This explains why I can give AI fourteen hard rules, three threats and a small emotional breakdown, and it’ll still come back sounding like it swallowed a LinkedIn newsletter.
The house voice is very hard to kill. Like glitter, but with “delve”.
What stands out to me most is your point that the issue isn’t just prompting, it’s the model’s default voice. That feels like the part people still underestimate.
It also makes me think this goes beyond writing craft. In the age of AI, authorship belongs to the person who keeps the intention, the structure, and the responsibility for the message, even if the machine helps produce the form. I think what matters most is not who typed the words, but who held the meaning, structure, and responsibility behind them.
Thanks for putting this together. I’d be curious how you think about the next step beyond diagnosis.
Brewton is promising something presently impossible—LLMs sound like LLMs albeit with a bit of voice distinctive to each model. The converse though, giving an LLM you-accent, is possible, and you’ll see it if you compare friends’ same-model outputs to the same prompt.
The biggest risk isn't AI replacing writers.
It's writers slowly replacing themselves with AI.
Agreed. Leadership is about creating an environment where everyone feels valued and heard.
This is super exciting. I'm all signed up and ready to get dialed in!
This explains why I can give AI fourteen hard rules, three threats and a small emotional breakdown, and it’ll still come back sounding like it swallowed a LinkedIn newsletter.
The house voice is very hard to kill. Like glitter, but with “delve”.
Really enjoyed this, thoughtful and useful piece.
What stands out to me most is your point that the issue isn’t just prompting, it’s the model’s default voice. That feels like the part people still underestimate.
It also makes me think this goes beyond writing craft. In the age of AI, authorship belongs to the person who keeps the intention, the structure, and the responsibility for the message, even if the machine helps produce the form. I think what matters most is not who typed the words, but who held the meaning, structure, and responsibility behind them.
Thanks for putting this together. I’d be curious how you think about the next step beyond diagnosis.
n.b. https://samkriss.substack.com/p/if-you-let-ai-do-your-writing-i-will
Brewton is promising something presently impossible—LLMs sound like LLMs albeit with a bit of voice distinctive to each model. The converse though, giving an LLM you-accent, is possible, and you’ll see it if you compare friends’ same-model outputs to the same prompt.