Comment by pinkcan

models are still having trouble reasoning – current tricks still depend of feeding back an intermediate answer with a different prompt, but it is still an LLM, and most times the same LLM