it enables the "precise" flag in embree. That causes embree to use slower ray-triangle intersection algorithm, that is watertight (does not let any rays through on boundaries).
I think you mentioned you study IT, so you probably know about problems with float imprecision. The fastest ray-triangle intersection method (moller-trumbore) suffers from some catastrophic cancellations in some cases - it calculates the volume of tetrahedrons created from ray origin, intersection point, and 2 points of the triangle. This volume is zero on edges, because all 4 points lie in single plane. This zero will easily ruin your day, because you need to invert it.