Noah Labhart

Why you think they're only as good as some of those nines?

Tell me a little bit more about that.

Certainly.

Now that makes total sense.

And that feels like kind of the root, call it growth area, maybe call it issue or limitation with AI in general, the probabilistic nature of

Over deterministic that we need for regulators, for compliance, for validation, for all those sorts of things.

So then what practical steps can organizations take to secure inference today?

We see the issues with the guardrails.

I think you outlined that perfectly.

And when you talk through compliance and why these things are critical to secure, how do they get started?

What's step one, two, three, as many as you can list?

So that makes sense.

A good starting point.

And I appreciate you mentioned doing the guardrails anyway, because that's what they are.

They're just guardrails.

They're not 27 nines of accuracy, six sigma level safety there.

And I think my last question is probably one that everyone is interested in, right?

Because

Security can come at a cost of implementation, at a cost of perhaps performance.

How can businesses balance performance with security when adding these guardrails?