Gary Sutton

So you saved a lot of money and time.

But, yeah.

But you didn't get the consensus that you were probably looking for or hoping for.

Right.

In my kind of world, in the data science world, I think that's especially true.

I was in a conversation a few weeks ago, and I sort of framed it as knowledge versus wisdom.

I think the language models provide the knowledge, but you still need human wisdom to really diagnose things.

I mean, if you were to deploy a model,

in production after you've trained it for however long.

And let's say the model doesn't perform in production the way it did during training.

who's going to detect that?

And then who's going to diagnose the problem?

And who's going to explain what's going on to executives?

And then who's going to apply the corrective action?

I still contend that those are going to be human-triggered activities.

Now there may be some AI interaction along the way, but I think they're going to be still triggered by humans.

The insight, there's still a human need for understanding the models,

Especially when a model doesn't perform as expected.

Like I said, who's going to detect that?

And who's going to understand why the model is performing now in production the way it did perform when it was being trained?