@m-j-r.eth

big news, probably well-aligned. Nvidia is well-positioned to lead harness optimization and further distributed training of local models after the fact (and better-specified, enduring incentive). if the chips are maximally utilized in smaller, more academic batches beyond gigawatt datacenters, then there may be longterm demand for replacing those general-purpose chips to remain in the conceptual frontier. ultimately, the question is whether an openweights platform like huggingface is compatible with shrewd self-interest and whatever proprietary tech-on-a-chip paradigm takes over when lithography ramps up over the 2030s. https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/

1 reply

0 recast

4 reactions