This is a surprising take to me! Would love to learn more about what you mean. I feel like the problems they solve seem so direct to me. For example: RLMs are an approach to long context problems. Not every problem is a good fit for RLMs for sure, but I can see some problems where I imagine it would work well!