A good reason why "model transparency" isn't as meaningful as it seems. Give us ways of evaluating systems, don't bury us in details of how individual pieces function.
Gentle reminder that ChatGPT is not a "model" ("foundation" or otherwise). It's a larger system, likely composed of multiple different models and traditional coded rules. Terminology is important, and breakdowns like these can happen when we lose clarity.
Nov 10, 2023 · 10:31 PM UTC