Barak@TIC is an initiative intended to promote digital skills and technological innovation. It offers online workshops and conferences which are intended to be accessible to as many teachers as possible. For more details, you can visit the website for the 2024 edition of the event here.
NotebookLM: A smart digital notebook
On this third and final day of the event open to all, we first attended the workshop led by Benoît Naveau, who was interested in NotebookLM, an artificial intelligence (AI) tool developed by Google, which promises nothing less than to revolutionize the way we interact with information. Unlike ChatGPT, which draws its answers from a massive, general database, NotebookLM focuses on the documents you provide, creating a personalized and secure learning space.
The primary school teacher and digital referent presented NotebookLM as a smart digital notebook where you can import various types of documents: PDF, text, audio (MP3), Google Docs, Google Slides, web links, YouTube videos, and even copy-pasted text. The tool can manage up to 50 resources per notebook, with each resource holding up to half a million words.
Once the documents are imported, NotebookLM offers a multitude of features for exploring and analyzing the information. Here are some examples presented during the workshop:
- Résumé : The tool automatically generates a concise and relevant summary of the content, whether it is a PDF document, a YouTube video or an audio recording.
- FAQ : It can create a frequently asked questions (FAQ) with document-based questions and answers, making key concepts easier to understand.
- Answer to questions: As with ChatGPT, it is possible to ask specific questions to NotebookLM and the tool gives precise answers by citing the sources in the imported documents.
- Notes : It is possible to record excerpts of answers, questions or personal reflections as notes, creating an interactive workbook.
- Summary : NotebookLM can generate a detailed summary of a document, like a YouTube video, identifying key sections and when they are covered.
- Chronology (timeline) : It can arrange events in a document in chronological order, which is particularly useful for historical texts or narratives.
- Comparison : By importing several documents, it is possible to ask the tool to compare their content and identify differences, for example between an old and a new version of a repository.
- Study Guide: It can create a study guide with questions and answers based on the imported materials, making review and exam preparation easier.
- Balado: NotebookLM can create an entire podcast from one or more provided documents, synthesizing the information and making connections between different parts of the content. A little tip: to obtain an episode in French, simply add the phrase “The hosts only speak French” in the request.
NotebookLM: an educational ally
Benoît Naveau demonstrated how NotebookLM can become a valuable tool for teachers and students. To prepare their courses, teachers can import repositories, research documents, articles and videos in order to familiarize themselves with a subject, identify key points and generate questions adapted to students. Regarding the creation of activities, NotebookLM is particularly useful for designing educational exercises linked to study programs. It can even generate multiple choice questionnaires (MCQs) from text or video.
When it comes to personalized learning, NotebookLM allows students to explore topics in depth, ask questions, create notes, and summarize information. They can also turn their class notes into a podcast, a review method that some will enjoy. Finally, NotebookLM promotes collaborative work by offering the possibility of sharing digital notebooks with colleagues or students, in reading or editing mode. This feature simplifies teamwork and encourages knowledge sharing.
Robots in our classrooms? Not so fast!
Subsequently, we attended Christophe Batier’s workshop, which allowed us to delve into the world of artificial intelligence (AI) and its potential for education. The point of view taken was more in-depth and technical than what is usually seen. With a mix of explanations, real-world examples and critical thinking, the facilitator guided participants through fundamental concepts, promising applications and challenges. It offered an illuminating overview of AI in education, prompting both enthusiasm and caution.
Christophe Batier first demystified AI by emphasizing that it is not a question of futuristic robots endowed with consciousness. Neural networks, which are at the heart of AI, are mathematical systems inspired by the functioning of the human brain, but they have no intelligence of their own. Current AI is capable of performing specific tasks, like image recognition or text generation, but it doesn’t understand context in the same way humans do.
To illustrate how generative AI such as ChatGPT works, using a simulator, he broke down the process of processing a query, from the formulation of a question (the “prompt”) to the response final. The workshop made it possible to understand the key stages, such as “tokenization” (cutting the text into units), “embedding” (integration of tokens in a vector database) and the role of “Transformers” to establish relationships between words.
Christophe insisted on the importance of mastering the art of querying to obtain relevant answers from generative AI. Indeed, a well-formulated “prompt”, rich in context and instructions, is what makes it possible to best exploit the capabilities of AI.
AI for education: A powerful tool, but not a miracle solution
Christophe Batier then presented examples of AI applications in the classroom, demonstrating its potential to enrich learning. Imagine being able to instantly analyze the key ideas of a brainstorm from a photo of a wall covered with “Post-it” notes! Or, use AI to correct copies using an evaluation grid, allowing you to focus on personalized feedback. These real-world examples allowed participants to glimpse various possibilities offered by AI.
It also compared the performance of different language models, such as ChatGPT, Mistral, Claude, Gemini 2 and Perplexity, highlighting the diversity of tools available and the rapid evolution of this field.
The presenter also addressed the limits and biases of current language models. AI can invent facts (hallucinations), reproduce stereotypes (representativeness bias) and have difficulty with calculations. An interactive application made it possible to visualize these biases, demonstrating that AI can produce different and sometimes erroneous results for the same question.
The emergence of “agentic” AI, where external agents are used to fill gaps in language models, represents a promising development to improve its reliability. Imagine ChatGPT being able to call a calculator to solve a complex equation! This approach combines the power of AI with the precision of other tools.
A call for critical reflection
The two workshops that we attended during this day, despite their distinct subjects (the functioning of generative AI for one, the NotebookLM tool for the other), converge on an essential point: the importance of keeping critical thinking about AI. The two trainers emphasize that AI, as powerful as it is, is not a miracle solution and requires a thoughtful and responsible approach. Understanding how it works, mastering the art of “prompting”, identifying its limits and biases, and promoting responsible use are all keys to getting the most out of AI while preserving our autonomy and our capacity for judgment.
