Get the latest tech news
Can LLMs write better code if you keep asking them to “write better code”?
Most coders want AI to write code faster: I want AI to write FASTER CODE.
For this experiment, we will give Claude 3.5 Sonnet an interview-style coding prompt using Python: one that is both simple and could be implemented by a novice software engineer, but one that can be heavily optimized. “Planning” is a long-used trick to help align LLM output for a first pass — the modern implementation of “ let’s think step by step.” Some LLMs ignore it, but Claude 3.5 Sonnet always obeys. Real-world systems are obviously much more complicated than a job-interview-esque programming problem, but if a quick for-loop repeatedly asking Claude to implement a feature provides any hint which can speed up the code by 100x, the pipeline is more than worth it.
Or read this on Hacker News
