Simon Willison has compiled a list explaining how large language models (LLMs) can be used to help with coding, illustrating it with examples, strategies and tips for using AI assistants such as ChatGPT and Claude effectively.
While LLMs can assist with programming work, human oversight is crucial and testing is essential.
LLMs can also help with software development more generally, for example by offering interactive exploration of unfamiliar codebases, aiding research or speeding up the prototyping process.
LLMs can degrade in performance when trained on AI output, so it is important to use reliable human sources of information.
Currently, LLMs work best with JavaScript and Python code rather than with less common languages or less mainstream coding languages.
Even when using forums as a data source, the information is still useful provided it is framed correctly; knowing something doesn’t work is as useful a data point as knowing something that does.
