RLHBARF - Reinforcement Learning from Human But Actually Robot Feedback
There is an actual interesting question here if BARF loops can create new synthetic knowledge, or if humans need to be the 'spark of novelty' in that loop to drive it to new knowledge.
There's another thread running at the moment about BARF degrading the machine's performance. I don't have enough information to be confident, but it sounded plausible and likely.