What happens when you train a modern AI on etiquette books from nearly a century ago? In a fascinating short video, the channel AIbyteswithApoorv explores an unusual experiment: feeding large language models (LLMs) the refined social rules of the 1920s. The goal is to see whether AI can adopt the polished politeness of a bygone era — and whether such training reveals deeper insights into AI personality.
The experiment uses a vintage dataset of 1920s etiquette guides, challenging AI to balance courtesy with utility. Key questions arise: Can an AI be too polite to be helpful? How do historical texts shape the tone and behavior of modern AI? This quirky blend of tech and history highlights the flexibility of AI personalities and the creative ways researchers can 'engineer' them.
"We’re using historical texts to shape the future of AI interaction," the video notes.
While the experiment is lighthearted, it underscores a serious point: AI behavior is malleable, and the data we choose to train it on can produce vastly different personalities. Whether teaching AI to say 'please' and 'thank you' makes it more useful — or just more stilted — remains an open question.