Study Finds Politeness with AI Models is Futile

A new study by researchers from George Washington University reveals that politeness towards AI models is a waste of computational resources.

Adding words like “please” and “thank you” to prompts has a negligible impact on the subsequent quality of chatbot responses.

Experts found that polite language is generally “orthogonal to the substantive good and bad output tokens” and has “minimal impact on the dot product”—meaning such words occupy separate areas in the model’s internal space and hardly affect the outcome.

The article contradicts a Japanese study from 2024, which claimed that politeness enhances artificial intelligence performance. That research tested GPT-3.5, GPT-4, PaLM-2, and Claude-2.

David Acosta, Director of AI at Arbo AI, noted that discrepancies in results are due to the overly simplified model from George Washington University.

“Contradictory results on politeness and AI performance are generally due to cultural differences in training data, nuances in task-specific prompt design, and contextual interpretations of politeness, necessitating cross-cultural experiments and task-adapted evaluation systems for clarification,” he commented.

The team behind the new work acknowledged that their model is “intentionally simplified” compared to commercial systems like ChatGPT. However, they believe that applying the approach to more complex neural networks will yield the same result.

“Whether an AI response is inadequate depends on the training of the LLM that forms the embeddings of tokens, and the content of the tokens in the query—not on whether we were polite to it or not,” the study states.

In April, OpenAI CEO Sam Altman stated that the company spent tens of millions of dollars on responses to users who wrote “please” and “thank you.”