>>16754894
Kind of makes me think that the "ultimate limit" of LLMs trained on feedback is just picrel.