Google has announced the development of a new family of generative AI models called LearnLM, which have been fine-tuned for learning. This collaboration between Google’s DeepMind AI research division and Google Research aims to provide conversational tutoring to students across various subjects. These models, built on top of Google’s Gemini models, are already being used in several Google products, including YouTube, Google Search, and Google Classroom.
According to James Manyika, SVP of research, technology, and society at Google Research, LearnLM is grounded in educational research, making learning experiences more personal and engaging. During a keynote at Google’s I/O developer conference, Manyika highlighted the potential of LearnLM in simplifying and improving the process of lesson planning in Google Classroom. By working with educators, LearnLM could help teachers discover new ideas, content, and activities tailored to the needs of specific student cohorts.
LearnLM is already powering Circle to Search on Android, a feature that helps solve basic math and physics problems. In the future, it will be able to understand more complex problems involving symbolic formulas, diagrams, and graphs. Additionally, LearnLM is being used on YouTube (only on Android in the U.S. for now) to enable users watching academic videos to ask clarifying questions, get explanations, or take quizzes based on the content they are viewing.
In the coming months, LearnLM will be integrated into Google’s Gemini apps, allowing users to create custom chatbots that act as subject-matter experts. These chatbots will provide study guidance, practice activities like quizzes and games, and will respect each learner’s individual preferences.
Google plans to extend LearnLM beyond its own products through partnerships with organizations such as Columbia Teachers College, Arizona State University, NYU Tisch, and Khan Academy. This collaboration aims to explore how LearnLM can be utilized in different educational contexts.
While LearnLM shows promise in revolutionizing learning and education, a technical paper detailing its development highlights a few limitations. LearnLM struggles with maintaining an encouraging tone and has difficulty identifying correct answers compared to the vanilla Gemini models. However, it excels at identifying mistakes. There is also a risk of hallucinations, where LearnLM may generate false information in response to the prompts it receives.
The co-authors of the paper caution against using LearnLM in apps without further evaluation and analysis of potential harms specific to those apps. They suggest that Google has performed such evaluations and analyses for its own apps. It is crucial for the sake of students and educators that Google ensures the reliability and accuracy of LearnLM’s responses.
Overall, LearnLM represents an exciting development in the field of generative AI models for education. With the potential to personalize and enhance learning experiences, it has the power to make knowledge more accessible and useful. However, it is essential to address the limitations and risks associated with AI models to ensure their effectiveness and ethical use in educational settings.
