Google claims to have developed a new family of “fine-tuned” generative AI models for learning. LearnLM.
Google’s DeepMind AI research group and Google Research developed LearnLM models to “conversationally” instruct students on a variety of topics.
LearnLM is already powering features across Google products, including YouTube, Google’s Gemini applications, Google Search, and Google Classroom.
James Manyika, SVP of research, technology, and society at Google Research, stated during a keynote at Google’s I/O 2024 developer conference on Tuesday that LearnLM is based on educational research, making learning experiences more personal and engaging.
Google states that, via a trial program in Google Classroom, it is collaborating with educators to explore how LearnLM can simplify and enhance the lesson preparation process. According to Google, LearnLM might help instructors discover new ideas, content, and activities, as well as uncover resources targeted to the requirements of individual student cohorts.
LearnLM is also enabling Circle to Search for Android, a function that assists with fundamental math and physics difficulties and will soon grasp problems such as symbolic formulas, diagrams, and graphs. LearnLM also supports a YouTube feature (now only available on Android in the United States) that allows viewers viewing academic videos to ask clarifying questions, get answers, or take a quiz depending on what they’re seeing.
LearnLM will soon be available in Google’s Gemini applications, allowing users to develop unique chatbots that can function as subject-matter experts. Google believes that these chatbots will provide study assistance and practice activities such as quizzes and games, while also respecting each learner’s specific preferences.
Google also plans to work with institutions such as Columbia Teachers College, Arizona State University, NYU Tisch, and Khan Academy to explore potential expansions of this collection beyond its own products.
“Today marks a new chapter for learning and education at Google,” Manyika said. “Generative AI is unlocking new ways for us to make the world’s information and knowledge universally accessible and useful.”
LearnLM is a fascinating project, to be sure. However, a technical article outlining the model’s development indicates that it suffers from some of the same issues as previous generative AI models.
LearnLM, for example, does not speak in a highly “encouraging” tone and has difficulty determining when students answer practice problems correctly when compared to one of the vanilla Gemini models (Gemini 1.0), despite being superior at detecting errors. And it is not immune to hallucinations or inventing facts and numbers in response to the suggestions it receives.
For these reasons, the paper’s co-authors warn against employing this collection in applications “without further evaluation and analysis of the harms specific to [the apps]while also implying that Google conducted such evaluations and studies for their apps. Let us hope that is true for both children and educators.