Google says it has developed a brand new family of generative AI models which are “fine-tuned” to learning: LearnLM.
The LearnLM models are a collaboration between Google's DeepMind AI research division and Google Research and are based on Google's Gemini models. According to Google, they’re designed to show students “conversational” on a spread of topics.
LearnLM already supports features across all Google products, including YouTube, Google's Gemini apps, Google Search, and Google Classroom.
“LearnLM relies on educational research and makes learning experiences more personal and fascinating,” says JAmes Manyika, SVP of research, technology and society at Google Research, said onstage during a keynote at Google's I/O 2024 developer conference on Tuesday.
Google says it’s working with educators through a pilot program in Google Classroom to see how LearnLM could simplify and improve the lesson planning process. According to Google, LearnLM could help teachers discover latest ideas, content and activities, or find materials tailored to the needs of specific student groups.
Elsewhere, LearnLM supports Circle to Search on Android, a feature that helps solve basic math and physics problems – and can soon understand problems like symbolic formulas, diagrams and graphs. And LearnLM supports a tool on YouTube (currently only on Android within the US) that enables users watching academic videos to ask clarifying questions, get explanations, or take a quiz based on what they're watching.
In Google's Gemini apps, LearnLM will give users the power to create custom chatbots that may act as material experts in the approaching months. These chatbots provide learning aids and practice activities similar to quizzes and games, and respect each learner's individual preferences, says Google.
Google also plans to work with organizations similar to Columbia Teachers College, Arizona State University, NYU Tisch and Khan Academy to explore how LearnLM can expand beyond its own products.
“Today begins a brand new chapter for learning and education at Google,” said Manyika. “Generative AI opens up latest opportunities for us to make the world’s information and knowledge widely accessible and usable.”
LearnLM is actually fascinating work. But a technical paper An in depth description of the model's development shows that it suffers from among the same problems as other generative AI models.
For example, LearnLM doesn’t speak in a really “encouraging” tone and, in comparison with considered one of the usual Gemini models (Gemini 1.0), has trouble recognizing when students are answering practice questions accurately, although it is best at . And it will not be proof against hallucinations or making up facts and figures in response to the prompts presented to it.
For these reasons, the paper's co-authors warn against using LearnLM in apps “without further assessment and evaluation of the harms specific to (the apps)” – and in the identical breath indicate that Google requires such assessments and analyzes for its apps have carried out. For the sake of scholars and teachers, let’s hope that’s true.
