您现在的位置是:Harry Potter meets AI: Magical series fuel AI experiments >>正文

Harry Potter meets AI: Magical series fuel AI experiments

上海品茶网 - 夜上海最新论坛社区 - 上海千花论坛855人已围观

简介By subscribing, you agree to our Terms of Use and Policies You may unsubscribe at any time.In Harry ...

By subscribing, you agree to our Terms of Use and Policies You may unsubscribe at any time.

In Harry Potter, JK Rowling weaved a whimsical world of portkeys, dragons, pensieves, and invisibility cloaks. Although artificial intelligence (AI) isn't some magical invention (looking at you, futuristic AGI), Harry Potter serves as a popular literary reference in AI research.

Harry Potter meets AI: Magical series fuel AI experiments

Researchers are using the Harry Potter books for cutting-edge AI experiments. The series offers a rich source of language data and complex wordplay, making it ideal for testing generative AI technology.

AI researchers count on Harry Potter

A study titled 'Who's Harry Potter? Approximate Unlearning in LLMs' explores a technique to selectively make AI models forget information. Reminds us of the spell 'Obliviate,' which removes people's memories. 

See Also Related
  • Why don't we have invisibility cloaks? 
  • This 'Magical' Harry Potter Mask Reveals Marauder's Map When You Breathe 
  • Harry Potter: Wizards Unite Installed 400k Times, Lags $1.7M Behind Pokemon Go 

Remember that dreadful scene in 'Harry Potter and the Deathly Hallows Part 1' where Hermione changes her parents' memories so they don't remember her? Devastating.

The AI model, however, addresses concerns about copyrighted content in large language models. Developed by Microsoft researchers, they demonstrated that AI models can be altered to remove knowledge of Harry Potter books without compromising their analytical abilities, reported Bloomberg.

Now that one thinks about it, there might be some similarities between AI and Harry Potter. For example, the Pensieve, which allows memories to be stored and reviewed, is similar to how data storage systems collect and preserve information for future use.

Also, isn't the polyjuice potion, which allows one to take someone else's identity for a while, just like a deepfake? Deepfake technology in AI can manipulate videos and images to create realistic simulations of people. 

But why Harry Potter?

The choice of Harry Potter is based on its universal familiarity, facilitating evaluation and confirmation of the model's 'unlearning' process by the research community.

Another study developed a language model, Silo, aimed at reducing legal risks by removing data. When using Harry Potter books to assess the impact of individual text pieces on AI performance, researchers found that eliminating the books worsened the model's accuracy measure, called perplexity. 

Recent papers on arXiv, a scientific research repository, include topics like "Machine learning for potion development at Hogwarts" and "Detecting Spells in Fantasy Literature with a Transformer Based Artificial Intelligence," said Bloomberg'sreport.

"Harry Potter is popular with younger researchers," said Leila Wehbe, a researcher at Carnegie Mellon who conducted experiments in 2014 collecting brain MRI data from people reading Harry Potter books to better understand language mechanisms.

"They would have read them as children or adolescents, thus thinking of them when choosing a written or spoken text corpus," she added.

Tags:

相关文章



友情链接