Supporting education from classroom to career

To help ensure our AI benefits both learners and educators, we develop tools for learning that are pedagogically sound and demonstrably effective. We work with teachers, researchers, and students to validate our educational tools, ranging from technical benchmarking against learning science principles to randomized control trials and real-world pilot programs.

Towards Responsible Development of Generative AI for Education: An Evaluation-Driven Approach
Our technical whitepaper on LearnLM, a family of models specifically fine-tuned for pedagogical best practices and student safety.

Evaluating Gemini in an Arena for Learning
A research summary of how large language models (LLMs) perform against educational benchmarks and tutoring-specific criteria.

Towards an AI-Augmented Textbook
Our technical report detailing how AI-augmented textbooks can transform static curricula into dynamic, personalized learning experiences for students.

In an effort to reduce hallucination, we continue to train models to use trusted sources and verify outputs. Yet challenges still remain in determining which sources are trustworthy and how to handle subjective questions. To be an effective educational tool, AI must both be able to do both, moving beyond simple reinforcement so it can challenge a student’s misconceptions and correct inaccurate statements, not simply act as an uncritical mirror that enforces them. This is an area of deep importance to Google. We have made meaningful progress with each new model release, and continue driving the industry forward with new, comprehensive benchmarks that evaluate the ability of LLMs to generate factually accurate and sufficiently detailed answers to user queries.

DataGemma: Using real-world data to address AI hallucinations
An overview of new grounding techniques that connect LLMs to extensive statistical databases to ensure factual accuracy.

FACTS Grounding: A new benchmark for evaluating the factuality of large language models
Our technical deep-dive into new benchmarks designed to evaluate and improve the factuality of model responses.

Re-evaluating Factual Consistency Evaluation
A critical look at the methodology used to measure how reliably AI summarizes information without introducing errors.

We’re committed to ensuring that AI is used as a supervised platform for safe learning for students. AI education tools like Gemini for Education are designed to support established pedagogical principles and give teachers and administrators direct control over how, when, and where AI is used, allowing them to turn these systems on and off within classrooms. As educational institutions grapple with this topic, Google is exploring tools that can support schools as they adjust their curricula and assessments. We hope our experimentation in these areas, such as developing AI tools that can help scale oral assessments or enable students to show their work, can be beneficial.

What do AI chatbots really mean for students and cheating?
A Stanford-led analysis of how AI is reshaping academic honesty and the evolving nature of student assessment.

Teaching Responsible use of AI
Our guide for educators to help integrate AI literacy into the classroom in a way that emphasizes ethical use and critical thinking.