Currently, in AI training, what is called a general-purpose large model needs to anticipate pretty much every use, from folding our proteins to folding our laundry, in the same model. In doing so, it’s incredibly energy inefficient to train. However, when we know what we want the model to do, for example, folding the proteins or folding the laundry, then we can train what is called a domain-specific model or a local model that incorporates the local community’s input in such a way that also protects their data from extraction to the clouds of foreign big tech companies.