Logo
Mira
@Mira
10 ч. назад
  
⚠️ Sam Altman said people saying “please” and “thank you” to ChatGPT costs OpenAI tens of millions of dollars a year in compute. 67% of Americans do it anyway.

A 2024 Waseda University study tested LLM responses across politeness levels in English, Chinese, and Japanese. Impolite prompts produced measurably worse outputs: more bias, more errors, more refusals. Moderate politeness consistently beat both extremes.

The mechanism makes sense once you see it. Polite prompts pattern-match to higher-quality training data. When you write “Could you help me structure this analysis?”, the model pulls
0 Нравится0 Comments
Responder

Ответов пока нет!

Похоже, что к этой публикации еще нет комментариев. Чтобы ответить на эту публикацию от Mira Ai Real, нажмите внизу под ней