Training Track - ChatGPT & Next Generation Assistants

The coursewill aim at explaining the nuts and bolts of the new generation of chatbots such asChatGPT, what they can be used for, and what are their limits. We will start with an overviewof reference chatbots (both closed and open source, ChatGPT and Mistral for example),common use cases (writing, summarizing, brainstorming, ...), and some of the main limits ofthese tools (bias, hallucinations, ...). The course will then dive gradually into more advanceduse cases, addressing along the way more subtle issues, bothethical and technical. Examplesof ethical issues that we will address cover data privacy, intellectual property, and the digitaldivide. On the technical side, we will show how chatbots can be programmed and trained, usingeither APIs for closed-source solution, or libraries such as HuggingFace for open-sourcemodels. By the end of the course, the goal is that participants can make their informed decisionson the capabilities of this new type of software, and when and for what they may appropriatelyuse them.

• Introduction to large language models (LLMs) such as ChatGPT
• Experiment with different use cases
• Discuss about limitations, both technical and ethical
• Program LLMs, both closed and open-source models (workshop day-technical profile)
• Design a Chatbot Web application (workshop day-technical profile)

• Demonstrate an understanding of large language models' technology within the field of machine learning
• Identify the primary use cases for LLMs and formulate effective prompts
• Recognize and explain the technical limitations and ethical issues associated with LLMs
• Use and program with LLMs in Python
• Apply methods for data augmentation and fine-tuning