How to Evaluate a Machine Learning Model

When we're building these systems, especially if it's like an interactive system like a human dialogue system, or even building any machine learning model, we sort of use this gradient descent. How do we have an idea about the correlation between these intrinsic evaluation methods and later extrinsic evaluation? Like our good evaluation metrics in this initial intrinsic evaluation, is that a good proxy for later evaluation?Yeah, that's a very good question. And I that's what it should be. What you really need to do is to rely on less costly, not involving like non-human evolving evaluation metrics, just so that you can actually get better performance from your model. So once you do, then you