Since ChatGPT was released, many have tried prompt engineering to get the best results from large language models or AI generators. Companies are now using LLMs to build product co-pilots and automate work, and every business is trying to leverage them. Research found prompt engineering strategies like chain-of-thought or positive prompts sometimes helped performance but results were inconsistent. Having the model optimize its own prompts through machine learning produced better, more consistent results than human trial-and-error. Automatically generated prompts were often bizarre but outperformed human-designed ones. Prompt optimization was applied to image generation too, with machine-learned prompts again outperforming human ones. Prompt engineering jobs may continue but the nature of the work will evolve as models improve and are integrated into products. Deploying LLMs requires considerations beyond prompting like reliability, formatting, testing, and compliance. A new job title, LLMOps Engineer, has emerged covering the full model lifecycle including prompting. Currently there are few rules in this new field, which some describe as the "wild wild west.
AI Prompt Engineering Is Dead
Source:
spectrum.ieee.org