Open adamboazbecker opened 6 months ago
I'm not sure I quite understand the question. Are we talking about acceptance criteria for a finished model, or resource requirements going in to the development process, or something else?
Today I heard a story about a company that spent a good chunk of time trying to sort out edge cases with their evaluation test set only to find that when there was a new model release these problems went away.
I know it doesn't really answer the question and wait til there is a new model released but it does feel like there should be a way to game out what could change in the next 6 months that would turn everything we are working on up side down
How do we best specify requirements given the massive action space of Generative AI?