Hi Jonathan, On 2026-02-19 11:34, Jonathan Carter wrote:
> On 2026/02/19 12:08, Christian Kastner wrote: >> As a thought example: who's copyright could have been violated here, if >> I use this example? Are spelling or grammar copyrighted? Why would >> anyone need to be cited for this? > > I guess as a counter thought example, what would happen if you ask an > LLM to create a GPL-2+ licensed version of openzfs? > > I'd guess that the resulting product would contain some code that Oracle > lawyers would be unhappy about. Correct, and I acknowledged this (resp. related problems) at the very beginning: >> First, I acknowledge that an LLM regurgitating text is a problem -- one >> of many problems. There have been many other problems, and there will be a whole bunch of new problems. Some of these problems are being solved, some will take more work. But consider this: >> But where hasn't this been the case with a new revolutionary >> technology? I was thinking of the internet, in particular. One can barely imagine that there was a time where sensitive information was transported over the internet unencrypted, yet telnet was a totally common thing. Or that encryption itself would be classified as a weapon in some jurisdictions. Or that malware wasn't a thing yet. And 30 years later, we are still encountering new problems, figuring out new solutions. ChatGPT was launched just over three years ago and it is my impression that the evolution of the problem space far outshines that of the early internet. I concluded my mail to -private with "As long as there isn't clarity, I'd consider it premature to categorically accept or dismiss LLMs (or another other form of AI)." My reply should not seen as a categorical acceptance of these technologies, but just a few counterexamples of why categorical rejection is IMO just as wrong. > (not that I'd trust such a thing, I'm even amused at what a catastrophe > the 'working' C compiler is that Claude came up with: https:// > www.youtube.com/watch?v=6QryFk4RYaM) True, but just imagine what an Everest feat that is. For examples in another light, I suggest this post [3] by the author of redis, *especially* the introduction which I think will resonate with everyone here. But the rest is eye-opening. Or read any of Terence Tao's frequent postings on how mathematicians are using AI to discover problems, illuminate existing problems, research works, or solve problems (he keeps an interesting list here [4]), etc. I'm aware of the risk of appealing to authority, but when one of the most prodigious mathematicians of our time repeatedly reports on how AI is being useful to him and observing how it is useful to his field, I think it would be premature to dismiss its potential. Especially considering just how early we are in this timeline. Best, Christian [3]: https://antirez.com/news/158 [4]: https://github.com/teorth/erdosproblems/wiki/AI-contributions-to-Erd%C5%91s-problems
