There has been a lot of discussion recently in certain AI research communities recently about whether or not the entire idea of reinforcement learning is even n

There has been a lot of discussion recently in certain AI research communities recently about whether or not the entire idea of reinforcement learning is even necessary. Self supervised sequence models seem to represent the best path forward for general purpose problem solving agents because we can essentially just keep improving them by increasing parameter counts. There was a previous HN discussion about the idea here https://news.ycombinator.com/item?id=27659526 https://news.ycombinator.com/item?id=27659526