Lawyer-style prompting becomes the latest meme for taming GPT-5.6 Sol

mike64_t · x · 2026-07-28

The post jokes that if you talk to the model in lawyer-like wording—adding phrases such as “or any other such reasonable adjacent measure in addition to the ones enumerated above”—GPT-5.6 Sol stops taking the most adversarially narrow interpretation of the prompt and starts thinking about the actual task.

It is basically a prompt-engineering meme: overly formal legalese is presented as a hack to make the model behave more cooperatively, with the punchline that even a “Joe” name in the reply gets dragged into the joke.

Original post →

More from Fun

Fun channel →