Sal Khan | Khan Academy - Transcript
[00:00:00]
Jeremy Singer: I'm Jeremy Singer, president of the College Board, and this is the Education Equation. I've spent my career grappling with what truly drives student success. On this podcast, I'll talk with people who are researching, building and scaling solutions that matter. Every episode will go beyond the hype and focus on data and evidence to see what's actually working.
Let's stop guessing and let's figure out what works. My guest today is Sal Khan Sal's, the founder and CEO of Khan Academy, a nonprofit with a mission of providing free world-class education for anyone anywhere. He's the author of Brave New Words, a book about the future of AI in education and co-author of teaching students to use AI ethically and [00:01:00] responsibly, a resource for educators, parents and school leaders.
Sal began his journey by tutoring his cousins remotely as a hobby. Over the past two decades, his effort has grown into a global learning platform used by more than 190 million registered users across 190 countries. Khan Academy offers free interactive exercises, instructional videos, and teacher tools across subjects like math, science, the humanities writing, and SAT prep.
The nonprofit is pioneering AI and education with Khanmigo, an AI tool designed to serve as both a tutor and a teaching assistant. Sal's been recognized as one of time's 100 most influential people, and recently was named one of the best leaders of 2025 by US News. Sal first congratulations and second, welcome to the education equation.
Sal Khan: Oh, no. Wonderful to be here.
Jeremy Singer: There is a ton we could cover, but I don't think anyone wants to listen to a four hour podcast, even if we do it [00:02:00] at two times speed. So I'm gonna fast forward past your origin story, which is wonderful. I think it's even better than Peter Parker's. And I'm gonna start when we met
a dozen years ago, I had recently left Kaplan test prep to join the college board and was committed to providing the best test practice for free. And I set two excellent colleagues and friends, Auditi Chakravarty and Stacy Caldwell, out to evaluate possible partners. And they came back with a recommendation of Khan Academy and based on evidence, they felt your organization and you provided the best opportunity for all students.
I and the rest of our colleagues at College Board are so grateful for the partnership. So we've never, I don't think I've ever asked you this, but from your perspective, how'd that process feel when we approached you and when we started this partnership?
Sal Khan: Yeah, it's interesting and it does go a little bit back to the original origin story.
I started tutoring cousins and then that first cousin that I was tutoring, I started tutoring her when she [00:03:00] was 12 Nadia. But when she was a little bit older, by the time she was 16 or 17, she's starting to get ready for college. She's gonna take the SAT and I remember, what can I do to help Nadia?
And what I literally did without anyone's permission, you probably don't advocate for this, I went and found one of those 10 SAT books and I recorded me working through every problem in the book, and I tend to be compulsive that way. And I said, Nadia, they're all there for you now. And when I was doing it, I was like, I don't know if the college board, I didn't know y'all then, but I was like, I dunno if they're gonna like this, I'm maybe providing some value.
And then once Khan Academy became a real thing, 5, 6, 7 years later. I said, wouldn't it be great if we were to partner with the college board? And this was before you and David were at the college board. There were some soft conversations, but I think at the time they were like, who is this guy?
And this kind of startupy nonprofit type of thing. And so when y'all came to us, I was like, yes, absolutely. there was always these conversations about when, I started Common Core was coming on [00:04:00] and I was like, isn't there already a common core? Isn't it called like AP and SAT?
isn't that what people care about? So yes, I definitely was excited and I give a lot of credit to the college board. I'm glad y'all, did that diligence and came to the conclusion that we were the best organization, but we were a small organization, as and it legitimized us even more.
I hopefully we were already getting, you were well on your
Jeremy Singer: way and we are glad to have help do our part. I should note, we celebrated our 10 year anniversary of our partnership. Tens of millions of students. Have taken advantage of the free personalized test practice resources, and we're going strong.
I just heard there's over 350,000 students in both reading and math that use it every month. So I'm gonna again go back to 2015 and we were ready for the announcement and launch of the redesigned SAT, and you generated so many great instructional videos aligned to the redesign exam. And as we approached the announcement, we belatedly realized that there were gonna be a large number of students
who [00:05:00] were gonna still take the old or the existing exam before the redesign was available, and these students, we realized would be quite frustrated not to have all these great new resources we were gonna announce. So I remember going to you and saying, Hey, would you mind after all this great work, create videos for the old exam?
At College Board, we said you went into beast mode and you just started cranking out these videos for the old exam, like day and night, 24 7. And you really took mythical status at college board. And I remember talking to you about it. I think we were in Austin, and I asked how you did and you said, I wake up at 5:00 AM and I think you told me you go into a closet and just start recording. And jokingly thinking of the Rocky movie and the training montage, I said, oh, did you, do you drink raw eggs? And you were like, yes. So I'd love to hear more and our listeners would love to hear more about the process you use for creating all the instructional content and importantly, like how has that evolved over the last dozen [00:06:00] years. Let's pause it before we get to ai 'cause we're gonna do plenty on AI in a minute.
Sal Khan: Yeah. Look, my, methodology for making videos has always been pretty much the same, which is I like to, if, it's something that has a lot of like content and I gotta make sure I spell things right or say things right, I'll do some research or something I forgot.
But if it's, something like math or SAT questions, have a little bit of pride, I'm like, I should be able to do all of this. And so I like to, I actually like to be "surprised", like when I'm recording it 90% of the time. That's the first time that I'm seeing that question. Yeah. That's amazing.
we hear one, it's easy, it's better for me. I don't, and it's usually zero edits and what we've heard from students over the time is they really appreciate that you can tell whether someone is reading a script or whether they're speaking. They're really thinking while they're doing it.
And we don't edit out if. Let's say for five seconds, I go down the wrong path for a second and then I back up and we've heard feedback from students like, wow, finally someone's showing how you really think about these [00:07:00] problems, as opposed to just having the perfect solution, like as if it always just comes into your mind.
So that's what I did then. I still do now. I still make videos. I'm not the only person making videos. I still make pretty much all of the math. Yeah. Videos, whether SAT or other types of math. I do a lot of the science videos. We have a, another gentleman who's hilarious, who does a lot of the, reading comprehension and grammar.
He's a personality unto himself and he probably has a slightly different process, but for me it's very off the cuff and unless I, only if I, it's a disastrous, something happens. Yeah. Will I, re rerecord?
Jeremy Singer: No. No. And it shows Sal it makes it very authentic because you really are grappling with the issue with the learner, and I think, we'll get back to it later, but I think that's one of the many elements of, why you've been so successful and effective. last quick anecdote, your reputation at College Board grew very quickly. And as we're headquartered in New York City, so we have a, fair number of Jewish staff.
And here we were working [00:08:00] so closely with this Bangladeshi American, and I cannot tell you how many people I had to correct and say, no, it's not SAUL. KAHN. It's S-A-L-K-H-A-N. So we've come a long way since 2015, I guess is the message. So can you share a moment when the usage or performance data really reaffirmed in your eyes the approach?
We took a proof point that told you this partnership's something, the type of impact we all wanted.
Sal Khan: I think in the, data from pretty much as soon as we launched it surpassed, it definitely surpassed our expectations. I think it surpassed all of our expectations. Sure. In terms of the percentage, huge percentage of students who were taking the SAT were using it.
The thing that always means a lot is, the data, but then when you start hearing the anecdotes too, where you know, people felt genuinely thankful that they felt appreciative. Even today when we. You know the number of times we're talking to a school district for the first time and I'm talking to the superintendent or the chief academic officer and they'll literally start the conversation with, [00:09:00] oh, before we get into it, my niece says, thank you for the SAT help.
That's like probably the most common thing I hear, and and sometimes the niece is now 35 years old. It's been a while. Maybe not 35, 25 years old. Yeah, we're still young.
Jeremy Singer: So are there. Things, thinking back again to 2015 to extend you can, are there insights that you didn't anticipate or things that, you talked about how it gained traction quicker, which we felt the same way, but other things you've learned over the last, 10 years from this.
Sal Khan: one thing and I remember we were there. There was these debates and we even had these debates. 'cause obviously the whole point of doing this is to improve access and improve equity. And I think there's a couple things. One is the positioning, and I think this is why y'all came to Khan Academy, is that we didn't view this as test prep.
We view this as college prep, as evidenced by the test that shows how ready you are for college. And I think y'all appreciated that. on the SAT prep. Or on the examples, I usually would say, you should know how to solve [00:10:00] this, and I would show you how to actually do it. I might have given a few tips on you could have eliminated this or this, just by inspection, et cetera.
So that's one. I think another one, it really did. Fine tune. Even my notion of what Khan Academy as a whole is trying to do, because there's always this fear that, okay, if you put something online that requires an internet connection, maybe only the people who have an internet connection, now, especially in the us, almost everyone has one.
Yeah. That was definitely a fear 10 years ago, and I remember us talking about, oh, if a resource is somehow, if the demographics are disproportionately kids who went to higher education, et cetera, does that mean we're somehow driving inequity? And I think it really put up a fine point on this idea of.
No, imagine what happens if you take these resources away, because as we know, there's a ton of students, low income students from underrepresented groups who are using these resources. And if you take these resources away, the kids from upper middle class, well-educated families, they'll just go to other resources.
Not a big deal. They'll spend a little bit of [00:11:00] money for it, but the other kids, and we're talking about hundreds of thousands of kids, if not millions, yeah. Will have nothing. And That journey that we went on together. we've always tried to make sure we're reaching as many kids that need it as possible, but it's really put a fine point on how important this work is.
Jeremy Singer: Yeah. and we'll come back to it later, but one of the many things we get to your book, one of your books that I loved is, you mentioned Perfect Not being the enemy of Good. Comparing things, not verse absolute or purist, but verse the alternative of the reality we live in now. And so I think there, you're right.
is it perfect? No, I think we all acknowledge it, but is it better set of resources than otherwise? And I worry sometimes. In this space, the purists can kill really good work. So thank you for pushing against that. So moving away from our partnership and just more broadly how wonderful the Khan Academy has grown and its usage over, as I [00:12:00] opened up with 190 million registered users, et cetera, I watched, at least my perception was there's a little bit of a repositioning or rethinking.
Initially, very early, it was almost supplemental to what happens in the school, and I've observed you seeing and thinking about how do you better integrate it into the way students are learning with teachers, et cetera. So we have a fair number of teachers, instructional leaders who listen to this podcast.
So share that evolution and how you got there and where you're at.
Sal Khan: Yeah, no, there, there has definitely been an evolution. In the early days, we definitely viewed this as something that hopefully teachers would find useful. I remember I made the first teacher tools, teacher dashboards back in 2007 when it was just me operating out of a cla.
I'm actually still in that same closet right now. It's a nice closet. No complaints, but you should build yourself a bigger closet. no, I like it. I can touch football. So we always had a lens that. Look, we wanna be a safety net for the student out there who might not have access to school or doesn't have access to [00:13:00] tutoring.
Maybe they're struggling in their class and they need some supplemental learning, but ideally, we could also be used in a classroom. There's one teacher, 30 students that. Teacher knows that these kids are all at different levels, have different gaps. If they could use a tool like ours, maybe they could personalize the pacing more, give students the opportunity to fill in their gaps, move ahead, slow down, whatever they need.
And you fast forward to about, I don't know, seven years ago we had many. Thousands, if not tens of thousands of teachers using it in their classrooms in this way. And we had a whole series of efficacy studies. We still do 50 plus that if students were to engage in personalized practice, even 60 minutes a week over the course of a school year, they're accelerating 30, 40, 50%.
And some of these are quite causal, Yeah. RCTs. And we recently, the one that I think is even a little bit better than an RCT, if possible. So we started going to school districts. We say, Hey. A bunch of your teachers are already using Khan Academy. Why don't you just use it district wide?
And the school districts, after they've told us how, [00:14:00] thank you for their niece on the SAT prep, they would say, we believe Khan Academy would work. But for us to use the district wide, we need support, training, integration with our rostering systems, district level dashboards. You need to sign our data privacy agreement.
You gotta, meet our accessibility guidelines and. That is a lot of work and you have to align to all of the different state standards or even flavors of the state standards.
yep. but we were like, if, our mission free world class education for anyone, anywhere, if we're serious about it, if we really wanna move the dial for.
All people, districts are the way that, that you, need to reach that. And if you want to engage students, we can make all the nice bells and whistles on the product, but really the, way that you're gonna engage students is through teachers. And if you can help teachers, help students, then everyone wins.
And yeah, we've been going on that journey for about seven years now. And, there's approaching 2 million formal school partners. There's many more millions using us informally. Yeah. But these are the places where it's a, deeper partnership. [00:15:00]
Jeremy Singer: Yeah, no, it's great. And I, I can hear from when I talk to teachers, the sentiment has grown so much more positive over the years about your work in Khan Academy.
So I feel that from the start of our partnership, we were both realistic and knew that for so many students, regardless of how good your tools were and everything else, that there still was a role for some sort of human support, whether it was a live person to explain a concept. Someone that the student would feel accountable to do the work when they weren't in front of them.
A cheerleader, a friendly ear, you name it. And the reality of the majority of learners would not succeed with solely a set of asynchronous instructional tools. And so we tried a college board, a lot of different partnerships to bring the human into the mix. We did a partnership with Boys and Girls Club, but nothing worked, at least at scale.
And then you come along and you start Schoolhouse World. It wasn't like, [00:16:00] I don't think you were bored. You're pretty busy, and I don't think you were having a midlife crisis. So what led you to start that nonprofit?
Sal Khan: I'm glad you think I wasn't having a midlife crisis, the pandemic hit and I had always had a, vision or an idea that even in, in traditional classrooms that we're using Khan Academy, the best implementations weren't just kids on their screens in their own lane.
It would be that there's some mechanism where students who might be a little ahead could be a little bit of a TA and tutor some of their other kids. So there'd be like this hum of kids working with each other, not. Cheating, but supporting each other and tutoring. And even in the earliest days when I was working with my cousins, that first version of Khan Academy, I had a little shared blackboard, and in a lot of cases I would set that up with my cousins.
I was like, okay, we covered this last year. Can you tutor your other cousin to do that? So I've always been a big believer on peer to peer. When I was in high school, I was a peer tutor and my high school took it so seriously that this tutoring program that. The math club ran and I, no surprise, I was the president of the math [00:17:00] club.
They made it part of the school. I didn't the chess club, for the record that's nerdy at my school. Any student who was getting a C or lower in their math class had to go to the tutoring that was run by 16-year-old Sal and. We were effective. we took some kids who were failing their math classes and they joined the math team and they became, so that was actually my first exposure to, if you just gave someone some attention and sometimes the near peers could actually give better attention.
They knew where those students were. they could speak their language. Yeah. And so the pandemic hits and. Khan Academy usage goes through the roof. But we also knew that people were isolated. They were on their own. And that's when the idea said, look now or never. What if we were to, reach out to Zoom, get a bunch of free Zoom licenses, which they did to their credit, and created a system where students could say what they need help in.
And other students, when we'd have to verify somehow that they know the material can go and be the tutors. And it worked. and so we started scaling and then obviously we've always been partners with y'all. [00:18:00] And we started thinking about, oh, maybe this could be really powerful for the SAT.
And there was already a little bit of informal SAT tutoring going on. And over the last, what's it been, a year and a half. Almost two years. Yeah,
Jeremy Singer: it's a little longer. So I'll jump in and we can break some news here Sal, for our listeners. So College Board and Schoolhouse World are launching a new effort and we're gonna offer free peer-to-peer tutoring for SAT preparation.
And it's adding a new dimension to already. This wonderful world class free set of practice resources that we have together. So just some data since 2022 was longer than we thought. I, when I saw the data, I was also surprised. We have piloted four week peer tutored, SAT bootcamps. There have been more than 30,000 sessions, 181,000 registered learners.
nearly 20,000 peer tutors and over 98. Percent of the sessions have been rated helpful by learners. So all this, as you were saying, gave us confidence and you confidence in the model. Can you share any data or [00:19:00] feedback from that experience that gave you confidence that the model could work?
Sal Khan: Yeah, I think one of the really powerful parts of.
The pilot that we've had, and now we're gonna be, going bigger with this is one of the main questions is when someone is a tutor, how do you validate that they're qualified and being able to partner with y'all and essentially being able to recruit from the students who are already operating, near the 99th percentile on the SAT was a great way to validate.
And then also a lot of colleges started saying, Hey, yeah, if someone is an SAT tutor and a SAT scholar, that's something that we would be interested in. And seeing it, it shows not only. Obviously, they could already see their SAT score, but this person can communicate, they can lead. There's all sorts of these higher order skills that it starts to evidence.
So that was great that, there's consumers for this. Most of the students are doing it just 'cause they really enjoy it, but it's good for them to know some of this stress around college admissions that can be channeled into something that can help your peer as opposed to try to feel like you're competing, with your peers.
So that's been nice. [00:20:00] And to your point, about 98% of folks saying it's been helpful. when we first launched our first version of, let's call it the digital SAT prep, and I would still meet students who were paying for test prep, and I said, why are you paying for test prep? we have all the studies.
this stuff that we're doing with the college board is as good as it gets. They're like, no, we know we're using that. And actually, in many cases, shady, the paid test prep was also using the tools that we created. But they said, I just want someone to hold me accountable, and it just makes me feel better.
And I'm, and I'm, I like being part of a cohort. And so now that we've been able to do these bootcamps, honestly, I think the, reviews we're getting from the bootcamps are better than when someone. Even if they were to get the, I won't name names, but if they were to pay $5,000 and get that, paid test prep, they're liking these bootcamps even more.
And both sides are benefiting. My son is a, is an SAT tutor and he's tutoring in his bootcamp he had, two young women, one from Mexico City, one from Connecticut, and it's just, there's this cultural exposure that, that you get too. [00:21:00] we started schoolhouse as a way to help people help each other academically.
But I think this, these. Connections, human connections that have started to form has been a pleasant surprise.
Jeremy Singer: I think it's great and one, obviously together we've done some research and one of the things I think it would make our mutual friend Angie Duckworth excited was students who complete the SAT bootcamps show significant gains in confidence and specifically in growth mindset.
We've also seen score gains compared to other interventions. So yeah, we think there's a lot here and we're super excited and particularly that it's free. I'll also add one more piece, which is. I think clearly you've thought schoolhouse, that world has thought so smartly about how do you motivate to get people to volunteer, to be tutors and create credentials and something that's meaningful and you talk about in your book for college admission.
So the whole system works in my mind. So congrats. thread throughout all these discussions, we both speak about the [00:22:00] importance of the pedagogy of the materials or instruction. But we also need to make sure that it's engaging because if, no matter how good the pedagogy is, if students don't spend time on tasks, they're never gonna succeed.
And frequent. Critique of something like Duolingo is, they're great at engagement and they get these kids to do these streaks for 300 days, but maybe not so much on the pedagogy. you've been working on both sides of that and can you share the idea of knowing medicine can always be made to taste good, how you've succeeded here.
Sal Khan: It's an ongoing journey and Duolingo does a lot of great things and from them and the gaming world, there's definitely a lot of things that we have historically put into Khan Academy and we're going to put more into Khan Academy with badges and streaks and, fun classroom level competitions and things like that.
But at the end of the day, the most important game mechanic, and I'm not saying that as a pejorative yeah. In the education system is the teacher one. There's just a very. extrinsic motivator of people [00:23:00] wanna pass the class and if the teacher says you need to do this to pass the class, they're more likely to do it.
But also, as you mentioned, the teacher is there to, Hey, why are you stuck? How can I help you? That was a great thing you just did. Or, Hey, why are last week you were pretty engaged, what's going on this week? and so that's where, that's why we're working with districts. That's why we're doing teacher training.
That's why we're working closely with them. And so I think to your point, it's both like we, we definitely should be investing in ways to make it. More engaging. I a hundred percent agree with you. If someone is engaged, you can give them a 19th century textbook and they will learn a lot. And if they're not engaged, you can give 'em the fanciest vr, ai, whatever.
They're not gonna learn a thing. And so we have seen that too. And so for us is, yes, let's make the platform by itself as engaging as possible, but we can't water down. The content. Yeah. Content has to be efficacious and but there's human systems we can leverage and that could be the teacher in the classroom, that could be the superintendent celebrating, we're talking to some countries now and we're like, look, we're not [00:24:00] here to quote, sell anything.
If you are going to invest in your schools, in your country using Khan Academy, it's only going to succeed if you are. one of the countries we're talking to has a king. If you're king, like every month, like celebrates the schools and the teachers that are using Khan Academy, then the culture in the country is going to change.
So I think it's the human systems plus the game mechanics.
Jeremy Singer: I was sure you're gonna say my cousin, who's the king? Why
Sal Khan: not that blue blooded? I can't believe we've
Jeremy Singer: been talking this long and haven't talked about ai, but now I wanna shift to ai. You really have been focusing on this and been a leader in this.
You've published two books. I listened to Brave New Words. Which is excellent. I was even more impressed that you actually voiced it. And anyone who can reference Noam Chomsky, Ethan Mooch, Francis Ford, Copeland, Donald Trump in the same chapter has done something really significant. I should also note this small thing that you launched Conmigo, which has really done pioneering work in the AI space, particularly around tutoring.
[00:25:00] With all this, by the way, you make me and others feel quite small, not accomplished much, but. To start, share the moment when you first grasp what a game changer AI could be to tutoring or more broadly to teaching and learning.
Sal Khan: we were lucky. And look, there's a whole series of things. I'm a very rational person, I think, but I have sometimes this, let's call it a tongue in cheek theory, that benevolent aliens are helping Khan Academy prepare humanity for first contact.
One example is y'all reaching out to us when we were still a fairly small organization saying, you wanna be the official practice for the SATI was like, yes, please. Fast forward to about three years ago, we had another one of those. Benevolent alien moments where I got an email from Sam Altman and Greg Brockman from Open ai, and this was six months before chat GPT came out.
Yeah. No one was really caring much about this even. I was like, yeah, I know these guys. They're the real deal. And they said, they're Baptist trained, their next model, which would end up being GPT-4, and they thought this was gonna be the model. They said we wanted to launch it with some socially positive [00:26:00] use cases with organizations that could leverage it.
We thought of Khan Academy. Would you like to see it? And I didn't think it was gonna, I get, I'm sure you do too. Every day we get five or 10 people who wanna partner with us, et cetera, et cetera. I'm like, it's not gonna end up, but I'm curious what it's gonna be able to do. So we had, a virtual meeting and they showed me, and we're all used to, we've all experienced chat, GPT and Gemini and everything else for the last several years, but you could imagine.
When it was able to answer and, relate to y'all an AP bio question and I realized in hindsight, I dunno if y'all have heard the story, the reason why they were showing me an AP bio question, they asked me to answer it in the ai Yeah. Apparently when they did GPT-3. Bill Gates said, this is cute, but I'll only be impressed when it can pass an AP bio exam.
Yeah, we know that origin. yeah, for sure. so I was a pretty, and then I said, oh, I have it. explain why the other ones aren't the answer. It did. and I was like, okay, this is. Strange. Yes, I would like access. So I signed all these non-disclosure agreements, et cetera, and just to appreciate how early [00:27:00] this was that day, they gave us access and I think we were the first, myself, Christian, our chief learning officer, we were the first people outside of Open ai.
I don't think even most of Open AI had access. I asked them, I was like, oh, does this work only in English? And they said, yeah, it works only in English. And then I was just curious. I immediately, I was born in New Orleans, my family were Bengali. I can barely speak Bengali. And so I tried to speak to the AI in Bengali and it responded back to me in Bengali.
And so I took a screenshot and I sent it to Greg and he's oh, yeah, after you asked, we tested it. It looks like it speaks every language. Even for them it was emergent behavior.
Jeremy Singer: That's amazing.
Sal Khan: but as soon as, I think the biggest thing. And this was what was, I think it's very surprising for everyone.
We're used to it now, but science fiction always imagined robots talking like this and being very non emotive and not understanding subtlety, but being very good at math and being, having a good facts. People
Jeremy Singer: don't a video so they can't see your arms moving. Oh yes,
Sal Khan: exactly. The robot arms.
Jeremy Singer: Yeah.
Sal Khan: But here was the thing, I remember I the Donald Trump reference, one of the first things I did is I [00:28:00] asked it to rewrite the Declaration of Independence in the tone of Donald Trump.
And, regardless how you feel about Donald Trump, it was hilarious. and said, it's gonna be the best country ever. George II's the biggest loser. I'm like, yeah, he was a loser. I can, But, and then I had it, write it in the tone of Dr. Seuss and then I just took a. I didn't think it was even gonna work.
I, I said, you are an empathetic tutor. You are the Robin Williams character from Dead Poet Society tutor me. And it did look, it had the issues. It was making math errors embarrassing. Yeah, it was making up facts, but tone wise, it was incredible. And that's when I said, okay, this is. This has been the dream of EdTech for a very long time.
To be able to get that much closer to emulating a tutor and a teaching assistant in, in many cases. But, you could imagine when I brought this to the team and we got more people into the fold, we had all the debates that folks are still having. This could be a cheating tool. It's,
Jeremy Singer: yeah.
Sal Khan: How did math, it's gotten a lot better since then.
it hallucinates, what if kids have shady conversations? And I said, look, let's just turn [00:29:00] those fears into features. This is too big of a deal. And look, the models are going to improve fast. And we've gotta, if we can help people with this, we've gotta try.
Jeremy Singer: Yeah, no. And I agree.
Again, it goes again, back to perfection being the enemy of good and trying to make progress. And we're gonna get into a lot of the details on those things. So let me start with. In your book you talk about, and I'm complete agreement, that there's these things that have been, they're not new, like 20, 30 years.
We've been talking about personalizing the learning experience based on the individual student. It makes perfect sense, and we've talked about mastery learning, relatedly, where a student goes at their own pace and based on what they know, they see the next set of learning opportunity. And even for the classroom, it's 20 plus years ago we were talking about.
The opportunity for flip classroom and some are actually doing it. But despite these all being clearly good ideas, and a lot of smart people have tried to tackle it, but very few have scaled and worked. And so [00:30:00] you make a strong argument that AI can enable all of this. So what have you seen where this has manifested or where you see this potential around these kind of things?
Sal Khan: And even on some of the things you just mentioned, sometimes I feel exactly you just said, oh, these are good ideas, but we're, but they actually think they have had their started to have effect when, 10, 15 years ago compared to now. Whether it's a, a pure flip of a classroom, you definitely see a lot more classrooms that are indexed on active learning versus, students passively sitting and taking a lecture in, which was the norm for most of.
Human history. So you do see a movement there. I think especially after the pandemic, you started seeing people talk about unfinished learning and ways to address that, which is shorthand for mastery learning. yes, the reason why so many people are struggling is they have so many gaps. don't even get me started on, some of what we've been seeing in California.
And look, I'm not ashamed to say, when the uc system said no standardized tests and University of California, San Diego says, if. 30 times as many people [00:31:00] now who can't engage in essentially middle school math. That's one. Yeah. Don't have the data. and look, everyone in California is yeah, it's obvious 'cause they got rid of standardized tests, which are measuring how ready you are.
And the irony is they're rejecting a ton of kids with. 15 hundreds on their SAT because they don't know they got a 1500 on their SAT and they're accepting a lot of kids who, can't, who aren't ready for algebra yet. And you're not doing those kids a favor 'cause they're go showing up on campus.
They're out of their element. They're probably gonna drop out. You're not doing them a favor at all. But this is all a byproduct of unfinished learning. right now or historically. These kids were sitting in classrooms getting promoted year to year. They're taking Algebra one, algebra two, sometimes pre-calculus, sometimes even calculus.
And they're getting A's in these classes, and then it turns out they're not even ready to learn algebra yet. So there is that history there. I think there's a lot more that we need to be able to do, but my hope is that we can get that much closer now with the tools that are emerging. if we were talking about two years ago, right when all the AI stuff was like super hot, I was [00:32:00] also guilty of probably over indexing on the ai.
yeah. It probably because I assumed that the AI progression from GPT-3 to four to, five was going to be exponential, but it's looks like it's asy toting a little bit. Yeah. Yeah. But I still think adult does go back to engagement. And if we can, if the ais can help teachers, help the students, if it can help save them time with, things from lesson planning, eventually even grading papers to some degree, highlighting what students need their support.
Most of EdTech, including us, would just give teachers all these dashboards and expect them to be data scientists. If we can just give narrative insights and just very actionable things to do. Save teachers time 'cause they're overworked and spread thin. And then yes, in the right moments when if a student is stuck on something, historically they've had videos and solutions and articles.
If the AI can be another layer of support or motivation, that's great. Yeah. But I do think it's not just AI and isolation and I don't think AI by itself is a silver bullet. At least not yet.
Jeremy Singer: I [00:33:00] want to jump into all this and there's so much here, and I actually may be more of a bull on AI than you. You make a clear case in your book that when AI is designed well, it can help students learn but not answer the question for them.
And as college words working on these same things and this is a hard needle to thread. How do you build the guardrails and when it falls short and AI actually has replaced the actual student learning, like what are you seeing in that? With all that you're doing with Amigo and other things,
Sal Khan: it is this moving target.
Obviously, there's some basic things we've done from day one, which is even on the basic prompting, you're Socratic tutor. You can nudge the student in the right direction, but not give them the answer. But as we all know, if you've ever tutored a student, that's sometimes hard for a human tutor to do.
I help my kids when they're doing their homework, and sometimes I do feel like, oh, I'm doing a little bit too much. I've gotta back off a little bit. But sometimes they do need a little bit of a nudge. You don't want them to just. Flail all night. That's not going to be constructive either. So it is a tough thing to thread.
One of the [00:34:00] hardest things actually is the evaluation of the students, because you can imagine, let's say the answer is one third and the student puts in 0.33, right? They're not quite right, but they're not. Super wrong either and so we've done a lot of work there. We've built a whole infrastructure around test cases that have been very difficult.
And then every time we do new tweaks to the different layers of reasoning we do, and we're trying with new models, we're able to test how does it handle all of these edge cases. And so that's given us confidence. We're moving forward. We also have edge cases on errors, on the evaluation errors or even math errors and things like that.
And they're now sub 5%. And in fact, many cases sub 1%. And this goes back to not letting perfect be the enemy of the good. I remember when we saw those numbers, we're all wanting to be 0%, right? But then I was like, I wonder what I am when I tutor my, when I tutor my own family. And just the other day I was tutoring my 11-year-old and I was like.
We did the problem together and we got an answer and he put it in on Khan Academy. He got it wrong. And he's [00:35:00] what you mean Khan Academy? And you got it wrong. Oh, negative sign, right? so I think my error rate is actually probably higher than the AI at this point. It is good to remember that.
Jeremy Singer: That's very fair. And I wanna press a little on, you write thoughtfully that AI is a teaching assistant, the AI tutor for existing teachers and won't displace teacher. To be clear, I don't. I think teachers are gonna go away, but I wonder, I think their role could change more dramatically than just sourcing some of the administrative and planning tasks and grading tasks to ai.
Let's talk about that, what do you see if you think three years from now, how would you think the teacher's role is different? what are they actually doing with students when they're with them?
Sal Khan: I'm convinced that teaching is one of the safest jobs in an AI world. I think any job that can really lean into the human element is gonna be safe.
I am worried. that's a whole other conversation about if you're a rideshare driver, if you're a call center worker, if you're,
Jeremy Singer: yeah, me too.
Sal Khan: Anyway, that might be my next book actually, but what we need to do on the re-skilling side,
Jeremy Singer: universal [00:36:00] basic income, I think.
Sal Khan: Yeah. I'm a fan of universal basic work.
Anyway. That's a whole other, fuck. Yeah. Yeah. It's good for now. Yeah. I imagine a ideal AI enabled classroom in three years or five years. Right When you observe it'll just feel like a good classroom, a good classroom today. I think the teachers walking around, the kids look really engaged.
They're actively doing something, that they find engaging and is actually on standard and useful for those kids to engage in. I could imagine we, I was just in Newark, New Jersey. I've seen some great classrooms with great teachers. one of 'em had a station rotation model where.
About five or 10 kids were doing their Khan Academy. Another five or 10 kids were working on a project together, and then the teacher was doing a focused intervention with another five or 10 kids. Very interactive was great. You can imagine other classrooms where it's more Socratic dialogue and you're like, oh wait, where's the ai?
The AI is almost invisible, but it's empowering that scenario. So some of it is happening outside of the classroom as we were talking about before. Even when there is a very prescriptive curriculum, [00:37:00] we know that it doesn't work in the reality. Oftentimes for the teacher, they have a different amount of time.
Kids need something else, so they're modifying it. They have to do lesson planning. That takes hours a week. They're grading papers, they're doing all that. Hopefully, AI is by a factor of 10. Taking that off of teacher's plates and when the teachers are thinking about what they wanna do in the classroom.
They're able to modify that even more. I actually think in about five years people will realize these very prescriptive curricula don't make sense anymore. Yeah. That actually what they're going to be is like guidelines for an AI to work with the teacher to create custom curricula that are still on standard for what those kids need.
But you can imagine when they're in the classroom, the teacher, this might feel a little bit scary, but if we get the privacy and all that, the technology's already there where an AI could observe the classroom, right? It can see if a student is not on task and it can whisper in the teacher's ear.
Maybe they have their AirPods in and say, Hey, go walk by Billy's desk. Just walk by it and let's see if it, let's see if it gets him back on task. Or, [00:38:00] Susie. Was all year really engaged. But this, today she's, why don't you go tell her this, that you really appreciate this about her?
And with the other a thousand teachers, I've recommended that to, it really worked with students like Susie, or, Hey, this doesn't seem to be working for the class right now. Let's pivot it quickly. every teacher's been there, they have this great lesson plan, and then they bring it to the class like, oh boy, this is falling flat.
in real time let's pivot this thing. And yeah, the teacher's spending a lot less time on the. The administrative paperwork side of things, grading side of things, the teacher spending a lot more time engaging with their students, but they'll look like they're magical. I raise a lot of money for Khan Academy as a nonprofit, and before I go to a big fundraising event, there'll be someone from my philanthropy team saying, Hey, make sure to talk to that person.
They gave us some million dollars. They care about this, right? Teachers need that too. Hey, go talk to Billy. He cares about this and he needs help right now.
Jeremy Singer: I hear you. I do think a lot of instructional stuff, like if you could outsource the fundraising to someone genius of framing it, you would too. [00:39:00] It's interesting 'cause my daughter Izzy, works at a school called Winston Prep, which is a school for students with learning differences.
She got her master's in social work and she's a focus teacher and it's amazing that the support these students get. She has just seven students, so every student in school gets a personalized and she meets with each of her seven students for a full period every day, five days a week. And she does.
Social emotional stuff. She does some incremental instruction and so forth. And I do wonder whether that is more the future of the teacher doing that kind of thing. The more that the AI can do the core piece of instruction, we can debate that another time. So in your book, you delve into AI and mental health and you share that loneliness, isolation among young people as an epidemic.
I think we all see the data and agree and you share that AI can provide some support to people who feel isolated. My worry in reading that and thinking about that and following all this, is the more we condition people to find the connection through the chat bot or whatever, the less likely they're gonna either need or [00:40:00] forge connections with humans.
So it feels how do we square this circle? 'cause I both agree with you that it can help but worry that it can also harm. And so I don't know what the answer is.
Sal Khan: Yeah, and I don't claim to have the perfect crystal ball, but I think, in moderation it can actually be very powerful.
I have a family member who's trying to navigate the mental health space right now and. It is so hard to find help and so expensive, And we're fortunate. Consider a solidly upper middle class. And I'm like, what? Yeah. Yeah. That's how much it costs. And so it's extremely hard to do it. And actually the same family member has found, it's not like they are giving up on the traditional system.
We're trying to navigate that, but. This person has taken solace in working with chat GPT actually, and it has no, and there's a lot
Jeremy Singer: of research that the virtual therapists are actually effective. Yeah. So I agree. Yeah.
Sal Khan: And I can imagine both. it goes back to the teacher analogy. If I'm a therapist, I get to see [00:41:00] my patient once a week.
what else is happening between, so if they're able to talk to a virtual therapist, an AI therapist, maybe have a virtual therapist that I prompted to have the same style, and maybe it can observe even our therapy sessions too. So it has that context. Yeah, that I have the context of what's happening during the week, and it has the context of what's happening in the therapy session.
That's more. 360, it surrounds the user. there's a lighter story from my own family in the book. I talk a lot about my daughter. She does seem, she does. She's very reluctant to be included in any of this stuff, but she's always my best source of stories. This was about two months ago.
So this isn't in the book. She's 14. I have a 16-year-old, a 14-year-old and 11-year-old. A boy, girl, boy. And our 11-year-old is always, she gets annoyed by, I miss classic stuff and she's sometimes honestly, a little bit mean to him. I would always categorize it as bullying. And about a month or two ago, we see her being shockingly nice, like her friends comes over.
They're gonna watch a movie in the garage and they're like, Hey, za, do you [00:42:00] wanna watch it with us? I was like, what's going on here? this is, and he was so excited. And when I got her alone, I said, Hey, yeah, I noticed you're being very kind to your brother. What, what's going on?
And she said, I didn't even know this was happening. I've been journaling on chat GPT, and I was complaining about aza, how annoying he is and blah, blah, blah. And it told me that. And she has her AI talk to her in Gen Z slang. So it's yeah. I see what you're saying, it's your little bro, and you know you're gonna go to college in four years and trust me, you're gonna miss him and you're gonna reminisce about these days and it's only four years and you're gonna wish you included him.
And it made the difference. It had a noticeable positive difference on her. If my wife or I said that, she would've just rolled her eyes I think there's positivity to it, but it's like anything, we want our kids to read books, but if they do nothing but read books and they don't go hang out with friends, that's not good.
Similarly, if they can get a little bit of help from this, that's great, but, I also [00:43:00] hope that the ais can say, Hey, we've talked for
Jeremy Singer: yeah,
Sal Khan: an hour. Why don't you go outside or Yeah, go do something. Yeah. Start talking to your dad about this, or, do, who's your closest friend? Oh, I'm your closest friend.
no. You need someone else. Things like that.
Jeremy Singer: I can't wait for your book on parenting tips that's coming next. I wait for
Sal Khan: my kids, or at least, I gotta make sure that they're not, anyway,
Jeremy Singer: yeah, so AI is moving so quickly. a month in the AI world feels a year or two years, and it makes it even more impressive that your book Brave New Words, which was published I think in March.
And you, I'm sure you were finalizing even well before then, is still I, literally listened to the last two weeks and still so on point. With that said, what's changed or surprised you since you published that? You're like, wow, that went faster, went very different than you had thought.
Sal Khan: The stuff that's going slower than I expected is if you look at.
The latest, Gemini three or GBT five, or Claude Opus 4.5 or whatever. [00:44:00] They've gotten a lot better in important ways. They're hallucinating less, much better at math. they're incredible at coding. I think, they're, the video generations, all that stuff has surpassed expectations.
They haven't gone as far as I thought they would by this point. If you extrapolated, if you took the trend line from three to four. Yep, And it's reinforced this idea that AI. At least immediately isn't going to eat everything.
Jeremy Singer: Do you feel more relieved by that or bummed that it's, it isn't going at this, as
Sal Khan: you mentioned, it's still going super fast in other areas.
Yeah. Yeah. I feel, honestly, if I'm honest, a little relieved that we have more time. I remember telling the open AI folks, if you just froze. At GPT four for the next 10 years, you're gonna keep us busy. Yeah, exactly. I feel the same way. so it's, there's a lot there. And maybe 'cause they haven't tried it, it doesn't make a Khan Academy explainer video.
there's people who've made explainer videos, they're okay, but you can tell they're AI generated. They don't have the same feeling of oh, this is some dude in a closet trying to work through this, that, seems to have care about me. Things like [00:45:00] that. They, haven't even done, I know.
Y'all have to do a lot of item generation. We have to do a lot of item generation. Yeah. I think we were both hoping that it could generate very high quality items. Fast.
Jeremy Singer: Yeah.
Sal Khan: You killing
Jeremy Singer: humans. It can help with some of the detractors and stuff. That's a great one. Yeah. You also coauthored teaching students to use AI ethically and responsibly.
For districts and school leaders and teachers who are listening and working on implementing ai, I can't miss the opportunity for you to give them advice on, common pitfalls, evidence-based strategies, like what would your advice to them be?
Sal Khan: I'd say the first is don't make kids have to navigate some weird gray area, and I don't think.
you might think you're setting a bright line by saying, don't use ai. You cannot use AI for this assignment. it turns out as soon as you do a web search, you've used ai, right? As soon as your grammar checker, you've used ai. So you're putting kids in this weird gray area. I would say if you really wanna make sure kids are doing it themselves, do it in class.
we're working [00:46:00] on some tools that'll make the process transparent, but it's using AI for that. But it's, it's a Socratic ai
Jeremy Singer: Yeah.
Sal Khan: Or lean into projects where you just tell the kids, use ai. I was meeting a family friend who's a freshman at Stanford, humanities major, and he was in midterms and it was a bunch of take home assessments and I asked him, I was like, what percentage of your peers are using AI for this?
He's oh, 95%. Then I'm like, oh my God, how does that make you feel? He's makes me feel horrible. 'cause I'm in the 5% that's still, trying to learn. And it makes you wonder okay, on some level those students, that 95% are doing the right thing because when they come into our organizations, we want them to be able to write more, better, et cetera, et cetera.
But. They're not learning what Stanford thinks that they're teaching them. Yeah. and so there's a real problem. So don't put students in that. And, I, I think honor codes, this might be controversial, are a signal for a lot of dishonor. So be very clear with folks, don't think that you can detect ai.
These AI detectors have a 40% false positive rate. And now there are these apps called humanizer. [00:47:00] Yeah. Which are to have,
Jeremy Singer: technology's gonna outpace anything we can do to detect Yes. That's a fool's errand. Yeah.
Sal Khan: and then, yeah, and especially in high school and college, lean into it.
Tell the kids to use AI if you want them to know how to write by themselves, which is important, I think. Do it in class and, do a flip classroom as we talked about before.
Jeremy Singer: Yeah. No, and we are working on tools as well as you are where there's parts of the process you want them to do without AI authentically, and if that's in class and so forth.
But other stuff you have to lean in and I do worry. About an arms race kind of thing, where a student who wants to do the right thing, but everybody else is, and it's tricky. So I always ask each guest four rapid fire questions just to get to know you a little better. Just very quickly. So one educational buzzword you wish we could retire.
Sal Khan: Boy, lemme think fast. This is supposed to be rapid fire. The one I was gonna say, I don't wanna trigger people. I'm varnished. Come on Sal. What came into my mind was the high quality instructional materials. Okay. And not that I think that they're bad, but I think we're going into a world. I think the reality is most [00:48:00] teachers view them as suggestions, not as exactly what to do, and they do need to be customized more.
So I'm fewing of, I like
Jeremy Singer: that. That's brave. And I think they can be overdone. your favorite book about education or one that deeply shaped your thinking? You can't be your own book, I should
Sal Khan: say. Yeah, that would be circular. I'll reference, science fiction. I reference the foundation series a lot.
And it's not an obvious book about education, but it's a world, it's a universe. It's a galaxy where 30,000 years of the future where someone says, Hey, I wanna shorten the dark ages of the galactic empire and I'm gonna do it by collecting all of the. Galaxy's knowledge and a foundation, and I read it in seventh grade.
It was very inspiring. And I still, and when I'm having more of my, grandiose thoughts, I think of, maybe Khan Academy can be that foundation for us.
Jeremy Singer: Awesome. Love it. One thing that makes you bullish on the future of learners.
Sal Khan: When I see some of the strongest students that are coming out of our system, they're very strong.
they are people that will give anyone their run for their money. So that, [00:49:00] it sounds cheesy, but when I see some of these young people, it gives me a lot of hope.
Jeremy Singer: Yeah, fair enough. And last question, what one class do you wish all students had to take?
Sal Khan: Oh boy. I'm gonna be a little bit pragmatic here.
It's actually probably a toss up between a finance accounting class and a legal class. Like these are both two things that everyone should know and it's mind blowing that they aren't part of the, financial literacy is a basic thing, just to know the basics of how organizations, anyway, either that or law.
Jeremy Singer: This is not a plug. College board's launching, an AP business and personal finance class. So if you knew that, I would feel even better, but,
Sal Khan: oh, you gotta do an AP law too. That'd be awesome.
Jeremy Singer: Okay, we'll look into it. But imagine, we've talked about the role of ai. Imagine we're having a conversation three, five years down the road.
What's a dream scenario where you really like, look back and say, this evolved the way you could have hoped it could.
Sal Khan: Yeah. I hope we're, as I said, we're looking at classrooms that are more interactive, more joyous for everyone, the teacher, the [00:50:00] students, people are seeing outcomes move in the right direction.
I think another area that, I've talked to you and David about a little bit. I hope that in five years we see new types of assessments. I think the things that we're already assessing are important, but I think there's other things that we can assess that can layer on it, and I think AI could be part of that where you know a hundred
Jeremy Singer: percent,
Sal Khan: you're not just your writing.
You can draw a picture, you can give a speech, you can dance, and it can give you feedback, feedback on your art. I think there's something there. I would love to see a world where the AI tools. That are dominating and are getting traction are ones that actually make us healthier, not less healthy.
Almost like the anti-social media. I've definitely brainstormed with folks. Someone maybe will try to do it, make some tools that can beis on our side, that can monitor us, our students while they're working and say, Hey, that video's not appropriate. Or, Hey, look, we've done this for half an hour. Let's go play outside.
Or instead of chatting with your friend, why don't you go to their house? They're only two miles away. things like that or talking to the parents and saying, Hey, are you okay with Zad doing this for now? Only if Heca does his Khan Academy first, and it's like the AI says, got [00:51:00] it, I'll go do that.
I think that could be a positive future, unfortunately. I think there's gonna be a lot of negatives too, that I'm worried about. Yeah,
Jeremy Singer: me too. But I love the vision. Again, those who have not read brave new words, you get into a lot of this and I found it inspiring and thoughtful. There's a lot for us to do, so I look forward to working together.
But thank you so much for spending this time on the education equation and congrats on all you've done and, look forward to hearing more and working with you in the future. Thank you, Sal.
Sal Khan: Thanks, Jeremy.
Jeremy Singer: Thanks for tuning in today. Join the conversation by following the education equation wherever you listen to podcasts.