Microsoft Research settled the KLS conjecture, open since 1995. This is not an applications announcement — a wall on the theory side came down.
Microsoft Research announced a proof of the KLS conjecture, unresolved for 31 years since it was posed in 1995. The question asks how small a cut can be when a high-dimensional convex body is split into two equal volumes, sitting across convex geometry and probability. It matters wherever you sample uniformly from high-dimensional space, and it has long held the place of a basic constant in theoretical computer science.
The result is neither a product nor a model; it changes how a constant used in complexity estimates is handled. Algorithms that use sampling internally — volume computation, convex optimisation, parts of probabilistic inference — get cleaner complexity bounds. That does not make existing implementations faster. It bites where you argue about how fast something could be in principle, and rewriting code is separate work. Note: the announcement does not say when this reaches applications.
Whether follow-on papers cite this to tighten upper bounds. Whether convex optimisation libraries that lean on sampling change their implementation strategy.