Is It Possible to Write the Ultimate RL Algorithm on a Whiteboard?

The amount of RL algorithms is exploding exponentially and every there's variants of sub variants. People think they can actually write the algorithm themselves, that's not possible. If you want to build an AGI, we need the thing that uses all possible strings in one model. That's likely to be more of a solution to the RL problem than beating the score on Mujoko with like CQL plus plus.