Coursera Connect Kazakhstan

Key Takeaways

foree speeech fore spee foree speech fore speeech fore on fundamental soft research I found that uh very useful to have St Lo courses of Cera on this particular issue but was what was very useful uh specialization when you take couple of courses which gives you like comprehensive view of for example in our case of usage of artificial intelligence in educational process so I can see say for example this certificate about uh generative AI leadership and strategy it consists of three Pro courses taught by Wonder University and it gives you comprehensive overview uh about different aspects of under artificial intelligence from uh introduction uh to prompt engineering and all the way to using uh artificial intelligence in different ways in educational and research process so I found it this was very useful and I believe that I can use uh knowledge what I got from corsera courses on on artificial intelligence in my everyday work in my teaching and my researching and in my project management capacity teach English now second language in conversation English foree foree fore spee spee spe speech speech fore foreign k resarch I found that uh very useful to have stud Lo courses of Cera on this particular issue but was what was very useful uh special when you take couple of courses which gives you like comprehensive view or for example in our case of usage of artificial intelligence in educational process so I can see say for example this certificate about uh generative AI leadership and strategy it consists of three Pro courses uh taught by Wonder University and it gives you comprehensive overview uh about different aspects of under artificial intelligence from uh inter ction uh to prompt engineering and all the way to using uh artificial intelligence in different ways in educational and research process so I found it this was very useful and I believe that I can use uh knowledge what I got from corera courses on artificial intelligence in my everyday work in my teaching and my researching and in my project management capacity for andish now second language in conversation English foree foree foree fore speech speech speech foree spee spe on fundamental soft research I found that uh very useful to have St Lo courses of Cera on this particular issue but was what was very useful uh specialization when you take couple of courses which gives you like comprehensive view or for example in our case of usage of artificial intelligence in educational process so I can see say for example this certificate about uh generative AI leadership and strategy it consists of three Pro courses uh taught by Wonder University and it gives you comprehensive overview uh about different aspects of under artificial intelligence from uh introduction uh to prompt engineering and all the way to using uh artificial intelligence in different ways in educational and research process so I found it this was very useful and I believe that I can use uh knowledge what I got from corera courses on artificial intelligence in my everyday work in my teaching and my researching and in my project management capacity for and simple present tense of teach English now second language conversation englishe foreign spe foreign foree speeech for Fore for fundamental of research I found that uh very useful to have stand Lo courses of Cera on this particular issue but was what was very useful uh specialization when you take couple of courses which gives you like comprehensive view or for example in our case of usage of artificial intelligence in educational process so I can see say for example the certificate about uh generative AI leadership and strategy it consists of three courses uh taught by Wonder University and it gives you comprehensive overview uh about different aspects of under artificial intelligence from uh introduction uh to prompt engineering and all the way to using uh artificial intelligence in different ways in educational and research process so I found it this was very useful and I believe that I can use uh knowledge what I got from corer courses on artificial intelligence in my everyday work in my teaching and my researching and in my project management capacity for and simple teach English now second language ination English hello everyone hello and a very warm welcome from Cera I am Vera igna customer success manager working with our wonderful team in Kazakhstan and with the ministry of higher education of Kazakhstan to deliver today's session uh we are broadcasting live from Kazakhstan from aana and together with me I have a great guest a distinguished professor of engineering at Oakland University in Rochester Michigan she probably is well known for quite a few of you from her course learning how to learn and giving you powerful mental tools to learn uh and she's one of our most renowned professors UNC course with more than 5 million Learners worldwide um so we are um thank you very much for everyone for introducing themselves in the chat we see a lot of professors from universities all over the country um and uh today we are focusing uh on our session with Barbara but before we start uh a few housekeeping moments so first of all please continue introducing yourselves in the chat uh we'd like to hear your questions so for this we have a special button with Q&A on the bottom of the zoom screen so please post two questions there and then we'll take them and try to answer as many as we can uh towards the end of the session uh today's session you can listen in Russian or in kazak or in English if you prefer so you've got a globe button again on the bottom of a zoom screen so if you click on it very simply choose the language you want to listen to so without a further Ado I'd like to pass it on to barara to start the session veryone welcome to barar play thank you so much Vera this is uh it's such a pleasure for me to be here in a this is my second time in Kazakhstan and I have to tell you it is even better than the first and the first helped my husband and I realized that this is one of the greatest countryes in the world so to help you to expand and become even greater let's remember the words of the preeminent AI researcher F Fe Lee she said AI won't replace humans but humans who use AI will replace those who don't so it's rather like learning to use any new tool when word processors came out people began to use them and it really helped them and now we have a revolution in what is happening with generative AI but before I dive into that I should back up a little bit and give you a sense of who I am and where I'm coming from I'm from Oakland University and people sometimes know the town of Oakland in California and they think oh you're from warm and sunny California with palm trees and all those kind of things but actually no I am from Michigan and Michigan is a very old State at least in winter so that's me going to work on a typical February day and I teach the course as Vera mentioned learning how to learn and my co-instructor is Terry sinowski the Francis CIT professor at the skulk Institute and this is Terry going to work on a typical February day in his job at at the Suk Institute which is in Southern California and that's one of the great things about online teaching and learning is that not only can students get together from very different places but professors can get together with very different backgrounds as well now to my shock I was invited to speak at Harvard I was a total nervous wreck because you know it's Harvard and I walked into the room and I became even more nervous because it was pack standing room only all around the edges and I thought why are there so many people here turns out our one little course which was made mostly in my basement which you can see a picture of right there for virtually no money at all had on the order of the same number of students as all of Harvard's massive open online courses and online courses Together made for millions of dollars with hundreds of people and what this tells you is that people are starved for that fresh new insight about how we can learn effectively that's based on neuroscientific insight about how the brain works this is very practically useful and this is the kind of thing you can convey broadly through massive open online courses so let us look now at chat gbt what does it do how how is it made often times people know that it works in part by predicting the next word in a sentence although good luck predicting the next word in that sentence it actually works there's there's some deeper aspects of how it works it relates words within a sentence to One Another now if you think about it um what I'm going to do is I'm going to anthropomorphize and that is a very fancy English word for treating in this case computers and generative AI as if it was a human and talking about it that way now we know that generative AI is not a human but if I speak of it as if it were it will make it a lot easier here on me so if you remember about five years ago if you took some English sentences and you tried to use Google Translate to translate it into kazak it would do a terrible job and the reason was that at that time by the time a computer got to the end of a sentence it would forget what was at the beginning of the sentence so it was very hard to do good translations but now of course not only can It remember what's at the beginning of a sentence but it can also relate those sentences to much larger bodies of text and even to much much larger bodies of text on the internet itself so how does this work I'm going to give you like a three three minute tour of Neuroscience the history of both neuroscience and generative Ai and in 2017 this paper was published called attention is all you need well that paper was submitted to the neural information processing uh Society conference which the president of this conference is Terry sinowski my co-instructor and Terry told me when this paper was submitted back in 2017 the reviewers looked at it and they said ah you know nothing special here and so when it came time to put it into the conference they put it off in a corner just a poster session nothing special now you should know that this conference in artificial intelligence is so important that it typically has about 8,000 registration spots within it and when registration opens typically it takes less than 10 minutes for all 8,000 registration spots to be taken so this is a really important conference but this paper was kind of put off in the corner as nothing important but it turns out that what this paper presented for the very first time was the concept of the Transformer and the Transformer as it turns out is so important that now this paper has been cited over 100,000 times what is the Transformer what does it do before we jump into what it does does or how it works let's let's give you a little sample of the kind of creative things that can develop for us so I'll show you uh just a little bit from a website calledo sunno is a music um generative AI app and what it does is you can put in like as you can see here I put in the top tile of that research paper attention is all you need and then I put a little bit of text like technical terms about how it works and then I asked it to generate a song and this is what it came up [Music] with L feedback Hoops decoding life through tangled webs so you can see that generative AI is not just taking you know uh it's it's not just this sort of semi creative wannabe but it actually is truly creative and in many of the same ways that human beings are truly created so do you have a a boyfriend or a girlfriend or a husband or wife or someone you'd like to really impress enter their name into sunno then put in a little description about them devise a song in their honor and give them that as a present and what they'll do is they'll think you are very creative in fact you are very creative you are co-creative with the Great uh insights and creativity that generative AI can give us so how do we know that Transformers are kind of behind all this what's a Transformer what's what is it composed of if we take a look it has two parts it has the encoder where you feed in the information and if you look at the bottom of that encoder it says the it it has a positional encoding that is where the magic occurs where a number is created a vector that helps that encoder and decoder of the Transformer to relate words and sentences to one another so anyway encoding you put your prompt into the generative AI engine then you get something out through that decoder that's the transformer in a very quick nutshell but what we're truly interested in here is that the flow of information through a Transformer is in fact remarkably similar to the flow of information through the human brain and that's no surprise because a artificial intelligence experts and neuroscientists have been working together hand inand through the Summers and the winters of the development of artificial intelligence so has informed the other and also what this means is if we understand a little bit about how the brain learns guess what we can actually use generative AI more effective acely so let's go look into the brain and see how it works at least a little bit it's very complicated but if we just focus focus in on one fundamental building block that is the neuron we can better understand how we learn and how to use gen more effectively so if you look that that is a neuron we have about 86 billion neurons in the brain and it's a little easier to understand these neurons if we use a metaphor for these neurons and the metaphor we're going to use is that of a alien so if you can see here we've got a space alien on the left of the screen and that space alien has three legs that dangle down those three legs are called dendroides real neurons have they can have dozens of dendrites even hundreds and on each leg there are these little spines like Toes that come out and those spines are called dendritic spines and then lastly there's an axon that's like an arm that reaches out from the neuron and what neurons like to do is they like to reach out with that arm and tickle the toe of an adjoining neuron what this really means they're doing is they are reaching out and they're sending signals from that neuron on jumping the Gap called a synapse to the next neuron and what happens when you're learning is you are actually creating by sending these signals when you're learning something you're creating a linked cluster of neurons in longterm memory in the neocortex and so this cluster of connected neurons is our memory of a skill or whatever we are learning whether it's learning how to um do a dance step or how to take a derivative in mathematics or conjugate a verb in a foreign language play a musical instrument whatever we're learning we're simply creating a set of connected links between neurons in long-term memory now unfortunately well-meaning Educators psychologists uh have for a number of decades now been saying don't need to remember things you can just always look it up but would I be able to speak kazak for example if I always just looked it up on Google translate of course not in fact you have to have those sets of connected neural links in long-term memory if you are to become an expert in anything and so in this day and age of generative AI it's all too easy for us to say okay here chat GPT give me the answer to this problem and not learn how to do it ourselves but if we want to think critically about whatever chat GPT is giving us we need to have those links of learning in our long turn memory so what professors and teachers are doing by ensuring that students learn well is all the more important now in this era of generative AI so ultimately how do we strengthen those neural links we practice and as we practice we are able to to um strengthen those those dendritic spine connections between Theon and the the dendritic spine of the the next n and if we don't practice though what can happen let's say we're busy we're we we learned something in class but we're too busy to go over and look at our notes for a couple weeks and then right before the test we try to review the notes and that's when we find out that because we have not been using those neural links they've been swept away and we can't remember what we might have learned before so it's very important to practice with whatever you're learning in order to retain and strengthen those neural links ultimately what you want to do is create a nice strong rich set of links about the various Concepts you're learning so that you more easily can draw them to mind when you need them so this brings me to the importance of metaphor so what why would metaphor be important it turns out that virtually everything we learn is related to material we previously learned and if we use a metaphor to explain something we say this is um like an analy this is similar to this which you already know it can help people learn those ideas much more quickly so let's say that the set of links I'm showing here represents the idea of how water flows current of water you you may think think well how water flows you know that current that's an easy concept to understand but when you were a toddler you played with water you would you would let it run through your fingers you'd stomp your feet and that's how you gradually acquired the idea of how water flows once you have a set of links about how water flows you can use that to more easily understand the concept of how electrical current flows so that metaphor that analogy of water flow is like how electrical current flow can allow students to much more quickly grasp the idea of how electrons and how electrical current works so I want to go a little deeper here when I was working on the manuscript for the book a mind for numbers what I did was I found the names and the emails of thousands of top professors and teachers of various subjects from economics engineering mathematics you name it and I emailed thousands of professors who were known for their high quality teaching with a copy of the manuscript for a mind for numbers and I asked them would you review this uh manuscript and can you give me a sense of you know a suggestions about how to improve it and any other ideas you might like to give me shocking percentages of these professors came back and said sure I'm glad to help you and they gave me great advice in the book but one thing that surprised me was many of these professors would come back and tell me independently you know there's this one thing I do in my teaching and that is the magic that makes me an exceptional teacher and I don't usually tell this magic to anyone else I don't tell the other professors I do this what was that one thing they did it was using metaphor and analogy and why didn't they tell the other professors it wasn't because they were trying to keep it secret it was because the other professors if they heard about this they would say that's why you're such a good teacher you've just make things so easy for your students but isn't that the isn't that what teachers good teachers are supposed to be doing making things easier for their students because learning something especially something difficult is hard and if there's a way to make it a little easier for your students that's a great thing for good teachers to be doing now metaphors always break down so um if you're using water as a metaphor for electrical current flow at a Quantum level that doesn't work but whenever a metaphor breaks down you throw it away and you get a new metaphor and that's how you can help people learn much more effectively help yourself to learn much more effectively and how you can teach more effectively but how do we come up with metaphors simple nowadays you just ask GPT or any of those generative AI engines so for example let's say that you are trying to explain the difference between mainu and scope in Python Programming well if you ask chat GPT it will come up with a very nice image and it will say that in Python main guard is like the gatekeeper of a castle and scope are like the rooms within the castle T and it will give you much more information besides so it's so using metaphors whenever you as a teacher or you as a student are trying to learn something difficult it can be a powerful tool to help you in the old days marketing groups would spend weeks coming up with just the right metaphor to for example explain how a pharmaceutical how a drug worked so that they can give a succinct explanation in 30 seconds in a television ad now we don't need to have teams of people like in the old days brainstorming for weeks to come up with a good metaphor we can just ask generative Ai and don't just ask for one metaphor ask for five metaphors ask for five V very different metaphors or tell it a little bit about you so it can put a metaphor a metaphor forward that that kind of sings to you about what you are interested in and what you know about now along these lines of thinking in terms of metaphors what I would like you to do is I would like you to think about um large language models like chat GPT like Claude Gemini and so forth as being like car engines so that means if you learn how to drive one car like chat gbt you can more easily switch to a car that has a different engine like Claude and still be able to drive it now there are handling characteristics and differences in how those cars how the car engines uh work a little bit and but generally if you just think of them as oh they're just slightly different engines that will help you realize that it's a bit simpler out in that world of large language models than you might think so so if it if you want to know about those handling differences this might give you a little sense of how this works so Greg Brockman um was formerly the president of open Ai and he famously put together he he he wrote in his notebook I'm I'm putting the text beside his notebook uh picture that that tells what he wrote in his notebook it said my joke website really does dumb joke one push to reveal punchline and the same for joke two so so if so Brockman took a picture of his notebook loaded it onto chat gbt lo and behold it created all the code he needed for his joke website so I took a picture of brockman's picture and I loaded it on to chat BT and it created the code that you see here but then just for fun I took that picture and I loaded it onto Claude that's another engine and it's actually one of my favorites and what did Claude do it came back to me and said you want to create a joke website jokes could hurt people I'm not going to do it and I had to argue with Claude to get Claude to acknowledge that no we shouldn't have a completely humorless world and yes it was okay to create this joke website so there are differences in how these large language models are tuned and created in fact um Claude here's a little story about the creation of Claude they were testing Claude uh which is from the company anthropic which is now affiliated with Amazon and they they put in a million garbage words and in those garbage words they seated one sentence and then they asked Claude um what's the answer to this question that could only be answered by that one sentence and Claude came back and said okay here's the answer it answered correctly but then completely unsolicited Claude hether said you loaded a million words of garbage just to see if I could pick out that one sentence amongst all the garbage words that You' uploaded didn't you it was almost humanlike in how it responded in a completely unsolicited way really scared the engineers that are working in it at anthropic so you can see there's very different handling characteristics and actually different insights that the different engines can give us and this is why it's worthwhile to not just default one favorite engine like chat gbt and try out some of the other ones to see how they respond to your problems so just to give you a worldwide perspective on what's going on there are seven major engines worldwide there's four that are coming out of uh Western sources and it can take hundreds of millions of dollars to create these large language models and so it's no surprise that some of the world's largest and most profitable companies are affiliated with some of those engines that we commonly see like chap gbt which is affiliated with Microsoft now and of course Gemini comes from Google Claude from anthropic affiliated with Amazon and meta is uh it produces llama interestingly llama is open that means you know sort of the weights the neural weights that create this we'll talk about neural weights later on but so that's it open they release how they form these large language models but open AI which you would think would be open is actually not open it keeps it closed so they don't tell what those weights are quite interesting when you really look at the the underlying what's going on but more broadly worldwide there are also three large language models that are prominent within China so that uh these are affiliated with companies like Buu Alibaba and tenens Cent so uh I just don't want you to forget that there's this whole other world of larger language models when you look from a worldwide perspective so so in interestingly also though mistol has recently come out from France and it is a very competitive engine um almost competitive with chat GPT you might wonder well how did that come out of France it seems out of the blue well two of two French Engineers were at Google deep mind they left to go form their own company and that is what mistal is so often times it seems that many of these um different engines are arising because Google's Engineers have kind of left and gone off and started and Seed some of these different large language model companies um Google itself interestingly when that paper came out in 2017 Google said looks too complicated for us we're not going to do it and so all of those Engineers eventually ended up leaving Google and seeding some of the different companies that you see uh on the screen so um I also want to make you aware of something called the token limit a token is about three4 of a word in English in Chinese it's roughly equivalent to a word and so you can see that Claude has a million tokens that means you can upload a lot of things onto onto Gemini and Gemini uh with its million tokens might seem to be that's the one you want to use because it has such a h a big um amount of tokens it can take in but you might find that Gemini in some doesn't answer questions as well as some of the others so I in particular like to have uh the essence of research papers I might ask for you know what are the key ideas in this research paper and Claude I find seems to do the best job with that but I I want to give you a sense of how you can think about these kinds of things to do with token limits in other words how much can you upload at one time and the context window context Windows related to the Token limit and it simply says during a long chat how much of that chat can it continue to hold in mind before it forgets you said something sometimes you'll notice if you have a long chat it will forget the earlier things you said in the chat and that's because it it can only hold so much in mind even though it's much more than before it's still got a little limit but I when after I wrote the book with my co-authors Uncommon Sense teaching about what's going on in the brain when you are trying to teach and how to teach and reach your students more effectively um what happened was the publisher of the Japanese translation of the book reached out to me and said uh would you write the forward for the Japanese edition of this book well you know I was taken back because I don't know anything about teaching in Japan but what I did was I at that time Gemini wasn't available yet but Claude with its 200,000 token limit or at least that's what it was at the time in fact those token limits may be larger than what I show here but it's pretty hard to get the current token limits out of each of these engines don't like to reveal it but um my book had about a 100,000 words so that that's about 150,000 tokens if you took the PD of it and so I took the PDF of the book I loaded it onto Claude so Claude had was able to take the entire book 100,000 words and and I asked Claude two questions I said how does my book support current methods being used in Japan to teach and explain why those those techniques work so that was one question and then I asked how does this book explain new and useful approaches to teaching that are not currently being used in Japan I got all this great information back I wrote up my forward I checked it with uh with people from my friends from Japan and then I said sent it off to the publisher of the Japanese edition of the book they got back to me and they said how do you know so much about the Japanese education system and I want to be like you know I'm just good but actually it was that I was good because I was willing to cocreate with generative AI so if you use it in your own work it can be tremendously helpful in giving you a much broader insight into all sorts of different things so let's look now though at what's going on in the human brain when you talk about something like a large language model and actually if we look at the human brain we can understand large language models more effectively because guess what we have a large language model built into our heads and that's why we can speak our native language so here's how that works I had mentioned before that you have these links in long-term memory right and you create them when you learn something but what I didn't mention was that there are two different Pathways that those links can be laid through so first first off if if you look um working memory is in the prefrontal cortex so let's say that you happen to be trying to translate a sentence from one language to another language so even if you don't speak the other language very well you're you're trying to hold something in mind temporarily while you work on it and trans at so this is what working memory does it it holds a thought temporarily in mind while you're working on it and this working memory like right now when you're taking in this information it's going through your working memory and through the hippocampus into long-term memory that's called the declarative pathway but there's another pathway and that is the automatic pathway which goes through the basil ganglia and the thing is you need Links of both types in order to be able to function effectively and to learn effectively so let's look at just a slightly look at this from a different angle if you look there that declarative pathway I'm going going to model that with a blue box and the automatic pathway is with an orange box and so if if you're if you are thinking about let's say a math problem that you want to solve you're conscious of this and your working memory says okay I want you to step through and solve this problem you're conscious of the whole process by large but that other way of learning is very different let's say that you want to learn to um to hit a hockey puck with a stick you know you've got your your your hucky stick you want to learn to hit that HCK in the right direction what you're doing as you're learning often your working memories conscious it says Hey basic glia which you are not conscious of but you say hey basil ganglia I want you to hit this hockey puck and you tell your basic Ginga that it hits the hockey puck and you see whether it went in the right direction but you are not conscious of the process of the learning that is taking place when you are hitting the the hockey puck you hit it you hit it again and again and again and you gradually are training that basil gang layer about how to hit the the hockey puck but you are not conscious of how it's learning it's almost like magic is happening in that basil ganglia what is the magic it is a deep neural network it's not to say that there aren't neural networks elsewhere in the brain but we are really beginning to understand it with that automatic system and if you look each of those circles represents a neuron and the line represents an axon going out hitting the uh dendritic spine going onto the dendrite of the next neural this is called a deep neural net because look there's actually five layers so it's pretty deep the human brain has six layers of neurons and what happens is when you're learning to hit that socer or that that hockey puck your your brain kind of um you you hit it and and magically inside there's a path through that neural net Network and then the value function that comes out Simply means did I hit it in the right direction you hit it and you hit it again and again and you're training your neural network some of those connections are being strengthened others are being weakened as you gradually improve your skill at hitting the hockey puck you're never conscious of how it's learning what it's learning you just tell it and then you see the outcome so this is not only how the brain the human brain learns it's also how large language models learn via an automatic process this large this basil ganga Learning System is where we learn our native langu language it's trained through the basil glia this system is really strong in youngsters as you grow older this basil ganglia system is not as strong but the declarative system gets much stronger that hippocampal system that's why we struggle sometimes to learn a new language and speak it with the automaticity that that automatic basil ganglia system can give us but we can still when we're thinking consciously about it speak in that new language but what happens when uh in humans when that conscious system might get destroyed the hippocampus is destroyed in fact when people have Alzheimer's and what do people with Alzheimer's do they confabulate they hallucinate they make things up and that indeed is what we're seeing in larg language mods they confabulate they hallucinate and it's because they don't have that conscious reasoning uh embedded within them now researchers are trying to emulate that hippocampal conscious system through Chain of Thought reasoning for example but it's still not conscious so this is one of the drawbacks of large language models these engines don't have they they can hallucinate because they don't have like an overview Consciousness that catches things that are incorrect so that's one of the drawbacks of these engines but there's of course more drawbacks one of the drawback back is for us teachers it has pulled the rug out from under us the methods we previously used to force students to learn and indeed learning is really hard and so some students will kind of use whatever tricks they can to avoid that hard thinking that's involved in learning but uh so what chap T and others uh other large language models have done is if if we ask for an essay we can get generative AI to do it if we ask for homework we can often get generative AI to do the homework and so this means that there is an easy way out for students and that means it is all the more important nowadays to have professors and teachers watching over to ensure that students still learn no matter what we are at The Cutting Edge of a revolution it's really uh hard to know how to get students to learn even when they can do things more easily on chat GPT I say the number one thing to do teach students how their brain learns we never teach students it's amazing to me students will have 8 to 18 years of learning education and we never give them a course in how their brain learns so if we can teach students how they learn then that it will make more sense to them when we're asking for them to push through and be able to do things on their own not just asking chat gbt so that's that's an important aspect of um of teaching but there's more than this and that relates to something called the Flynn effect in the Flynn effect um IQ scores were found to rise from the 1930s all the way to the 70s and this has been shown in a number of studies uh and and so researchers of course looked at this and said well what can explain it obviously it's great teaching people were exposed more to good educational opportunities and that helped um students kind of in some sense get smarter but since the 1970s copious research has shown that IQ scores are beginning to go down why is that what happened in the 1970s well you might think about it this way in the 1970s calculators came out and suddenly it wasn't calculators that that might have caused a problem but what probably caused a problem was the tens of thousands of Educators psychologists teachers who all well-meaning said you don't need to remember things you can always just look at up and what happens then you're not using that cognitive skill involved in being able to pull ideas from your own memory and cognitive skills as a consequence May well be the cause of the decline so that means it is all the more important for us as Educators to not repeat what happened in the 1970s and say oh it's perfectly okay um we should be teaching students higher level things that chat GPT and so forth can't do no we need to be sure that we are teaching students to be able to check and verify and do some of those same things that generative AI can do because otherwise they will not be able to think critically at what chat gbt is giving them it can give them all sorts of malarkey they can enter in the wrong number and they'll accept the result because after all it's from generative AI chap GPT so we need to get instill those th that new learning into students if we want them to be able to think critically in this new era I should point out another thing so remember how I said that uh teach teachers and psychologists were often saying oh you don't need to remember things you can always just look it up well of course that was wrong and in fact a lot of educational research literature should be looked at very carefully the reason is that only 0.13% of all education research literature is replicated that means less than 1% of what is published in the research literature is actually checked by anyone else so you can p if anything that you can get published it's unlikely that somebody else is going to check it and that's part of why we get faned and uh and sort of going off onto uh ways of teaching that are not really effective in fact some of the the most traditional ways of teaching are truly effective if you look in my uh School of Engineering for example or in schools of engineering science mathematics across the United States many of the professors are from countries that still respect the importance of rote learning and so rote learning yes remembering things is really important along with using being able to use those ideas creatively so let's return now uh just so I can give you an even better overview of what's going on in the world of the many large language mods I'm going to focus in on the west now because I don't speak Chinese and I'm more familiar with these but just remember there is this other world of engines out there and a lot of these engines have an interface that builds out from one or more of these engines that's been developed many of these and um apps and tools grow from chat GPT because that was first out of the gate so for example there's lots of writing assistants like um grammarly Jenny I kind of like right Sonic a new one is um notebook LM and that will even take a research paper and turn it into a really funny podcast so it's it's pretty amazing but I want to introduce you to the idea of a rapper a rapper is not a singer a rapper is like a candy wrapper that goes around the piece of candy so many of these apps and tools are like wrappers and underneath the wrapper is a large language model an engine and most of the Eng or most of the apps use as their engine pgbt again because it's first out of the gate so um what you find is you can't buy subscriptions to all these different large language models I mean if you keep buying those and you buy some of the tools and apps that are rising from them you'll bankrupt yourself because there's so many so what you want to do is think critically about some of these different apps that are coming out now right onic I find is a kind of a cool website and it will help you um you can formulate a press release turn it into a blog post have it as a Facebook post you know it will help you compose all sorts of things very quickly but Jasper on the other hand is has been in trouble in the Press because people are beginning to discover that it is a writing app it can help you but actually if you go to chat GPT directly you can do many of the same functions right there on chat gbt and you don't need to buy Jasper so you you want to just be aware that um you know some of these apps that are available don't really add that much more than chat GPT or one of the other engines themselves there's a lot of research or apps out there and these are often trained differently than um for example chat GPT is trained pretty much on the web itself everything it could scrape up at the time legally or now it's a little bit questionable sometimes whether it was done legally but it was a lot of stuff and Gemini uh also trained on the internet although it can use YouTube but Gemini famously got in trouble because some of the stuff it scraped up was from humor sites that were sarcastic and so it it would put Gemini would put out advice like this if you want to have a more nutritious meal put rocks on your pizza and it will give you more minerals it will be you know healthier for you I was scraped up from a humor website but it didn't know the difference so anyway these research apps are trained on Research literature oftentimes so semantic scholar is trained on over 200 million research articles I particularly like the website site that sore that particular website you can ask it a research question it will go and answer it to the best of its abilities it will site a bunch of citations but then you can watch it it will go back and check the citations and rewrite the uh you know whatever it is told you so it can be um it's a way of having something that's a little better that doesn't just straight out hallucinate but is checked for hallucinations although of course you should always check it as well there's also live to internet so like co-pilot or PowerPoint makers there's lots of PowerPoint makers personally so far I find that they often just give you a clump of bullet point text and then some generic pictures so I'm I'm not real happy with what a lot of PowerPoint makers can give you um but there it might help you on specific a aspects but be careful that what it can give you can be really boring for people there's also all sort of good apps for teachers and instructors like magic school teach mate AI you can take a picture of your notes upload them and then ask it to generate a lesson plan for example there's good upskilling so like cor corsera has really rich um ways of interacting now with generative AI can help you create University level classrooms even high school level classrooms through something called horse Builder it also can you can interact with it it can quiz You by asking you for example to give an example from your work about uh that that illustrates some idea that you should have been learning in the moo it's pretty a massive open online course it's it's really pretty cool there are other websites that for example GTH the first math website uh mentioned there many high school students I hate to say it are uploading a homework problem and getting a solution this can be bad but it also can be good because you can get you can check and see if you are learning it yourself and just check and see whether you got the right answer there's up there's all sorts of um kind of upskilling platforms there's or or um you know uh there's different coding and so forth that are coming from llama there's imagery like Del U which comes from chat gbt there's also mid Journey Fable diffusion dream studio uh there's also uh um some good video producers and both video and imagery is coming from gemini or from Google as well and I have to so I if this is a place where you want to take a screenshot this is the place to take the screenshot and uh I I would like to say this is inspired by a wonderful book called teaching with AI so I I highly recommend this book and in fact let me give you a few other books I I'll put them all on the screen at once here and so um there's not only teaching with AI but also chat GPT and the future of AI which is by Terry sinowski my co-instructor and uh it's a wonderful book Ethan mck's co-intelligence is an awesome book uh that also gives you a great deal a practical insight about how to use generative Ai and how to think about it and if you want to dig a Little Deeper but still very readable books are two of my favorite books in the world and that's the worlds I see by F Fe Lee and also the alignment Problem by Brian Christian uh the alignment problem you might wonder um you know sounds like maybe a boring title but it explains things like reinforcement learning inverse reinforcement learning how do we model human curiosity It's a Wonderful book as is f Fe Lee's book which is very autobiographical and gives you insight as to why she is now called the Godmother of generative Ai and um gives you a lot of insight into how generative AI was created in the first place and how it works finally AI superpowers gives you a worldwide um sort of context to help you better understand generative AI there's lots of Great Courses on generative AI including on the right two of of my most recent that the top accelerate your learning with chat gbt uh is a um we just produced it I produced it with Jules white who's a wonderful instructor on corsera and and then there's a good um uh course about critical thinking growing from a neural perspective and using Insight from generative AI I do want to to say that large language models can be used to to help us evaluate and kind of hold together um insight about something called collective intelligence which is like all the knowledge of many humans put together so you could for example create a large language model that is trained in different multimodal aspects so you could train it on kazak test texts old ones new ones imagery and color symbolism Melodies you could train it on all of these things and it would be a repository of sort of the kazak way of thinking culture and this can be useful both for kazaks and for people who are looking uh outside from outside H Kazakhstan to better understand what is going on with kak ways of thinking it is easy and important to train your faculty your teachers uh professors at universities and teachers at schools on the Neuroscience of teaching and learning it can be incredibly helpful and uh for example what has been done in a one Hungarian university university of Zed is they've kept the videos in English but they've translated the course beautifully and it is now uh when students apply to the university they get extra points if they have completed the uh learning how to learn in Hungarian course uh there's also possibilities to refilm any of these courses and put them in a truly kazak perspective so you can use kak Stars so for example learning how to learn was refilmed with Spanish stars uh made in a nice uh Spanish from Spain context and it's a beautiful course so learning about teaching there's so much insight for example that we can get from the Neuroscience of movie making how do we get and make engaging materials and if if you might want to have a neuroscience of Education Neuroscience in teaching program you could base a master's or micromasters or even an undergraduate off of some of these Great Courses on corsera because it's pretty hard to find uh a courses that where you've got instructors that are well-versed in Neuroscience in generative Ai and in how to teach effectively but on corera you can find for example this specialization of three courses quite easily I I should note that the Chinese versions of these course courses that were refilled with a confusion con uh context have won best moo Awards in Taiwan so as I'm closing I just want to remind you that of course as you're learning you're creating connections between neurons in long-term memory now we used to think that you were born sort of with all of the neurons you'd ever have but then they would gradually begin to sort of die off and as you got older you learn lose more and more neurons and you get dumber and then you die and it was really depressing but fortunately it was completely wrong now we know that new neurons are being born every day especially in the hippocampus and those new neurons have special qualities number one they help you learn better and number two they help you to feel better and how do those new neurons survive thrive and grow it's like new learning is like a a Travis that helps these new neurons stick in the trellis survive thrive and grow so if you as a student are learning new things you may be grumbling uh about it but it actually helps to boost your mood and if you're a teacher you are doing great things to lift the moods of your students and to help Society in Kazakhstan to be even happier and better so I salute you all for your great work in learning and teaching and I thank you so much for your attention thank you very much Barbara for a very inspirational fantastic presentation um I hope you all have lots of new insights about Jen of AI and