1,370 views
In this video, the third episode of an 8-part course for students of the various masters programs at HES-SO (Switzerland), Richard-Emmanuel Eastes dissects how chatbots work and presents their principles as well as best practices for use through various examples of tools and uses. He begins by explaining how chatbots, such as ChatGPT, generate responses. Based on advanced language models and neural networks with trillions of parameters, these tools analyze the prompts provided, break the text down into tokens, and predict each subsequent word to construct coherent responses. He emphasizes that while these responses are plausible, they are not necessarily true, which poses a verification challenge for the user. The course also addresses potential biases in chatbot responses. Richard-Emmanuel highlights how prompts influence responses, sometimes leading to unintentional biases or incomplete responses, especially on sensitive topics. He also explores the political coloring of language models, presenting studies that reveal political tendencies linked to training data and designers' preferences. To promote safe and informed use of chatbots, Richard-Emmanuel proposes several methods. First, he advises limiting their use to areas where the user has sufficient expertise, in order to be able to validate the answers obtained. By introducing prompt engineering, he then shows how to structure queries to improve the accuracy of answers by providing clear contexts or by breaking down tasks into simpler steps. Finally, he recommends feeding chatbots with specific knowledge bases, such as documents or articles, in order to align the answers generated with reliable sources. He then presents advanced techniques, such as thought chains, which consist of guiding the chatbot step by step in solving a problem, as well as the use of Notebook LM, a Google tool that responds only on the basis of documents provided, thus guaranteeing precise and referenced information. In conclusion, Richard-Emmanuel highlights the relevant roles of chatbots as assistants and thought partners. He closes the course by announcing that the next episode will focus on the ethical and societal issues of artificial intelligence, inviting students to reflect on the implications of these tools in our society. Test your knowledge on the video with Wooflash: https://app.wooflash.com/join/HNAP6DA...