Hacker News Reader

Comment by andersource | Hacker News Reader

Most of the time, but it can, and I think that's the issue a lot of people have with it.

https://twitter.com/mitsuhiko/status/1410886329924194309

https://news.ycombinator.com/item?id=27710287

I don't understand why anyone has an issue with that. You know what else can copy and paste code all the time? Humans. But we have various ways of stopping employees who copy and paste code from stackoverflow and github without checking the license, so it's the same thing if you use one of these tools. There's nothing new I can see here to be upset about.

This would be a lot more interesting if it showed the various GPT-3 experiments at generating music and used that as a point of comparison.

> But we have various ways of stopping employees who copy and paste code from stackoverflow and github without checking the license

What would those be? I’ve worked at a number of organizations that were (rightfully) paranoid about accidentally incorporating GPL code, but even there I wasn’t aware of automated tooling to prevent it, it was only enforced through developer vigilance.

If you actually want a paid service, there are plagiarism detectors like Fossa and Codequiry. Although in my opinion, code review should be enough to catch any "accidental" incidents of plagiarism, the differences in writing style should make it very obvious when the employee has copied something. That of course probably won't apply if you suspect the employee is intentionally changing it around to obfuscate the origin of the code, but it seems that wouldn't be the case if they were just committing the output straight from a neural net. But automated scanners probably won't be able to catch those well either -- the way to catch that would be to make them do pair programming a lot.

>Although in my opinion, code review should be enough to catch any "accidental" incidents of plagiarism, the differences in writing style should make it very obvious when the employee has copied something.

You must do some CSI level code reviews. Best I'm able to do is figure out if code will work and if something can be done obviously better. Stylistic calls (beyond lint enforceable) are up to authors as far as I'm concerned.

And even then it's trivial to fix up naming schemes and such to march codebase - doubt that gets you out of copyright issues.

I mean in cases where someone just copy and pastes something without making any effort to match the style, or in cases where they can't explain what a piece of code does or how they came up with it. You should be able to spot those very easily in code review. If somebody is trying to fix up the naming schemes to avoid being detected and for whatever reason is able to explain the code perfectly, then I'd imagine that person would probably be doing the same bad things regardless of using copilot -- it's not like it's hard to search stackoverflow and github for code snippets.

But I think copilot matches code style, no ?

This fast inverse square root function is very well known, with even a Wikipedia page, and it is more than 20 years old. My country doesn't have software patents but it seems that the standard duration of a software patent is 20 years, so even if this function was patented, the patent would have expired by now.

Can a few lines of technical code hold a copyright for that long?

Copyright and patent are different. Also, while you can’t copyright an algorithm, your specific source code that implements it is copyrighted (assuming it’s sufficiently original).

In this case it’s not implementing the algorithm, it’s copying a particular famous implementation, down to the comments.

There is no real reason for copyright terms to exceed patent terms.

(And FWIW, patent terms should be inversely proportional to the number of patents issued in that category the previous year. This would automatically reduce terms in categories where innovation is rapid, promoting competition and drive to get to market, but preserve maximum protection for inventions in mature categories with a slower pace of innovation.

> This fast inverse square root function is very well known, with even a Wikipedia page, and it is more than 20 years old.

The Copilot algorithm didn't take this into account though... if it did, we probably couldn't complains about copyright infringement.

This truly is the engineer's disease. Hundreds of incredibly strong opinions about the legal system derived almost entirely from a few tweets and zero experience outside of software engineering.

Copilot is neat. If you are concerned about it, talk to a lawyer and get their opinion.

Reply on news.ycombinator.com