AI and the Science of Language Learning with Ben Whately of Memrise

Then Ben Whately is the chief strategy officer and co founder of Memrise a psychologist by education with a master's in experimental psychology from Oxford, where he focused on neural networks and computational modeling of human learning, then has spent the last 15 years looking at ways to learn and teach languages. While at Oxford, Ben became fascinated

by how we learn languages. In the past, he had always been able to pass language exams, but would find himself totally unable to actually use the

language abroad. While studying psychology he formed a theory for why this is and memorize has been a journey of bringing that theory into reality, then believes that people fundamentally learn languages to connect with other people, or because they're enamored of the culture of that language, and Memrise strives to embody this experience of connection.

Building on his neuroscience and language acquisition background, then pioneered the use of LLM to create the world's first AI language partner using GPT three technology. Min bot, Ben is also a prolific angel investor investing in and advising more than 50 companies in the UK startup ecosystem with a focus on AI and machine learning and climate tech. Ben, welcome to Tech insiders.

too much money. But I did bang on Oxford, I got very interested in how neural networks learn how the human brain learns, and particularly got fascinated with the way we process produce and learn language. And a lot of what you see in the generative AI models now large language models, and the way that language is kind of inextricably tied up with thought and the way that we express ourselves that that was a lot of what I was into at the time, but in a much simplified and more rudimentary

form. Well, we can do today was imaginable but felt a very long way away. What interested me a lot about language was the way that it was such a fundamental property of our brains to learn language. And yet, my brain, if I'd learned one thing in language lessons over 10 years, it's good it was that my brain was not very good at learning languages. And that was literally my one takeaway. I am

crap at this. And it's quite a disappointing realization to say that we're human brains fundamental property is to be good at languages and yet mine isn't this really helpful. So after I graduated, one of my co founders got deep into the study of memory and memorization and became a grandmaster of memory. Greg, my other co founder, went and did a PhD in Neuroscience at Princeton. And I went away to see if I could learn a language

to a good standard. If I was unencumbered by people trying to teach it to me, I had this theory that actually my brain wasn't bad at learning languages, it was just that it was bad at learning languages the way I've been taught. And the maybe the problem was with the teacher, not with my brain. And specifically, maybe a bit simpler than that, that language acquisition. It turns out, it's about 20%, about learning stuff, and about 80% about effective

practice. And my hypothesis was, I just never got any effective practice. So what if I just go and get effective practice, have a bare minimum of teaching? Well, I ended up there. So I went to live in the north eastern corner of China up on the Siberian border. I wanted to go somewhere where no one would be there. Speak English to me. Turned out being the only native English speaker in a city of 3 million has is a double edged

sword. Yes, there are no other English people to speak to, but anyone in the city who wants to practice their English, you want a guy I was heavily sought after for English conversation. And I found that after about six months, I had learned enough Chinese to be able to then go and set up a business restoring motorcycles, only using Chinese like no one that spoke any

English at all. And that kind of belied both my previous understanding of my ability, but learning languages, and the reputation of Chinese as being a hard language, clarify, I couldn't, at that point read and write very well, I then use a bunch of mnemonics and memory techniques in order to learn to read, which I got pretty good at pretty quickly as well, again, it's that is a real memory task that can be so optimized by the

use of mnemonics. And then also been there for a few years, we then three of us, the three amigos came back together in the UK, to set up memories. And the idea was really founded around language learning. And our first content was around Chinese. And that was, what I've been using, the content I created to teach myself Chinese was then what we used as our first product. That is

personally. But my family spoke many, many languages historically, and I never liked language learning in school, I took years and years of multiple languages, and as you say, walked away with virtually nothing in terms of actually being able to speak it. So I really identify with,

There's a lot of evidence that no one actually came up with this as a clever way to learn languages teaching people to grammar, like there's 70 years of studies showing that ability to do well in a grammar test and ability to speak the language are uncorrelated. You can see it in children. They can't pass a grammar test, they can speak the language, you can see it I had friends at Oxford who studied Latin and Greek, they knew they could pass every grammar test, they couldn't say a single

sentence out loud. From their mind, it's so obvious that these are orthogonal factors, and yet we persist in trying to teach this way. No one ever really thought it was a good idea. We can see how people actually learn languages. There's a theory that the only reason that we teach through grammar is that it comes from teaching Latin that when Latin was a spoken language, everyone learned it as you learn other languages by

talking to people. And then they created Latin grammar books, so that people who were already fluent speakers could become better translators, by getting more nuanced and more precise in the way to interpret things. And then when Latin stopped being a spoken language, all we were left with with these grammar books. So then people were like, Oh, crap, let's try and work out how to do this, instead of doing

Latin like a puzzle. And then they were like, well, we better make some more of these textbooks, learn French grammar books, that's how we do it. And then this whole thing grew up, not because anyone actually

thought it was a good way. And then we're stuck in this kind of self fulfilling prophecy that the people who become language teachers are almost by definition, the people who have succeeded in this, the other small 1% of people who've succeeded in learning through this method, which is not how most people learn languages, you

look globally. Most people that still learn languages by living this someone who speaks a different language and trying to communicate with most people globally learn English by watching TV. The thing that predicts if you live in a non English speaking country, the thing that predicts whether you speak English well or not, is not your education system. It's unrelated. Sadly, for all those hardworking English teachers out there, it's irrelevant. It's whether your TV is dubbed or

subtitled. Like when you watch American TV, is it dubbed or subtitled? If it's subtitled into your language and you listen to the Rachael in France, you learn English, if it's dubbed you don't. Really, which comes back to this point of language learning is 20% about learning 80% about practice. And if you can get that practice by watching TV in the target line gritch by talking to people regularly, the learning bit is like, well, it can be a multiplier can help you move

faster. It's not necessary, and it's certainly not sufficient. Whereas just getting the practice is necessary and sufficient, but maybe not optimal. And the optimal is this kind of 20%. Let's learn some vocab and some phrases because you'll go faster. But don't forget about this other part. And I think that's really what we're trying to do in memorize is giving you that ways to engage in that 80%. And that's really what we're building is that 80% practice.

grandmaster. And I think, you know, the DNA of the company comes from both language learning and hardcore, you know, understanding memory, tell our listeners who have never used memorize what it's actually like to learn on it, and how you get to this sort of 8020 split.

performance and so on. We then open that up to users to be able to create their own list of vocab. And then people started using it to do other things. And we thought, Well, why not try this a bit? Actually, I think that cheese course I may have made that. It was great. It was pretty fun. So we've tried briefly doing that. And then we're like, Well, hold on, this is a distraction, guys. This is not the reason we want to turn

about. And the real issue there is that memorization can be powerful, but it's kind of in a language learning context. Memorization is important as a stepping stone to practice. And it's really useful because pedagogically, the only way you gain an ability in a target language is by processing target language content at just above

your current level. So in that experience of listening to language, trying to work it out, taking a guess having it confirmed, that is your brain actually learning so that next time it hears that meaning just appears without you having to translate it all the time, when you're still having to translate it, you're not doing the language. And what memorization can do is mean that there is more target language content that is just above your current

level. Because if I just go now and try and understand some Turkish, I didn't know any Turkish. So everything is basically way above my current level. So if I go and learn 100 words and phrases, then there's a little area of content that is just above my current level, and I learned more than, then there's more, I still have to do that practice. And that practice is still 80% of my time. But I can get more content that is more interesting can come into my sphere of practice, by

learning quickly. But what we found was that these memorization tools are super effective. And you can use kind of rich encoding and mnemonics to learn words really quickly. But in a sense, it becomes a bit of a just a trick and just to show pony, because it's not actually what you need to do. It's sort of false progress, you get to a point where you're like, I know 2000 words and phrases in this language like, Okay, can you have a conversation? No, you can't. That's not a useful place to be.

So then we, we shifted the emphasis over to how do we give you more of those immersive experiences. So the first thing that you'll see in memorize, and by the way, we've left all those cheese courses and whatever, they are still there on the website, you can't get them on the mobile apps. They are still there, but they are relatively hidden. And they're a very small part of the focus on the

business. The main thing that we then shifted to was creating our own courses so that we can control the content and make sure that that content is really high quality. And then that we can make richer and richer experiences around it. So the first thing we did was add in videos, we we bought a double decker bus, filled it with

videographers. We got a Hollywood director called Doug Liman to come and train me like The Bourne Identity and things and he came and trained our videographers, and advise them on how to how to set things up as well as possible, had a good time with him, sent the bus off around Europe for six, seven months, with his team of videographers just collecting 10s and hundreds of 1000s of videos of native speakers from

all over Europe. And the idea was, let's bring the immersion to you so that you've got it on your phone and you've got all of these videos, every word every phrase, used in context by a real person actually using it in their normal life. That's the first thing giving you, which is why we then removed actually mnemonics the memes from the app, because they're kind of a bit too much of a distraction, what you want is to get as fast as possible to, I hear the words, and I understand them.

And if we're giving you a mnemonic about what it sounds like, it's kind of a bit of a confusion, it actually becomes extra baggage. Mnemonics, or memes, by the way, are super important for learning to read Kanji, Chinese characters, reading other alphabets that can be super powerful that and we will bring them back in for those use cases. But that's not the central, the 99% use case on memorize these days. So first thing was we put those videos in to that kind of learn

experience. But that was still that was a slightly more immersive learning experience. But still, fundamentally, that was about learning words and phrases. Then we build on the two, we call this our pedagogical stool. It's a three legged stool where we have, learn, immerse and communicate. And these are the three legs of our stool said learn leg is where you memorize words and phrases, you've got a course you learn sets for things, you pick your topics that you want to

learn about. And then you get the words of phrases related to those and you know, the grammar is explained and so on. Kind of standard language course. Then we've got a maths where we speak to this need, that a language learner has to just process a huge amount of content at just

above their current level. And the way we do that is that we look all across the internet, YouTube, Instagram, Tik Tok, we curate 10s of 1000s of videos, in every language, we then generate transcripts of those videos, and we use that you do that using Whisper AI, this has only recently become good enough to the AI has only recently become good enough to be able to

do this. So then we get a very high quality transcript, then we pause that transcript pulling out which of the chunks of language that are most important to learn. So if in the video I said, By the way, you wouldn't want to learn the words, by the way, because I wouldn't get you anywhere near. So we need to chunk it up into the right bits of language. So you need to learn and then put those into a word list. Then once we've done that, it gives you two options.

Either you can go to the immerse tab, and then you're going to watch a video let's find a video. I like bad bunny, let's watch his music video, and I'll learn some Spanish. And we can then tell you, Okay, go and learn these words. Once you've learned these words and phrases, you'll understand that video. And so you can either do it that way around, like led by your interest or like, watch a

football game. And then we can say these are the words you need to learn to understand the commentary, and you can then learn the words and then go and watch a football game and you're like, Okay, I can now do that. But also the other way around.

It means if you've been doing the course, we can at any point you've learned 50 words, we can say okay, here's some videos we found that you will just about understand, because even when you've learned 50 words, we can go and look on Tik Tok, and find some videos that are only 15 seconds long, but they are the videos that 1000s of people have liked. They're entertaining, they're funny, and they're in

the target language. So rather than being constrained, to content that was created for language teaching, which is as a general and violated rule, which is unbearably boring. Language learning content is just eyewatering the doll. I remember once you're trying to hire a videographer as you to go on to go on the bus around Europe, who had been making educational and Language Learning Videos content for Pearson for years. And I said to her like, cool. So the videos you make Pearson would

you? Would you watch them yourselves yourself? Would you enjoy watching them? And she just laughed. It was like, no, no, no. There is no videos to be just absurd. It was like you seriously, you think all of these people who will try to learn English can fight themselves to be interested in the video that even you it's maker don't think it's interesting. How have the temerity to think that someone could do that. Anyway, I

digress. So we've got the immerse leg where we go out onto the internet, get videos on any topic they are interested in. So you can also say like, I'm interested in knitting, well, we got videos for you. You can use that as your miliar of language learning because maybe I'll come back to the point about why I think syllabuses and curriculums are wrong. I'll put a pin in that and come back in a moment. So that's the first bit and then the communicate bit is where we created this language partner

AI. We built it on GPT. Three, a year ago now. So we had this out. You can talk to it, it talks back. We set up these situations where you have to achieve something so you like you're in a garage, you've got to try and persuade the car salesman to sell you this Audi for $500. And there has to be a bit of tension got a bit of fun in it because a bit of a challenge. Because if as some of our esteemed competitors do, they've built language bots, on GPT, that are much more

functional. It's like order some coffees, and you order some coffees and it gives you some coffee. I was fun. It's just, it's not very exciting. And one of the problems we have with practicing online is that when you practice using language in real life, if you order a coffee, the reward you get is you get a coffee, and you wanted a coffee, that's why you asked for it. So you're rewarded, I get a coffee, that was good.

Whereas when we asked you to do it online, you don't get a coffee, you just get a green tick. Well, that was rewarding, except it was I wanted a coffee. And so instead, what we do with the member, which is what we call a language partner AI is to create these challenges where it's a bit difficult, it doesn't just give it it won't let you have the car for 500 bucks straight off, you've got to convince it, you've got to give it some good

reasons. That makes it a kind of a fun challenge that you can kind of get involved with. And so those two bits, the immersing communicate, and of course, the words that you've come across in those conversations, you can put back into your learn stack, so that then you learn those words. So these three legs of the stool

are all really integrated. And they all feed off each other, you watch some more videos, you've created a bigger stack of words, to learn, it learns, you go back to learn, you learn all those words, then you come back, and you can then have a richer conversation and communicate. So they're all bouncing off each other all the time. And it's this that is it means that the learning experience gets tuned to you and whatever capricious whim drove you drove your interests like what do you want

to do next? Okay, well, we can line up the language learning so that that works for you, which is in stark contrast to syllabuses where someone has tried to work out the right order for you to learn stuff. And it gets lumped together. So it's like you learn all of the future tense at once. It's like you don't need to, you're learning way too much. Because you probably don't need to say

the third person plural. For the future tense straight off, you're not going to come across it immediately, you need to work out the things that you need to learn to do the things you need to do. Because language learning is a slow process. Anyway, every single thing that you learn, needs to take you on your journey to that first successful conversation. Once you've had that first successful conversation like well, I believe I can do this, then you need to get the next successful

conversation. And anything that you learned that is not in that conversation was a waste of your effort. And it made it less likely, it's not just a waste of effort, it actually made it less likely that you would have success, because some of your finite score of time and energy just went into doing something that you didn't need to do. And instead, what we need to do is focus you so that everything you do gets you closer to your aim, which is having a conversation.

And our belief is that the best way to do that is to work out what it is you want to be able to do. And actually the member situations are really good for doing this, like getting a haircut. When I looked when I was in China. My Chinese got good. But it got good mainly and motorcycle factories. I was gonna ask if I could, I could talk motorcycles. I didn't know anything about motorcycles. When I went out there. I knew more about motorcycles in Chinese

than I did in English. So actually, when I came back to England, I used to have to point at parts and I'd be like, and then have to word for word translator. And what I remember particularly is the font the NT is working divide electricity machine and the mechanical distributor cap. I've heard of that. It was but it anyway, I totally lost my train of thought in that random walk into Chinese motorcycles. You were saying

haircuts? Yes. So I still couldn't get a haircut I was kind of terrified of going in because the most of the haircuts that I saw coming out of the Chinese barbers were not the kind of haircut that I wanted, or that I thought I could carry off. So I was always terrified of those moments, and never quite got the right vocab. And it's quite hard vocab to look up of like, how do you say a short back and sides in Chinese like what? Because it isn't just a

direct translation of that. And so what we do with the member set up the barber shop, where you have to try and get certain kinds of haircut and you can go in you have the conversation, the hairdresser keeps asking you questions about your holiday, did you have to try and fit those off and get him to actually give you the haircut.

Just have those kinds of experiences which are what happens because when I sit in the barber's chair, I didn't know if he was asking me something about my haircut that if I got this wrong, it was kind of cut my head off or give me a buzz cut. Or if he was asking to shave my beard or if he was just asking me where I was going on holiday. And being able to understand those things, it comes from a very subject

specific knowledge. So arranging instead of defining a rigid syllabus, which by the way, also proven to be the wrong order to learn things, but rather than having a rigid syllabus, we arranged around what you want to do, and then just in time deliver you all of the information that you have to have in order to do that thing. The image I have of this is that if we spoke about cooking, like we speak about language learning, then we would say things like, Hey, have you

learned to cook eggs? And you'd say, oh, no, I haven't learned to cook, I keep meaning to. And I start learning about oven structure, and about the theory of knife sharpening. But you know, I never, I never actually get beyond that, which, of course, what you do is like, I'm not a great cook, but I learned scrambled eggs. And then I learned fried eggs. And then I learned to make pasta. And, you know, I dish by dish I got better. And that's actually how

you learn languages. You know, you learn to get a haircut, you learn to buy a bus ticket, you learn to order a coffee, you learn to do the things that you need to do to have your life. And then gradually, you generalize between them and you start spotting the patterns. And memorize is designed to be a tool that enables that journey. Rather than being a tool to force feed you a syllabus, that is proven to be a poor way to acquire a language.

syllabi, forget grammar. This is really not what this is about. It's about practical learning. And I would imagine, I know that I hear this as a failed language learner. You know, and I'm sure many of our listeners hear this and say, Yes, like, I've been waiting for somebody to say that, because I've been trying the babbles of the world, I've been trying to Duolingo for the world, and they just don't do it for me, I can never have a conversation, I can never get to do the simple things I want to

do in that language. So it's really refreshing. Yeah,

So it's not enough on its own, but it can be really engaging as well. So I don't I don't want to get into the kind of trash talking competitors. There no, I know you didn't. But I just wanted to make really clear that that's not my position in

mentioned whisper AI. And you've talked a little bit about Memebox, which is your, you know, DBT three based conversation tool, which can do does both, you know, text and speech and everything's instantly translated, you have a background in neural networks, you know, you've obviously been thinking about this since way before it was popular. We're at this crazy moment, of course, where suddenly the tools are getting better so quickly, especially around translation. Where do you see language

learning going? In this moment, where suddenly, you know, they're just incredible sets of tools we've never had before?

interesting question. My understanding of this is when we talk to people about why they're learning languages. They're not learning languages to order coffee. They're not learning languages to buy a bus ticket. They're learning languages to connect with other people. Be it a partner who speaks a different language upon his family who speaks a different language in place. They go on holiday and

they want to make friends. It's human connection and and there will always be a weakness to human connection when you're talking to someone through a machine. That's one thing. Second thing, most of the situations where you're having human connection is not actually just one on one, it is multiple people involved. And it's a significantly harder technical problem to overcome to direct

attention. And so just from a brain standpoint, you've got the way that the neural network of the brain works and processes information. And then you've got how attention is directed through the brain, and the different sort of feedback mechanisms on that. And so what LLM some generative AI is very good at is producing language and behavior, mimicking a lot of those thought processes, but trying to sit at the dinner table with 10 people speaking in Spanish, and translate all of

them at once. How do you direct it, how to choose which one separating out one person's voice from a stream of audio? Very difficult task, and not as the, as far as I'm aware, we're very close to being able to do at the moment. So there are a bunch of things there that are still very difficult technical problems to overcome. But I think if you wanted to do it, just through a device doing the translation, but I think the overwhelming thing is that people are learning languages

for human connection. And actually, I think that what I observe in myself, is that I think it could well go in the opposite direction, that I find when I visit countries that I don't speak the language at all, like when I was in Japan, don't speak any Japanese, I would start using Google Translate to talk to people. By doing that, I would start paying attention more to what I was saying what they were saying, the bits that sounded that I started to learn

some basic phrases. And that stoked my desire to actually learn the language. Rather than just being like an sp Japanese, I'm gonna have two weeks on it, I'm not gonna speak. It actually got me engaged in doing it. And I think that we're human connection is the thing that you're looking for, which is what language learners are looking for. I actually think that these tools are an aid to that rather than a hindrance. That's the sort of main answer

to your question. I also think that there is how we apply generative AI to help you in the learning process. So there's a broad edtech level, we've got the applications of AI, where AI can do stuff, so you don't have to learn how to do it. And we've got the applications of AI that helps you to learn faster, and to learn more. And in the specific context of language learning, I think there are some things functional language, you don't need to learn that

anymore. But honestly, no one was learning languages to do that, anyway. No one was really learning, partly because courses weren't structured that way, if you wanted to learn to order a bus ticket, you had to learn so much around the side. That worked as obviously it wasn't

worth the payoff. So for language learning, I don't think there'll be a huge replacer except for sort of short trips, and very functional language, it will be useful, and it's good for, but then the really exciting thing is, how does AI help us learn faster, and that's really what we've been doing with the member and with the immerse content,

learning is about. I'm actually curious from the learning side from the you know, if somebody imagined, you know, we're in 2040, our, we can pick different years here, but let's say 2040, and you have a, you know, 12 year old who says, oh, you know, I'm really getting into manga, and I'd really love to learn Japanese and travelled to Japan and meet people and really get deep into it. What will that learning experience look like?

In a world where there's so much amazing tech, like you've been obviously, innovating within memorize in all sorts of interesting ways, being able to do speech recognition and being able to do instant translation between languages. But where is this all going in terms of the people who do want to learn language?

phrases. So your first experience and this is pretty much there at the moment is that you would watch a few videos with the subtitles in English. So you get into watching the videos and start like building that appreciation of back engage the storylines, the narrative. Then once you've done a bit of that, just watching and the sort of background, cultural osmosis, then start learning some of the words and phrases, and then go

back, watch the manga again. And then you're seeing, okay, you're now actually understanding some of what's going on. And then see I'm doing that bounce back and forth, and then drop into the communicate, stop role playing some of those situations, we'd have conversations where you had to pretend you were in a particular situation in a mega, and how do you roleplay out that situation, start having that conversation is just bouncing back and forth between those

experiences. And then, crucially, at the point where you personally feel comfortable to do it, connect you up in a situation where you can actually use it, use the language live with other people that you need to take that final step, different people, it's very personality, lead decision of when you're comfortable doing that. Some people are happy talking from day one, those people are the people who today end up as polyglots generally.

But people like me who are much more reticent until you're more confident there are people will spend longer speaking to the member. But actually, once you speak to the member for a bit, and you get used to just talking, you don't just like I got this, it's no longer a big stress. And actually, stress is such an important part here, you asked earlier about memory techniques. I mean, what really simple thing about memory is the stress is good for memory formation, and terrible for

memory recall. And the problem that we have when trying to speak a language for the first time is that it's really stressful, you're in front of someone, you're looking like an idiot. And it sounds like a word. Or the more stress you get, the less you can remember anything is disaster. It's an absolutely terrible situation.

And so the stress of that situation means that your chances of remembering it just like fall off a cliff, which is why people go and have a couple of drinks, and then find they're no longer stressed by it. They no longer worry that anyone thinks. And then the language flows. And what we're trying to do well, actually, we also did a study on the member looking at stress levels, while people were talking to the member versus that language teacher. And it was like 40% reduction in

stress. And I can imagine when you're talking to the bot, which means you can successfully recall much more effectively, when you're talking to the bots, it's like training wheels that let you actually get good at doing it. So that then when you go to talk to a person, not only are you not so stressed, because you're confident in your ability, but you're able to do better, because you're not so stressed that so it's kind of you're more likely to do it and you're like, wow,

this vision. You know, people listeners may be thinking of the Babel fish, right that when you mentioned the the instant translation, that's the classic Hitchhiker's Guide, you know, model of the future of language, or Star Trek has this as well.

But I think this vision you're painting of having a, you know, friendly, low stakes, comfortable AI bot, that you can converse with that knows your interests that knows, sure, of course, knows the language knows any language, that can get you comfortable enough to jump in and have conversations with real people. That's a really cool and very natural when you hear it vision of how AI can write. And that's just based around.

But the core game thing is like built on what interests you not on doing a boring activity. Curiously, though, one of the biggest challenges that we have is people's expectations. So people kind of have such a strong expectation that language learning is going to be boring and painful, that if it isn't there, what's going on here? And people are so used to learning points. As grammatical structures, like the crocodile is wearing a yellow hat, it's totally worthless thing to

learn. And yet, that is what people expect. And so we need to get very good at explaining to people why this is actually the natural way to learn. And I think I am hearing that people are getting it more and more intuitively that it is like, people know, when you actually ask them about it, people know the only way, they really believe that they'll end up learning a language is to go and live in the country. That's why I want to move to China is the only way I thought I'd learn the

language. But then why is that? Well, that's because you get a lot of really targeted practice, practice, at the language you need, in order to do the things in your life. And kind of life creates its own spaced repetition. Because the things that are most important to you come across every day, you say hello to people multiple times a day. So you get good at that first. And you don't need to think about a syllabus, and what should I be learning, because you should be learning what you

need to do. You need to learn in order to go about your day life. And so we're just recreating that experience. But without you having to leave home.

naturally. And it is in a natural order, according to what is useful to you. Well, how do we support you in that, given that you are at home, in a country that doesn't speak German? How can we create that situation for you, so that you can have those benefits from

where you are. And that's actually turned out to be much more important faster than the memory techniques, which at first blush, these memory takes these and they are really powerful, it's just they're not, they don't get you all the way to the end, it's kind of like any of the things that are happening in the learning space. You know, like Duolingo is good at the part that it does. But it only does a part, right? It doesn't complete that full cycle of practice.

And you've invested in over 50 companies, including certainly AI and machine learning companies, because you've been in that space for a while, just I'd love to hear a little bit about how you think about angel investing and how your experience building memorize for over a decade relates to your investing thesis.

is to make number go up. But it's also the fact I'm not just betting on a horse, I'm owning a part of that company, that company is able to do what it's doing, because I've given them my money. And that leads to a kind of uncomfortable place where I had 30 companies in there, and I didn't know everything that all of them were doing. And yet they were doing it with my money. And I was therefore responsible for some of that. So that stuck with me

for a long time. And then, through the process of memorize I discovered this kind of excitement I have for values before memorize as well starting early stage companies and it's that kind of going from infinite possibilities, but no clear idea of the direction and starting to pull together. What is the thread that Most important, what's the really powerful thought and insight that we have here that we want to test? How can we test that? Can we build

something around it? How can we build that momentum, I just find that process really exciting. And then it struck me that this is a way that you can have such a huge impact on whether a company exists or not, like partly being the builder, but I can't build 50 startups. But you can have a bit of an impact of do a bit of an investment, and help them along, help be part of that kind of forming journey. And it's just so incredible seeing them grow from idea to

something. So I've been doing that a while starting, perhaps unsurprisingly, in ad tech. But then moving. Over the last few years, I've really shifted towards climate tech entrepreneur in residence at a program called carbon 13. That is a venture builder for the

climate emergency. So it takes on people who have different skills or their technical scientific business, commercial, and then they form into teams, and then dive into problems, try and solve them try and create solutions that are businesses out of them. And that just watching people go from no idea, forming ideas, businesses raising money, it's a really exciting process.

see your square on the map. And then you earn carbon credits as the trees grow. So you can offset against your flights or whatever. But also so that you could if you want to reset if there's a sort of interesting play there that and this is a very experimental one, this is playing with concepts. But if the price of carbon credits goes up in future, which it may well do, then the value of your token may go up, because it's going to keep emitting carbon credits for

40 years. So there could be a speculative element where it's a token that actually is underlied, by something non blockchain. And we're kind of playing with this idea of how does the provably unique facet of a blockchain token? How can we tie that to a particular site in the real world to create a forest? And how do those things tie together and it's that technology is very, very early

on. But I think the opportunity there is to create something where normal people like me who want to do something about climate change, at the moment, like I don't, I'm not going to buy carbon credits myself that I just don't know where they came from it. It seems like a bit of a waste. But I might buy a tokamak and be like, That's my square of forest. I can look it on Google Earth, I can see where it is. And I cover credits from it. And I can see all the legal documentation for why that is

mine. And that's pretty cool. I think it really

That enables your learning journey and I think we will see this huge shift away from syllabuses and away from exams, which examine performance on their syllabus, it's towards just in time skill delivery, for what you need to be able to do. People being able to see where they want to go, I think, got the skills to go there. It's a bit of a cheat in that that's definitely AI empowered, but surely memorize we were doing that before GPT three, we were doing Yeah. GPT three medigo much faster than we

to do it. And of course, I can support that like crazy. And finally, what is a resource that you would recommend for someone who wants to dive deeper into any of the many topics we discussed today?

language lessons there. And I think that's something that I feel like we've bought to the world that is extremely valuable.