Can LLMs write better code if you keep asking them to “write better code”?
In November 2023, after OpenAI added the ability for ChatGPT to generate images from DALL-E 3 within the ChatGPT web interface, there was a short-lived meme where users gave the LLM a base image and kept asking the model to “make it more X”, where X can be anything.
The trend quickly died as all of these images were very samey and uninteresting, aside from the unexplainable trend that all of the examples eventually converged into something cosmic, irrespective of the starting image and the prompt. Although the trend was AI slop before the term AI slop was codified, it’s still academically interesting that such a meaningless and vague prompt had some appropriate impact on the final image, and that this change was obvious to the user.
What would happen if we tried a similar technique with code? LLM-generated code is unlikely to be slop (although not impossible) as it follows strict rules, and unlike creative outputs such as images, code quality can be measured more objectively.
