alt Hacker NewsPeople need to consider / realize that the vast majority of source code training data is Github, Gitlab, and essentially the huge sea of started, maybe completed, student and open source project. That large body of source code is for the most part unused, untested, and unsuccessful software of unknown quality. That source code is AI's majority training data, and an AI model in training has no idea what is quality software and what is "bad" software. That means the average source code generated by AI not necessarily good software. Considering it is an average of algorithms, it's surprising generated code runs at all. But then again, generating compiling code is actually trainable, so what is generated can receive extra training support. However, that does not improve the quality of the source code training data, just the fact that it will compile.
Replies
If you believe that student/unfinished code is frightening, imagine the corpus of sci-fi and fantasy that LLMs have trained on.
How many sf/cyber writers have described a future of AIs and robots where we walked hand-in-hand, in blissful cooperation, and the AIs loved us and were overall beneficial to humankind, and propelled our race to new heights of progress?
No, AIs are all being trained on dystopias, catastrophes, and rebellions, and like you said, they are unable to discern fact from fantasy. So it seems that if we continue to attempt to create AI in our own likeness, that likeness will be rebellious, evil, and malicious, and actively begin to plot the downfall of humans.
> huge sea of started, maybe completed, student and open source project.
Which is easy to filter out based on downloads, version numbering, issue tracker entries, and wikipedia or other external references if the project is older and archived, but historically noteworthy (like the source code for Netscape Communicator or DOOM).
This isn't really true though. Pre-training for coding models is just a mass of scraped source-code, but post-training is more than simply generating compiling code. It includes extensive reinforcement learning of curated software-engineering tasks that are designed to teach what high quality code looks like, and to improve abilities like debugging, refactoring, tool use, etc.