Every regular expression I’ve seen produced by a large language model has either being broken, or done an unacceptably bad job due to being the wrong tool for the job (e.g. validating an email address with a fairly short regular expression, which I’ve seen people share the answers of at least three times).
