Hacker News Reader

Comment by abrookewood | Hacker News Reader

I don't think the Open Source claim is accurate. From their repo "Magicoder models are trained on the synthetic data generated by gpt-3.5-turbo-1106 developed by OpenAI. Please pay attention to OpenAI's terms of use when using the models and the datasets."

View on HN

OpenAI behaves as though everyone else's outputs are fair game for training LLMs, including GPT, but doesn't want others using GPT's output?

So, when OpenAI does it, it's transformative, but when we do it, it's not?

That's not right, and I don't think the courts will rule in their favor.

> That's not right, and I don't think the courts will rule in their favor.

Especially if you consider that all of OpenAI code training data comes from open Github repositories.

Yes, I totally agree. They are more than happy to make money off the backs of copyright holders.

> Please pay attention to OpenAI's terms

Just because people at a company tell you how to behave doesn't mean you need to comply

Or you can do the right thing and avoid distilling that stinky model, like they’ve asked.

Does the fact someone asks you not to do something inherently make it the right thing? Especially when what they've asked is hypocritical?

What makes it the "right thing" and who decided it was right?

All model outputs (text, image, video) are considered public domain in the US, so they can be used for whatever.

It's not true. Compiler output is not public domain. Why LLVM output should be different? Both ingests some and output some code as result.

Compiler output isn’t owned by the compiler writer, but the compiler user usually.

Not sure why you brought up compiler output, I'm talking about generative machine learning models.

Reply on news.ycombinator.com