Google has unveiled LearnLM, a new family of generative AI models specifically fine-tuned for educational purposes.
Developed through a collaboration between Google’s DeepMind and Google Research, LearnLM builds on Google’s existing Gemini models. These models are designed to tutor students across a variety of subjects in a conversational manner, enhancing their learning experience.
LearnLM has already been integrated into several Google products, including YouTube, Google’s Gemini applications, Google Search, and Google Classroom.
“Rooted in educational research, LearnLM aims to create more personalized and engaging learning experiences,” stated James Manyika, Senior Vice President of research, technology, and society at Google Research, during his keynote speech at the Google I/O 2024 developer conference.
In an effort to streamline and enhance lesson planning, Google is piloting the use of LearnLM in Google Classroom. Through this initiative, they are collaborating with educators to explore how LearnLM can assist in generating new ideas, content, and activities, as well as providing materials tailored to specific student groups.
Elsewhere, LearnLM is driving the “Circle to Search” feature on Android, which addresses basic math and physics problems and will soon tackle more complex issues involving symbolic formulas, diagrams, and graphs. Additionally, LearnLM powers a YouTube tool (currently limited to Android users in the U.S.) that allows viewers of academic videos to ask clarifying questions, receive detailed explanations, or engage in quizzes related to the content they’re watching.
In the coming months, Google’s Gemini applications will incorporate LearnLM to enable users to create custom chatbots tailored as subject-matter experts. These chatbots will offer study guidance and practice activities such as quizzes and games, while also accommodating individual learning preferences, according to Google.
Moreover, Google intends to collaborate with institutions like Columbia Teachers College, Arizona State University, NYU Tisch, and Khan Academy to explore how LearnLM can be utilized outside its own suite of products.
“Today marks a significant leap forward for education and learning at Google,” stated James Manyika. “Generative AI is creating unprecedented opportunities for us to make information and knowledge universally accessible and practical.”
While LearnLM presents exciting possibilities, a technical paper on the model’s development indicates it faces challenges common to other generative AI models. For instance, LearnLM is not particularly “encouraging” in tone and struggles to accurately recognize when students answer practice questions correctly, as compared to one of the standard Gemini models (Gemini 1.0), although it is better at detecting mistakes. Furthermore, it is prone to generating inaccuracies or “hallucinations” when provided with certain prompts.
As such, the authors of the paper advise against deploying LearnLM in applications without thorough evaluation and consideration of the potential harms specific to each application. They also noted that Google has conducted such evaluations for its own apps. For the benefit of students and educators, one would hope this is indeed accurate.
