In prompt: You will be penalized if…”

Hello,

On a number of occasions, I have seen people recommending to include in prompts the words “You will be penalized if…”.

I wonder if this makes any sense, considering a LLM will never be actually “penalized”.

Regards,

Martin

Hey Martin,

Great question! The reason people recommend adding words like “you will be penalized if…” is because there was a recent study done by AI researchers that showed that adding emotional language to an LLM prompt can improve responses. In fact, they found a “10.9% average improvement in terms of performance, truthfulness, and responsibility metrics.”

Here’s an article that describes it further: Emotional Language Improves AI Responses

And here’s the original paper in case you want to read it: [2307.11760] Large Language Models Understand and Can be Enhanced by Emotional Stimuli

I’d suggest trying it out for yourself and seeing if it improves your responses. And to your second question, yes it is strange that it works considering an LLM will never be penalized, but there’s a lot we’re still learning about how LLMs work!

Best,

Paul

5 Likes

Thank very much for this insightful answer.

2 Likes