1
Why Generative AI Today?
- A timeline outlining the evolution of various AI paradigms, highlighting major inventions and innovations.
- A refresher on key concepts, including deep learning and neural networks (with emphasis on the number of parameters in a model).
- How ChatGPT was built: the fundamental differences between the Large Language Model (LLM) component and the conversational component.
- Potential limitations and biases arising from training data.
- Limitations related to content moderation and OpenAI’s policies.
- Pricing models based on token usage, along with their advantages and drawbacks.
2
Basic Techniques in Prompt Engineering
- What is prompt engineering? Why is it essential for optimizing the use of language models?
- Structures of prompts: open-ended, closed, pre-filled, multiple-choice, confirmation, and error prompts.
- Creating effective prompts based on examples of tasks to be accomplished.
- The roles of context and task specificity, including considerations for prompt length.
- The Randomized Controlled Trial (RCT) method.
- Analyzing the pros and cons of AI-assisted decision-making.
- Identifying key biases (such as gender bias) and issues related to hallucinations.
- Recognizing topics that ChatGPT restricts and analyzing its sensitivity to context.
Demonstration
Creating and testing prompts for various tasks, with and without applying steerability; comparing responses; analyzing different AI outputs to identify inconsistencies or errors.
3
Enhancing the Quality of Interactions with an AI
- Generating prompts using ChatGPT.
- Exploring advanced use cases, including an example using a “thought tree.”
- Detailing techniques for crafting effective prompts (precision, context, follow-up queries).
- The Princeton React method.
- How to use AI for writing assistance and learning.
- Automating data collection with ChatGPT.
Demonstration
Experimenting with complex prompt techniques in concrete use cases, comparing the performance of various prompts on given tasks, and raising awareness about chatbot vulnerabilities (e.g., Gandalf.AI).
4
Recognizing and Evaluating the Limitations of an AI
- Learning to identify the markers of AI-generated text (style, coherence).
- Proposing constructive improvements rather than merely dismissing AI outputs.
- Developing a multidimensional evaluation framework.
- Generating website code from a sketch.
- Understanding the concept of narrow AI versus strong AI and the future of Artificial General Intelligence (AGI).
- Comparing responses between ChatGPT and other tools with fewer parameters.
Demonstration
A comparative analysis where participants identify whether texts were written by humans or an AI, explain their reasoning, refine an AI-generated text using examples, and evaluate responses based on established criteria.
1d
- 7h00 
Download in PDF format
Share this course by email