Problems solved with Monte Carlo methods are often ones where you need to draw random values from oddly-shaped multidimensional distributions.
The naive method of doing this is just to cover the space evenly and weight each value by its probability. This is often computationally inefficient because you spend lots of time generating samples that don't actually contribute much to the result.
This is why we have methods like Metropolis--Hastings or Gibbs: these take an appropriately encoded description of the desired distribution and, through clever means, generate samples from it relatively cheaply, in proportion to how likely they are.
This part of the problem is likely what the GP alluded to.
The naive method of doing this is just to cover the space evenly and weight each value by its probability. This is often computationally inefficient because you spend lots of time generating samples that don't actually contribute much to the result.
This is why we have methods like Metropolis--Hastings or Gibbs: these take an appropriately encoded description of the desired distribution and, through clever means, generate samples from it relatively cheaply, in proportion to how likely they are.
This part of the problem is likely what the GP alluded to.