- We begin with the pedagogical observation that learning can be more effective when the experience is adapted to the characteristics and needs of the lea - We begin with the pedagogical observation that learning can be more effective when the experience is adapted to the characteristics and needs of the learner - The value of multiple representations of content has been studied in learning science. For example, dual coding theory suggests that multiple representations have the advantage of forging links between different encodings of the same concepts, thus reinforcing the corresponding mental conceptual structures. Learn Your Way is therefore augmented with multiple views of the material (audio lessons, narrated slides, and mind maps), which learners can interact with and choose from - It is also motivated by theories of self-regulated learning (SRL) and visible learning that acknowledge the importance of supporting learners with cognate control of their learning process - Gen-AI offers four key opportunities in this context. First, it can generate such content for any material the learner is interested in. Second, it can do so while adapting to the specific attributes and needs of the learner. This is in contrast to the generation of personalized learning material by human educators, which is a much longer process and is impractical to do at scale. Third, AI can be used to generate different representations of the material, including visualizations and audio-based formats, which are known to further enhance the efficacy of learning Finally, AI can generate formative assessment elements tailored to the learner, allowing them to monitor and regulate progress. As notes, formative assessment is a critical driver of learning, and in particular self-regulated learning - The text is generatively adapted, with the goal of matching the Flesch-Kincaid Grade (FKG) for that level, while maintaining factuality and coverage of the material. This is known as re-leveling and is part of Gemini 2.5 Pro core education capabilities. - Our Gen-AI rewriting is done in a focused manner, by first selecting parts of the text that are particularly amenable to personalization, and then replacing only these parts with an AI-rewritten personalized version. This has the added advantage of highlighting the personalized text, thus informing the learner that it has been specialized to their interests - Visual learning is broadly recognized as a powerful medium, and many textbooks include explanatory diagrams and drawings. It is natural to use AI image generation tools to produce such visuals. However, our initial exploration found that even the most advanced AI image generation models struggle to produce these types of images. This can be explained by the fact that such models are trained to produce realistic images that are high on detail, and sometimes return inaccurate images when asked for ‘simple’ or ‘educational’ ones.1 To overcome this, we fine-tuned a model specifically for this task. This model is applied to parts of the material that Gemini identifies as worthy of illustration. ---- (Team et al., n.d.)