When AI Gets It Wrong: Learning Generative AI Through Its Mistakes

Generative AI tools often make mistakes, especially with historical East Asian sources. This hands-on workshop. led by Kwok-leong Tang from Harvard University, uses those mistakes to explain how the technology works, so that participants can decide when to trust it and when to use it.

No technical background is needed, and you can attend in person or register to join via Zoom. A laptop is required for in-person participants — the Hatcher Gallery Lab is equipped with laptops, but please feel free to bring your own.

When an AI tool gives a wrong answer, there are two possible reasons: the model itself cannot do the task, or we have not given it what it needs. The workshop teaches how to tell these apart. First, we will run a small AI model on our own laptops and look inside it: how it reads text, how it chooses each word, and why it can sound certain and still be wrong. Next, we will see how clearer instructions, more context, or a connected tool can fix some mistakes but not others. Finally, we will practice a simple way to test an AI answer: ask the same question several times and check the results against a reliable source.