How Do You Separate Probabilistic AI from Deterministic Planning?

LLMs are not very good at doing math.

They are good at constructing the narrative around it.

But for the calculation, we've always wanted to ground it into the reality of the

data, and at the same time, we didn't want to give those LLM access to the entire

set of raw data in Board because it would be also cost-wise prohibitive.

They would use a million tokens, billions of tokens to do mathematical operations

on all the data that we have. So we devised an approach where the agent always goes

through those data set APIs that you can define, and this way we use the

deterministic capacity of Board to do the calculation.

The reasoning on it and the narrative around it is constructed by the LLM, and so

that's where the probabilistic AI reasoning comes into play.

The result is that you have numbers that are grounded in reality, not made up by

the LLM because they come from a mathematical engine.