which -- for newer questions/concerns -- both SO and LLMs will be worse than they are now.