genai, alibaba "stealing" from anthropic, general musings
I do feel kind of good about Anthropic being upset that they are being "stolen" from. Because I like these companies having a bad time, and also because Alibaba has made all(?) of their models be open-weights, which is, technically speaking, slightly more beneficial than fully proprietary models... but of course still not nearly enough to call the Qwen models "ethical". Alibaba provides no information about the datasets they use by volume, no crawler name or opt-out protocol (which is also backwards, it should of course be opt-in if anything), no specific licensed data disclosures or anything like that.But yeah, I think open weight models are interesting because when the bubble pops, proprietary models will vanish. Open weight ones would at least remain, and you can also perform post-training on them to make them useful for specific tasks. Gemma4 E2B for example is a very smol model with vision capability which idk, that might be useful even after a bubble collapse.
Still not what I'd like to see at all though, don't get me wrong. AI development should, if done at all, be consolidated, done in a careful and science-driven manner (especially: social sciences to structurally eliminate bias), without state or capital interests involved, and done in a way that is both fully reproducible and also actually feasible to be used by the wider populace without requiring everyone to buy new hardware. For that to happen, we'd need a revolution or two, though
