WEBVTT

00:00:00.000 --> 00:00:02.760
 We need to think about gen AI models the same way we do our own

00:00:02.760 --> 00:00:06.080
 employees capabilities where we employ a variety of skill

00:00:06.080 --> 00:00:06.520
 levels.

00:00:06.520 --> 00:00:10.560
 We cannot hire the most expensive people and give them the most

00:00:10.560 --> 00:00:13.280
 menial tasks because they will refuse to do the job and

00:00:13.280 --> 00:00:16.400
 certainly we won't be able to operate our business effectively.

00:00:16.400 --> 00:00:17.800
 But that's exactly what's happening.

00:00:17.800 --> 00:00:20.240
 You think as a developer, well, why would I use the model that

00:00:20.240 --> 00:00:21.320
 doesn't work as well?

00:00:21.320 --> 00:00:24.320
 I want all the good luck and, you know, blessings I can get.

00:00:24.320 --> 00:00:27.120
 That's not how you can operate a business and really deliver

00:00:27.120 --> 00:00:27.520
 ROI.

00:00:27.520 --> 00:00:31.640
 So I think that a practical view of this really needs to be,

00:00:31.640 --> 00:00:34.800
 wait a minute, not even on just a use case basis, but the way

00:00:34.800 --> 00:00:35.760
 Krista works.

00:00:35.760 --> 00:00:39.670
 Every single thing that becomes a prompt to a large language

00:00:39.670 --> 00:00:43.800
 model gets evaluated on speed, accuracy, cost and risk.

00:00:43.800 --> 00:00:47.300
 And only then we pick a model family and a model within a model

00:00:47.300 --> 00:00:48.000
 family.