So what about LLMs training on free and open content? "I don't believe the answer is rejection," writes Hong Minhee. "I believe it's reclamation." The core question here, says Minhee, is that of who owns the models? "Who benefits from the commons that trained them? If millions of F/OSS developers contributed their code to the public domain, should the resulting models be proprietary?" OK, I take the point. But here's the thing. Why do we assume we're 'one and done' with models trained on open content. The content isn't going anywhere. Anyone can train models on this content (unless, of course, we lock it down, which would simply preserve the status of the existing proprietary models). We act today as though there could only even be one Anthropic Code or ChatGPT. But that's absurd. There are already many many properly open source models. We will be drowning in them! Open content ultimately means genuinely open AI - if we allow it to. Via Ben Werdmuller.
Today: Total: [Direct link] [Share]
