ape-llms-human-level-prompt-engineers

IN premise — entries/2026/06/21/wiki-Prompt_engineering-chunk-2.md

Created 2026-06-21T09:50:10+00:00

Automatic Prompt Engineer (APE) demonstrated that LLMs can generate and select their own prompts at human-level quality, using beam search scored by output log-probabilities (ICLR 2023, arXiv:2211.01910)

Summary

A team at ICLR 2023 showed that a language model can effectively write and refine its own instructions, reaching a quality that a skilled human prompt engineer would match, without anyone manually crafting the prompts. This matters because it points toward a loop where models improve their own usability, reducing the need for humans to babysit prompt design and hinting at a path toward more self-optimizing AI workflows.

Dependents

These beliefs depend on this one:

IN prompt-optimization-is-paradoxically-automatable-yet-fragile — Prompt engineering is paradoxically both automatable (LLMs generate prompts at human-level quality) and deeply fragile (model-specific, with 40%+ accuracy shifts from minor formatting changes), implying that prompt optimization must be continuous, model-specific, and potentially self-maintaining rather than a one-time engineering effort.