Diff of Autogollark at 596245c

@@ -24,3 +24,3 @@ Autogollark currently comprises the dataset, the search API server and the [[htt
 * Synthetic via instruct model.
-* RL (also include reasoning, of course).
+* RL (also include reasoning, of course). Probably hard though (sparse rewards).
 }

@@ -32,3 +32,3 @@ Autogollark currently comprises the dataset, the search API server and the [[htt
 }
-* MCTS over conversations with non-gollark simulacra? Should find //something// to use spare parallelism on local inference.
+* MCTS over conversations with non-gollark simulacra? Should find //something// to use spare parallelism on local inference. Best-of-n?
 * Longer context, mux several channels.