Hacker News Reader

Comment by todd3834 | Hacker News Reader

I wish I understood the commercial license. If you can’t host it, and they don’t offer it through an API, how can you use it commercially?

View on HN

> The use of the Work or Derivative Works to provide applications and integrated end user products which use the Work or Derivative Work in the background shall not be considered Hosting Use.

It almost sounds like they just want to avoid something like an AWS LLMA service to spin up off their hard work but building an app around it would be fine.

However, they have an Acceptable Use Policy that they can update at any time, and then you have to discontinue existing use of the output to match the update AUP, so there's no way to use the commercially (unless you have a backup plan, or decide that you're going to ignore the license).

You can use it if it's part of an application that uses the model. Not where you just wrap it around an HTTP API as a mere intermediary to the model.

That is not quite what they state, unfortunately.

Their language is a bit more suspicious:

Clause 9.1 - Subject to section 9.2, "Hosting Use” means any use of the Work or a Derivative Work to offer shared instances or managed services based on the Work, any Derivative Work (including fine-tuned versions of a Work or Derivative Work) to third party users in an inference or finetuning API form.

Well, if I send a message to my chat application "How to reset my password?" and it gets sent to the processing pipeline which gets a response out of Falcon LLM, is this not technically an "inference" context? I did provide access to their model, the model did perform inference, the user did get a result and the bot is a "managed service based on the Work".

This might be very problematic for commercial use.

Reply on news.ycombinator.com