AI Tutor for Students and Teachers

So there are headlines in education for the last few months: Students intend to use ChatGPT and other forms of AI to cheat on assignments - they learn nothing! It will dramatically change education as we know it. But there are ways to avoid this crisis if we implement the right safeguards and reforms now. We're on the brink of a new era in education, where an artificially intelligent teacher can help students everywhere reach their full potential. And we're creating a superhuman assistant for every teacher on the planet. And to give you an idea of ​​how big that undertaking would be, I'd like to share this video by Benjamin Bloom, a psychologist who worked in education and made a nice detour trying to understand what makes people good. at learning, but also studied mastery rather than intelligence. Bloom's theory, 2 sigma, is based on the concept of standard deviation, a statistical measure where most values ​​are close to the mean with fewer values, and he had good data to show that the world is actually divided into groups - as usual. . distribution , which can be seen to some extent in the bell curves so if you could give individualized instruction to students (and with the right distribution), you could achieve a two standard deviation improvement. In plain language, that can take your average student and turn them into an extraordinary student. Your smaller than average student may need it and make them more than an average student. The reason he described this as a problem was now, he said, well that's all good But how do you scale group learning in this way? How do you really give it to everyone financially? I'll show you what I think are the first steps to doing this. Obviously, we tried to approximate it somehow at Khan Academy for over ten years, but I think we are speeding it up dramatically. I will show you the early stages of our AI, which we call Khanmigo, what it can do now and maybe a bit about where it's actually going. So here this is a traditional exercise which you or many of your children have seen at Khan Academy. What's new, though, is the little robot on the right. Let's start by looking at one very important defense, in this case the conversation will be recorded and your teacher can review it. It is actually controlled by another AI. And it doesn't say the answer either. This is not a cheat tool. When the student says, "Tell me the answer," it says, "I'm your teacher. What do you think is the next step to solve the problem? " Now when a student makes a mistake, it surprises people who think that large language models are not good for mathematics, notices, just doesn't notice the mistake, it asks the student to explain their reasoning, but it actually does what I would say not just an average teacher, but a great teacher would do. It can predict what that student's misconception is likely to be, that they are unlikely to use distribution assets. Remember that we have to divide negative two both nine and 2 meters in brackets. This is a very, very, very big deal to me. And it's not just in math. This is a computer programming exercise at Khan Academy, where the student has to separate the clouds. And so we see how the student begins to define a variable, left X minus minus. It only did the left cloud part. But then they can ask Khanmigo what is going on? Why is only the left cloud moving? And it understands the code. He knows the context of everything the student does, and he realizes that these ellipses are for drawing clouds, which I think is kind of shocking. And it says: "That the true cloud may also be moved, try adding a line of code to the draw function which increments the right X variable by one pixel every frame. " Now it is perhaps even more wonderful because we have many math teachers. We've all tried to teach the world to code, but there are not many IT teachers. And what you just saw, even though I lead my children, when they learn to code, I can't help them so well, so fast. It will be a really great tutor. And it's not just about the exercises. It understands what you're looking at. It understands the context of your video. It may answer the age-old question, "Why do I need to learn this? " And it socratically asks: "Well, what are you interested in? " Let's say a student says, "I want to be a professional athlete. " And it says, "Well, if you know the size of the cell, what is this video which can be really useful for understanding nutrition and how your body works, etc. " It can answer questions, it can ask you, it can connect it with other ideas, you can now ask as many questions from the video how can you dream? There is another major disadvantage, I remember the high school I went to The student to teacher ratio was about 200 or 300:1. Much of the country is worse. We can use Khanmigo to provide each student with a tutor, academic coach, career coach, life coach, which is exactly what you see here. And we launched it with the GPT- release. We have a few thousand people here. This is not a fake demo, it really is in action. And then you know things I thought would be even harder it would be a bit science fiction even with a traditional teacher. We have an online high school with Arizona State University called Khan World School, and we have a student studying in this online school in India. Her name is Saanvi. And he did a report on The Great Gatsby. And when he reads The Great Gatsby, Jay Gatsby looks at a green light in the distance. And he says, "Why is he doing this? " He did some web searches and apparently people were looking into it and commenting on its symbolism, but none of them resonated with her. And then she had realized that she had Khanmigo and that she could talk to Jay Gatsby herself. And so: "Ah, great choice, old sport. Now I'm Jay Gatsby, the enigmatic millionaire. " From F. Scott Fitzgerald's classic. " And so: "Why are you looking at the green light? " “Ah, the green light, the old sport. It's a symbol of my dreams and hopes, you see. It is located at the end of Daisy Buchanan Pier across the bay from my mansion. I look at it longingly because it represents my longing for the past and I long to see Daisy again, the love of my life. " And what was cool, Saanvi then said, "I had this long conversation. "She called him "Mr. Gatsby" and finally apologized for his time, which I thought was very polite of him. But you can imagine that it opens up the study of literature, learning. . . You could talk to historical figures. We even plan to add an activity where we can talk about the Mississippi river, for example. It brings things to life in a way that is actually science fiction even six months or a year ago. Students can chat with artificial intelligence. And there we have it, a student discussion about whether we should cancel student debt. The student objects to the cancellation of the student loan, and we received very clear feedback. We started using it at Khan World School in our lab school, Khan Laboratory School. Students, especially high school students, say: “It's amazing that I can refine my arguments without fear of judgment. It makes me a lot more confident to go into the classroom and really participate. " And we all know that Socratic dialogue is a great way to learn, but honestly it is not available for most students. But now it is hopefully available to everyone. Much of the story we saw was in the headlines "This is writing for children. Children don't learn to write. " But we will show that there are ways that AI will not write for you, it writes with you. So it's a little thing, and my eight year old is addicted to it and he is not a child who ever liked to write, but you can say "I want to write a horror story" and it says, "Oh, a horror story, how spine-tingling and exciting. We plunge into a world of ghostly shadows and chilling mysteries. " And this is an activity where the student writes two sentences, and then the AI ​​writes two sentences. And so they make a story together. One student writes: “Beatrice was a misunderstood ghost. He wanted to make friends but accidentally scared them.

And the AI ​​says: "Poor Beatrice, a lonely spirit in need of companionship. One day she came across an old abandoned mansion," etc. I encourage you all to try it one day. It's surprisingly fun. Now touch on this use even more directly. And what I will show you is everything I've shown you so far is actually already part of Khanmigo and what I will show you, We haven't shown anyone yet, this is a prototype. We hope to release it in the coming months, but it's direct AI, generative AI, so as not to harm English or language skills but really makes it better somehow which we could not have imagined even a year ago. This is reading comprehension. Students read Steve Jobs' famous speech at Stanford. And when they reach certain points, they can click on this little question. And AI then Socrates, almost like an oral exam, asks the student questions. And AI can highlight parts of a song. Why did the author use this word? What was their purpose? Does it support their claims? They can start doing that again. We never had the opportunity to give everyone a teacher, everyone is a writing coach who can really handle reading at this level. And you can go to the other side of it. And we have entire workflows to help them write, help them be a writing coach, draw an outline. But when the student actually creates the sketch, and here they build nature, they can ask for feedback again, as you would expect from a good writing coach. In this case, the student says, it was said "Does my evidence support my claim? " And then the AI ​​just can't give feedback, but it manages to emphasize certain points and says: "It does not fully support your claim at this time" but again Socrates says: "Can you tell us why? '' So it engages the student and makes them a better writer, giving them a lot more feedback than they could ever get. And we believe it will greatly speed up your writing, not hurt it. Now everything I have talked about so far is for the student. But we think it can be just as powerful for a teacher to direct more personalized training and honestly save time and energy for you and your students. So this is an American history lesson at Khan Academy. It is about the war between Spain and the United States. And for starters, it's in student mode. And when you say, "Tell me the answer," it doesn't say the answer. It goes into teaching mode. But that switch that teachers access, they can turn off student mode and then it goes into teacher mode. And it changes… It could be seen as a teacher's manual on steroids. It simply cannot explain the answer, it might explain how you might want to teach it. This can help prepare the teacher for this material. This can help them create lesson plans, as you see being done there. Finally, it helps them create progress reports and finally to help them criticize. So again teachers use about half of their time planning a lesson with this type of activity. All that energy can come back to them or return to human interaction with their real students. So you know, one thing I want to do. These great language patterns are so powerful, it's tempting to say, well, all these people are just slapping them on their websites, and it commoditized apps themselves. And what do I have to tell you? It's one of the reasons why I haven't slept in two weeks when I first had access to GPT- in August. But we quickly realized that to make it magical I think what you saw with Khanmigo, a little bit it didn't interact with you the way you see ChatGPT interacting. It was a little more magical, it was more Socratic, it was clearly much better at math than most people are used to thinking. And the reason is that A lot of work has been done behind the scenes to achieve this. And I could go through a whole list of everything we've been working on, many, many people over six, seven months to make it seem magical. But perhaps intellectually the most interesting thing did we understand and this was the idea of ​​an OpenAI researcher, so that we can greatly improve its ability in mathematics and its ability to direct if we let AI think before it speaks. So if you are training someone and you start talking right before you decide on their math, you may not be correct. But when you construct thoughts for yourself, and what you see on the right is a real AI idea, something that he himself produces but does not share with the student. then its accuracy improved dramatically, and its ability to be a world-class teacher has improved dramatically. And here you can see that it speaks for itself. It says, "The student got a different answer than I did, but don't tell them they were wrong. Instead, ask them to explain how they got to that point. " So I'll stop, I hope You know what I just showed you is only half of what we're working on, and we think this is only the tip of the iceberg of what it can actually lead to. And I'm pretty sure what I wouldn't have been even a year ago, that together we have the opportunity to solve the 2 sigma problem and make it a 2 sigma option, dramatically accelerating education as we know it. I will now take a step back on the meta level. Obviously we heard a lot of conversation today, from both sides. Some people are more pessimistic about artificial intelligence, they say it's scary because of all these dystopian scenarios. Maybe we want to slow down, we want to stop. On the other hand there are more optimistic people who say, well, we've been through breaking points before, we lived through the industrial revolution. It was scary, but it worked. And now what would I argue is I don't think it's like flipping a coin or is it something where we just have to like, wait and see how it goes. I think everyone here and beyond, we are actively involved in this decision. I'm pretty sure the first thought is actually almost a self-fulfilling prophecy, that when we act with fear and when we say, "Hey, we just have to get this thing over with. "What actually happens is that rule enforcers can disrupt, can slow down but rule breakers, as Alexandr [Wang] mentioned, totalitarian governments, criminal organizations, they just get faster. And this leads to what I'm pretty sure is a dystopian state, that is, good actors have worse AI than bad actors. But I also talk a little with the optimists. I don't think it means that Yes, then we should just relax and hope for the best. That might not happen either. I think we all have to fight together to make sure we put the fence, we put - if problems arise - reasonable rules. But we fight like hell for positive use cases. Because it is very close to my heart, and of course there are many possible positive use cases, but perhaps the most effective way to use it and perhaps the most poetic use case is artificial intelligence. Artificial intelligence can be used to improve HI, human intelligence, human potential and human purpose.