A version of OpenAI’s GPT-4 will be ‘teaching’ thousands of kids this fall
Thousands of students heading into the new school year will arrive in classrooms from kindergarten to highschool alongside a new tutoring assistant: a large language model.
As CNN noted today, the education nonprofit service Khan Academy, is expanding its Khanmigo AI access to over 8,000 educators and K-12 students as part of its ongoing pilot program for the new technology. According to Khan Academy’s project description, Khanmigo is underpinned by a version of OpenAI’s GPT-4 large language model (LLM) trained on Khan Academy’s own educational content. Additional parameters are encoded into the product to tailor Khanmigo’s encouraging response tone, while also preventing it from too easily divulging answers for students.
But despite past controversies regarding the use of AI chatbots as stand-ins for various historical figures, Khanmigo reportedly embraces the concept. In its current iteration, users can interact with chatbots inspired by real people like Albert Einstein, Martin Luther King, Jr., Cleopatra, and George Washington, alongside fictional characters such as Hamlet, Winnie the Pooh, and Dorothy from The Wizard of Oz. And instead of glossing over difficult topics, AI invoking complex figures purportedly do not shy away from their onerous pasts.
“As Thomas Jefferson, my views on slavery were fraught with contradiction,” Khanmigo reportedly told a user. “On one hand, I publicly expressed my belief that slavery was morally wrong and a threat to the survival of the new American nation… Yet I was a lifelong slaveholder, owning over 600 enslaved people throughout my lifetime.”
But despite these creative features, Khanmigo is still very much a work in progress—even when it comes to straightforward math. Simple concepts such as multiplication and division of integers and decimals repeatedly offer incorrect answers, and will even sometimes treat students’ wrong inputs as the correct solutions. That said, users can flag Khanmigo’s wrong or problematic responses. Khan Academy representatives still refer to the software as a “beta product,” and reports continue to describe the pilot period as a “test.” Another 10,000 outside users in the US agreed to participate as subjects while paying a donation to Khan Academy for the service, CNN adds.
As access to generative AI like Khanmigo and ChatGPT continue to expand, very little legislation currently exists to regulate or oversee such advancements. Instead, the AI tools are already being used for extremely controversial ends, such as school districts employing ChatGPT to assist in screening library books to ban.
Although they believe AI could become a “pretty powerful learning tool,” Kristen DiCerbo, Khan Academy’s Chief Learning Officer conceded to CNN on Monday that, “The internet can be a pretty scary place, and it can be a pretty good place. I think that AI is the same.”