you will need to collect data for every decision/usecase and then train a model. But this can be used for different use cases with just a prompt. Founders resp

you will need to collect data for every decision/usecase and then train a model. But this can be used for different use cases with just a prompt. Founders response to a similar question on X: https://x.com/CompleteSkeptic/status/2100067328620896408?s=20 https://x.com/CompleteSkeptic/status/2100067328620896408?s=2... pasting it here: zero-shot + general == programmable I would assume any extreme scale narrow task could then be fine-tuned for, but we'll see - I suspect putting it all in shared cognitive core has bit maintainability/generalization benefits