Nowadays most of the spotlight seems centered on LLMs, CNNs, and other methods that are either human-labeled or at least reliant on human-created data, and have a static separation between "learning" and "inference".
I know that there are still non-LLM, non-CNN, non-anthropocentric topics of AI development currently, in RL and in other areas. Which would you say are the most prominent or promising today, or likeliest to come to fruition?