Everyone is stacking models to hedge bets, but it is often more pain than gain....

https://www.list-bookmarks.win/mixing-llms-like-lego-bricks-seems-smart-because-models-have-different

Everyone is stacking models to hedge bets, but it is often more pain than gain. Mixing models leverages specific strengths and catches failure modes, yet synthesis layers can hide vital dissent or leak private data across vendors