Building Duolingo Max with OpenAI's GPT-4
UX of EdTech
00:00
How to Be a Confident Learner With ABC
We're kind of a twofold approach the first one is a filter on the learners input. So we have a fine tuned GPT three model, which is basically moderating the learner so if they put something that's off topic and appropriate. We do a little bit of filtering where if their answers really far from the correct answer we don't even send it to the bot we just explained why they were wrong. And that way you can prevent them doing some like weird stuff there. Their role play which is very open ended is definitely the one we have more or have previously had more pause.
Transcript
Play full episode
