I am working on a business to provide support for independent musicians, and I would like to use this and this to essentially copy and paste this information as training data (I’m planning to use a fine-tuned GPT 3.5 turbo). It is important to note that I am not going to use this as code, but as written information for a large language model (essentially I want to avoid logging all this information myself).
This library has a MIT license, but I’m unsure how it applies to this specific use case. I eventually want to use this AI to be part of a for-profit website. This AI will be available to the consumer for free (the website will run off ads), so I won’t be making money directly from this. I’m pretty new to all of this, and there seems to be a lot of confusion regarding the use of licensed content in artificial intelligence, so I am rather confused when it comes to the use within a LLM.
Here is the library itself.
There are two possibilities here:
- The outputs of your LLM are not a derivative work of its input data - this is essentially what is being claimed at a larger scale by OpenAI, Meta, Google et al. In that case, the license of whatever input data you used has no bearing on what you can do with your model, so you can use it commercially (without legally needing to credit the library, but you can of course still do so if you wish).
- The outputs of your LLM are a derivative work of its input data. In that case, you will need to follow the MIT license when using them, but all that requires is for you to credit the library in the way specified in the license, so you can still use it commercially.
As of November 2023, there is just about zero clarity as to which of these two scenarios applies; multiple court cases in multiple jurisdictions are already underway and we can expect many more in the next couple of years.