Back to feeds

"prompt engineering" doesn't imply it's a solved, defined process. It leaves room to keep exploring and finding novel prompting techniques as the AI models rapidly evolve. A little bit of chaos is okay in an emerging field.

Was about to take up a course on it. Ngl this post has made me wanna think again. 😅

Prompt engineering is just a small part of the much bigger challenge of improving the general capabilities, robustness and alignment of AI systems

You make a fair point. For something to truly be "engineered" there should be standardized frameworks, extensive testing, and iteration based on results. Prompt engineering currently feels more ad hoc - people just trying things until they find decent prompts. It needs much more rigor.

I'd push back that there is actual prompt engineering research going on, even if it hasn't fully matured yet. Look at work from places like Anthropic on constitutional AI, recursive reward modeling, etc. They are actively designing and evaluating prompting techniques, even if not perfect.

I have to disagree that prompt engineering is fake. It may be overhyped or misunderstood by some, but crafting effective prompts is definitely a skill. We've seen huge performance differences based on how prompts are phrased.