@bkuhn not sure I understand the hack. Is the hack a persuasive argument to convince FOSS community to copyleft LLM output? @RichardJActon @cwebber @ossguy @evan @kees

@sfoskett neither scotus nor afaik any other US court has held this. I would argue that it seems to be the direction the US legal system is going in, but I recently heard a persuasive counterargument from a well regarded US FOSS lawyer @evan @cwebber @bkuhn @ossguy

@evan oh I mean of course you could use LLMs to help with the analysis @cwebber @bkuhn @ossguy

@evan I feel pretty confident in saying the abstraction-filtration-comparison test cannot possibly be automated @cwebber @bkuhn @ossguy

@cwebber to be clear compliance cannot somehow be built in to the LLM for reasons you stated, but ancillary tools for LLM users to reconstruct provenance exist and conceivably could be made more useful @bkuhn @ossguy

@cwebber I think adequate compliance might be possible with good enough detection/matching tools but I don't necessarily expect such tools to be developed (let alone available to foss projects) (my assumption is that the few such tools in use today are pretty bad) @bkuhn @ossguy

@cwebber I mean, as a practical idea worth contemplating. Could imagine it as an experiment by someone with sufficient resources. There were some highly ill-conceived efforts to create anti-copyleft models a few years ago @bkuhn @ossguy

@cwebber copyleft-only LLM is nonsensical , agreed @bkuhn @ossguy

@cwebber I truly don't think this is a new situation @bkuhn @ossguy