6 ms·
What is the licensing for code generated in this way? GPT-3 has memorized hundreds of texts verbatim and can be prompted to regurgitate that text. Has this mode
by iandanforth 5y ago
What is the licensing for code generated in this way? GPT-3 has memorized hundreds of texts verbatim and can be prompted to regurgitate that text. Has this model only been trained on code that doesn't require attribution as part of the license?
- richardanaya 5y agoYou're using the word "memorized" in a very loose way.
- delaaxe 5y agoHis point still holds, GPT-3 can output large chunks of licensed code, verbatim
- tsbinz 5y agoHow is it loose? Both in the colloquial sense and in the sense it is used in machine learning it is fitting. https://bair.berkeley.edu/blog/2020/12/20/lmmem/ https://bair.berkeley.edu/blog/2020/12/20/lmmem/ is a post demonstrating it.
- mempko 5y agoFinally, a faster way to spread bugs than copy/paste.
- Abimelex 5y agoNot sure how this is handled in US, but in Germany a few lines of code have in general not enough uniqueness to be licensed.
- hakre 5y agoSo the corpus has been compiled under license and the derivative work is eligible for distribution?
- billti 5y agoThe landing page for it states the below, so hopefully not too much of an issue (though I guess some folks may find a 0.1% risk high). > GitHub Copilot is a code synthesizer, not a search engine: the vast majority of the code that it suggests is uniquely generated and has never been seen before. We found that about 0.1% of the time, the suggestion may contain some snippets that are verbatim from the training set.
- jgworks 5y agoI'd be surprised if a company's legal department would be OK with that 0.1% risk.
- TedDoesntTalk 5y agoLike wouldn’t be entertaining without the License Nazis. No code for you! (Seinfeld reference)
- viraptor 5y agoGoogle already learned that one. "There's only a tiny chance we may be copying some public code from Oracle." may not be a good explanation there.
- varispeed 5y agoDid they have a license to use public source code as a data source for data set though?
- iandanforth 5y agoUpdate: Nat Friedman answered this as part of this thread on twitter: https://twitter.com/natfriedman/status/1409883713786241032 https://twitter.com/natfriedman/status/1409883713786241032 Basically they are building a system to find explicit copying and warn developers when the output is verbatim.