Greg Hogan <[email protected]> writes:
On Thu, Jul 9, 2026 at 2:52 PM Ludovic Courtès <[email protected]>
wrote:
Hi,
Steve George <[email protected]> skribis:
> I don't think there's any circumstance where we'd have to rip
> out LLM code from
> the archive.
One case is that of copyright infringement where the LLM
regurgitated
code that is proprietary or published under an incompatible
license.
From what I’ve seen (references are in the GCD and in its
previous
versions), this scenario is not science-fiction.
As discussed previously, the scenarios presented were targeted
extraction on old models. But it should be easy to demonstrate
incidental regurgitation for a model which would be used here,
right?
I'm sorry, but this burden of evidence sounds backwards to me.
If I made a consumer device that caught fire in many
well-documented cases, it would be reasonable to expect me to
demonstrate how the problem has been fixed. It would not be
reasonable to ask each burn victim to prove why their experience
indicates a systemic problem.
Jason
