Skip to main content


Bharat Anand, Vice Provost for Advances in Learning, and colleagues at VPAL, the Future of Media Lab, and the Harvard Teaching and Learning Consortium

When it comes to the future of education, virtually no recent technology has sparked as much debate as generative AI (GenAI) and large language models (LLMs). Some have seen this technology as destructive, with school districts from Los Angeles to New York initially banning its use, and others have touted its transformative impact and possibility of changing the game for educators and students alike.

Harvard has consistently tried to embrace new technology across our classrooms, residential and virtual. This past year was no different. In fall 2023, we introduced the Harvard AI Sandbox, enabling faculty to use LLMs in their classrooms while respecting copyright and data security policies.

Several faculty colleagues experimented with these tools in their classrooms during the semester. Here, we have surfaced the learnings from interviews with 30-odd faculty across Harvard, with short 3-7 minute videos, all answering three questions:

  • What is the challenge you were trying to address?
  • How did you use generative AI tools to tackle it?
  • What did you learn?

Our small team of faculty, students, and staff extracted from these a non-comprehensive answer through debate, discussion, and our own learning. I encourage you to do the same, perhaps with your peers or teaching teams, to learn about how these might be relevant for your own courses. Here are some of our initial thoughts:

  1. GenAI tools have raised concerns about how they may compromise student assessments, promote academic dishonesty, and facilitate “lazy learning.” Our faculty colleagues who experimented with these tools were not oblivious to these concerns; indeed, many share them. At the same time, what you’ll see is how they were also looking to understand how GenAI tools can enhance the educational experience and build more vibrant classrooms.
  2. Several colleagues are leveraging LLM features that everyone should keep in mind:
    1. Beyond text: For visual aids and images, coding, analysis, games, simulations, and more.
    2. Prompt design: There’s an old saying: “garbage in, garbage out.” The output of LLMs is only as good as the input, and it’s essential to learn (and perhaps teach) how to write a prompt that works. This is highlighted through discussions on the critical role of deliberate prompt formulation, from having students iterate on their prompts through the course, to engaging students in debate on the ethics of AI use, to making advanced statistical concepts accessible to diverse learners.
    3. Interrogate hallucinations: Errors arise not just because of algorithmic or data limitations but, importantly, because LLMs are fundamentally probabilistic. Faculty have found that errors can be reduced through detailed prompt engineering and balanced AI-human partnership.
  3. Some consistent patterns and learnings emerge from how GenAI has been implemented for use by our colleagues:
    1. Going beyond the simple question-and-answer interface: Sal Khan popularized the idea of using LLMs to ask questions of a student, not just answer them. Several faculty colleagues take this further, illustrating how LLMs can be used to simulate any persona you want, and to ask anything of them. Examples include simulating experts, peers, graders, course designers, and personal chatbots.
    2. More than the “first draft”: GenAI needn’t compromise student creativity; in fact, it can augment it. Some colleagues are using it to help students refine project prototypes and polish final drafts.
    3. Work alongside what you already have: Many faculty used LLMs to improve different (and sometimes mundane) aspects of existing teaching and learning “workflows,” such as producing course materials, personalizing feedback, generating assignments, summarizing real-time student responses, and tutoring students.
    4. Identify, and overcome, hidden or invisible barriers: Students and educators sometimes confront hidden prerequisites that present barriers for teaching and learning. GenAI can assist with overcoming these skill gaps: coding for a business class, foreign languages for research, art skills for building visual aids, and even building games for class engagement.
    5. Reimagining the classroom: While we’re still in the early days of GenAI, some of these examples already start to surface more profound questions: What does a class with GenAI at its core look like? Ultimately, what is the role of a teacher?
  4. Questions around genAI’s efficacy on learning arise: can we use GenAI tools—specifically tutor bots—to improve the way students learn? One faculty member created a tutor bot that answered questions like course staff. Beyond such research on students’ interactions with genAI tools, it might be helpful to imagine how it can help you and your students now, in other ways as well: by increasing task efficiency, improving student engagement, increasing their confidence, or even improving learning outcomes.
  5. The risks of LLMs present valid concerns. While popular debates often focus on “big picture” concerns like algorithmic biases, digital divides, and fake content, some faculty explore the risks at a micro scale, within our classrooms, such as hallucinations, failed reasoning, or superficial thinking, pushing students to understand these issues more deeply.

We hope this starts to give you a taste of the richness embedded in these videos. To start experimenting, request access to the Harvard AI Sandbox. As you think about what’s relevant for your course, support exists across Harvard to help you and your team experiment as well.

Looking ahead, it is clear we will continue to learn. We will experiment. We will refine. And we will advance our understanding. In the next months, we’ll continue to expand our catalog to include more innovations from Harvard faculty—please reach out to the Office of the Vice Provost for Advances in Learning (VPAL) via email if you would like to share your own experiences with colleagues across Harvard. We’re deeply grateful to our colleagues for sharing their experiments and experiences, featured here in the first Harvard GenAI Library for Teaching and Learning.

Frequently asked questions

The following information offers advice for educators interested in using generative AI tools in their teaching and course preparation. As this technology is constantly evolving, this page will be updated frequently with new resources and advice.