Shreya Shankar: evaluation

Transcript context

…Before we get into that actually, just to make sure people know exactly what you're describing there, these two types of evals. One is you said it's code-based and one is LLM as judge. Maybe Shreya, just help us understand what code-based eval even is? It's essentially a unit test? Is that a simple way to think about it? Yeah. Maybe eval is not the right term here, but think automated evaluator. So when we find these failure modes, one of the things we want is, "Okay. Can we now go check the prevalence of that failure mode in an automated way without me manually labeling and doing all the coding and the grouping, and I want to run it on thousands and thousands of traces, I want to run it every week." That is, okay. You should probably build an automated evaluator to check for that failure mode. Now, when we're saying code-based versus LLM-based, we're saying, "Okay. So maybe I could write a Python function or a piece of code to check whether that failure mode is present in a trace or not." And that's possible to do for certain things like checking the output is JSON, or checking that it's markdown, or checking that it's short. These are all things you can capture in code or you could approximately capture in code. When we're talking about LLM judge here, we're saying that this is a complex failure mode and we don't know how to evaluate in an automated way. So maybe we will try to use an LLM to evaluate this very, very narrow, specific failure mode of handoffs. So just to try to mirror back what you're describing, you want to test what your, say, agent or AI product is doing. You ask it a question, it gets back with something. One way to test if it's giving you the right answer is if it's consistently doing the same thing, that you could write a code to tell you this is true or false. For example, will it ever say there's a virtual tour? So you could ask it.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.