I see. Agree with the point about marginal improvements at a hefty increase in computational cost (I touch on this a bit at the end of the blog post where I mention that better performance requires better tooling/base models). Though I would still consider sampling and voting a “multi-agent” framework as it’s still performing an aggregation over multiple results.