Ask HN: How can an avg enterprise make use of LLMs without IP issues?

[tanmaydesh5189]

Many enterprises are worried that if they use pre-trained models to improve their organization's productivity, the training data used may be copyrighted. The other options could be using your own data/code reops to train these models, but if you don't have rich high-quality data/code repos then the model trained will produce suboptimal results anyways. Is there any better solution?

[unstatusthequo]

Why is the OpenAI API not adequate? Based on their documentation it does not keep / use data via the API. If that’s still not good enough, use offline models based on Vicuña or Llama. There is so much fearmongering that some will be left way behind due to over-cautiousness. I suppose if OpenAI is lying about API calls, then they are going to have a bad time. Seems like a huge risk to take on their behalf.

[Spooky23]

They should be - it’s nuts for a big org to do anything significant with this stuff until there’s some litigation settled.

The safe answer now is to consume services that use this tech and use indemnification and contracts to protect the business.