Hacker News Reader

Hacker News

Comment by bugfix-66 | Hacker News Reader

Parent Full thread

bugfix-66

It's the BSD 2-clause license, a very old, very widely-used, and very respectable open source license:

https://en.m.wikipedia.org/wiki/BSD_licenses

We've added one additional clause EXPLICITLY PROHIBITING use in systems like Copilot.

There's no ambiguity. Training your language model is a direct and unequivocal violation of clause 3.

If this is not adequate to prevent Copilot-style intellectual property theft then nothing is.

How else can we protect ourselves from Copilot-style use without attribution?

View on HN

Yes, the original license was open source. It's exactly your change that makes it not open source.

Please elaborate on this surprising claim, in detail.

Here is the OSI definition of Open Source:

https://en.m.wikipedia.org/wiki/The_Open_Source_Definition

Sure. Here's a quote from the Open Source Definition:

> No Discrimination Against Fields of Endeavor

> The license must not restrict anyone from making use of the program in a specific field of endeavor. For example, it may not restrict the program from being used in a business, or from being used for genetic research.

And here's the license term in question:

> Use in source or binary forms for the construction or operation of predictive software generation systems is prohibited.

I see, it's clause 6 in the OSI definition:

  6. No discrimination against fields of endeavor: The license must not restrict anyone from making use of the program in a specific field of endeavor. For example, it may not restrict the program from being used in a business, or from being used for genetic research.

So it's one or the other. Basically, Copilot makes all open source software (according to the OSI definition) necessarily use-without-attribution. Your code is my code. Your work is instantly my work, and no credit will be given.

I think this is a special case, illustrating that these new language models require us to reformulate our conception of what "open source" and "fair use" entails.

The laws and definitions need to adapt to a changing world.

> Basically, Copilot makes all open source software (according to the OSI definition) necessarily use-without-attribution.

No, Copilot is violating the existing open source licenses by not providing the attribution that they require. If Microsoft doesn't care about that, then they won't care about violating your license either.

I think that's misunderstanding of what the Open Source Definition refers to. The way I read it, you must be allowed to run the binary to further any goal.

But it doesn't mean you can share the source code however you want. Most FOSS licenses add restrictions on that, that's why there are so many of them. To be fair OP might need to reword a little to clarify that you can use the binary to accomplish training but doesn't allow you to create derived works of the source or binary by training on them.

Reply on news.ycombinator.com