3 ms·
Nothing other than Microsoft's attempt to stop AI from learning repo code.
by doublerabbit 3y ago
Nothing other than Microsoft's attempt to stop AI from learning repo code.
- 1231232131231 3y agoI doubt they would use the code search feature instead of just cloning the repo and feeding it to the AI.
- doublerabbit 3y agoI don't doubt it. If you wished to build a ML bot that taught switches/loops; instead of cloning every repo all you'd need to do is search for switches/loops within X lang.
- yodon 3y agoVery inefficient way to do data access for more realistic training objectives.
- doublerabbit 3y agoWhy else do you use search? If not to find a specific piece of code.
- yodon 3y agoHuman workflows are very different from large scale LLM training or augmenting workflows
- beeboobaa 3y agoThat would be ridiculously slow and inefficient.
- doublerabbit 3y agoAnd downloading a blind repo isn't?
- beeboobaa 3y agoMuch less so. It can easily be parallelized without running into rate limits of the proprietary search API. Then once you've cloned it once it's on your system and you won't have to talk to github again. So much more sensible.
- doublerabbit 3y agoSensible isn't a thing when it comes to malicious activity. Same could be used to cheat the search. Downloading a repo and than neither knowing the repo has the syntax you wish to learn is still going to be more resourceful than having a webpage thrown at you with all search results.
- dontupvoteme 3y agoThey probably want to use user info on what they search for, what they use, etc to better filter between good/bad popular/not-popular code for finetuning things. If people can search anonymously it gets a lot harder to datamine
- odyssey7 3y agoRetrieval-augmented generation could be the specific AI use case they’re guarding against if they’re restricting search but not cloning.