Nondeterministic algorithm

I'd be like, what percentage of what it's doing is deterministic and non-deterministic? Deterministic being like A always equals A.

Six times two always equals 12.

Non-deterministic is what ChatGBT or LLM outputs are.

When I say, where is the cat? Depending on the context, it might say it's in the hat or in the tree or somewhere else.

It will give you that answer depending on what context you feed it.

It's the same every time, like two plus two equals four.

Non-deterministic would be, it's a probabilistic process.

And this is the way that AI works.

It will take an input and it will say, okay, based on all of these various probabilities, what is the next letter that I should write? And because there's lots of confounding factors all the time, you can't always predict exactly what's going to come out.

It's really a, it's called a prediction engine.

And whether we get into the deterministic and non-deterministic outputs, when they start talking technical language, deterministic just means there's an algorithm used and it's going to give you the same answer every time.

Non-deterministic is going to mean that every time you ask a question, even if it's the same question in a chat or a prompt, it's going to give you a slightly different answer each time.

So that's a kind of split between different versions

of the same model.

But the way you get to that is very different from sort of past uses, and I think you talked about the sort of finite technology, like set it and forget it.

AI, I think the nondeterministic nature of AI in particular, and for the audience, right, deterministic is like you ask the same question, you're always gonna get the same answer, and that's your deterministic.

Nondeterministic is you ask a question, and much like an employee, you ask the same question to five people, they're gonna give you five slightly different answers.

Now, the nature of that technology therefore means that we have to treat the technology and engage with it in a different way.

And one of the things that we've learned over time, and we're seeing this also with our grantees and the feedback, one of these workshops around adoption was actually you need to manage your agent much like you manage an employee, which is a really, I think, novel idea, which, uh, I don't know that we've ever thought of technology the same way.