Hacker News new | ask | show | jobs
by ddtaylor 130 days ago
I look at this pretty similar to licenses that attempt to say something can't be used in some specific way, such as a "no evil" license. Famously, the JSON license:

    Permission is hereby granted, free of charge, to any person obtaining a copy of
    this software and associated documentation files (the "Software"), to deal in
    the Software without restriction, including without limitation the rights to
    use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
    the Software, and to permit persons to whom the Software is furnished to do so,
    subject to the following conditions:

    The above copyright notice and this permission notice shall be included in all
    copies or substantial portions of the Software.

    The Software shall be used for Good, not Evil.

    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
    FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
    COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
    IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
    CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
(Source: https://www.json.org/license.html)

This seemingly pointless yet straightforward addition of the license has caused problems, because it's highly subjective and therefore makes compliance with the license impossible to objectively measure - which is really important for a license!

I think LLM training seems right now a simple objectively measurable thing - and maybe it will always be that simple - but I could also see it becoming a subjective thing. At the very least the interpretation of copyright law has yet to be upheld, which historically was one of the most "powerful" license-like things in existence.

1 comments

How do you objectively measure if an LLM is training on something if you don't have access to its training data?
In theory the same way people are making those claims about "stolen" art, such as models that produced watermarks from Getty images or Shutterstock. Similar "watermarks" have existed in some LLM output.