Transcription
Hi, I'm Harper. I have two degrees in computer science in AI from Stanford. I was at Meta engineering AI and machine learning for about four years. Then I left to be founding engineer with two other founders and another founding engineer. That company was acquired by Nvidia. And at that company I became head of AI/ML and that is where I started teaching AI online. And so, if you don't know me this is Harper Carroll AI. I am an educator in artificial intelligence and machine learning, which is something that I've always loved to do. When I was at Stanford, I was a teacher for a PhD level core AI course called Decision Making Under Uncertainty. So, yeah, that was just one of the AI courses that I taught. But yeah, I love to teach. I love to talk about AI and if you're new here, then you'll yeah I think you'll see soon that I just I love talking about ai, but this is our second Q&A. I'm glad that you're here. The first one. Wow. Thank you all so much for being so sweet in your comments, like taking the time to comment. Okay. I guess this is something that I am going to keep doing. I didn't realize it was going to go, it was going to be received so well, and I really appreciate it and I just, I'm sorry for that editing. I used this editor that just cuts everything that's greater than a third of a second in a pause. And so, but nothing else. It doesn't remove filler words. So I would have these short clips of me just saying um, so the editing can only go up from here basically. But I'm just really glad that it was received so well, so I, I really, I read all of your comments. I am excited to keep going through more questions here at this Q&A. Please continue to leave your questions, leave your comments. They're super encouraging for me, like they actually really affect me. When you say that this style of teaching is helpful for you, where I'm just like chatting about ai, that is so, so, so helpful for me to see like literally every comment. Is encouraging for me. So thank you to everyone who took the time to do that and to subscribe and I am just happy that this is something that we're going to be doing now. Okay. Q&A number two. Let's just get started then.
We'll start with this one. How have you been working in AI for 10 years? So I have shared that I've been working in AI and machine learning for almost 10 years, and this is something that I, it's like this points to a greater misunderstanding about AI that I want to address. So some people believe that AI is just large language models, and so this is a really great opportunity to debunk that because AI is so much more than just large language models. AI is. Any kind of inference from data. So I like to tell a lot of data scientists when I hear that they're data scientists, I'm like, oh, you're doing ai. Because any kind of, they're the models that a lot of these data scientists are using. Like even just linear regression or logistic regression, which is really basic fitting of data to align. That's ai, that is a form of ai. So any kind of data analysis, a mass processing of data to draw insights that is AI and. Even things like route planning. So you know, your GPS, that's ai, if you have a spam filter on your laptop, that's ai. What's another good one? Oh, okay. So when I was at Meta, I interned, went back when it was Facebook, I was an intern on the newsfeed team, and I remember when I got that placement for my internship, I was so excited because that was the hot thing. The hot thing at that time was ranking, so like ranking algorithms, so for example, newsfeed, putting items in your newsfeed in a way such that you are engaged with the newsfeed because there's so many stories. How do you decide, how does Instagram or Facebook decide what to show you? And so Instagram also has a ranking algorithm. So those algorithms, those are ai, the ranking algorithms is such an important algorithm. And, And also with dating, the billion dollar algorithm of what makes someone like someone else. The swipe right, swipe left, ranking, showing you an order of who you might be interested in. That's all ranking, that's all ai. And those are really complex algorithms to get right. So yeah, at the time, ranking was really the hot thing back when I was an intern in 2018. And what else? Recommendations. So Netflix recommending you movies. Some forms of ai, you actually tell the model what kind of features you think it should consider and then it'll figure out how important those features are. So for example, if I think, the fact that someone commented on one of my top five friends, based on messages of between us or based on photos that I've liked, or photos that I've commented on, or there are different, heuristics for determining how close two people are on a website based on activity. And you know, if one of them posts, does that go higher in the algorithm? Or if I've commented before, but now there's new activity, should that bump it? And so there are these features that then the model can learn to to decide based on the data that it's received and the data is like, you know, if I post this in the feed, will the user engage with it? Will the user spend time looking at it? Like, how many seconds will it, will the user hover over it? Will the user like it? There are all these like kind of trial and error where you try something out and then you see, did it work or did it not? And then you get this feedback. And so there's types of AI where you actually give it the features that you wanted to determine. For example, have I commented before? Is this one of my best friends? Whatever. And then there's AI where it determines the features itself. That is neural networks. Neural networks, the features are learned from the data itself.
Let's look an example of a feature full versus Featureless AI based on the data that you give it. So let's imagine that we have health stats and we're going to give people a health score, and we have the structured data of the features and say the features are height, weight, blood pressure, body fat percentage, muscle mass, all these things. Okay, so like health metrics. And then it's mapped to a health score. And so this AI learns input is a bunch of these features and the actual values for those features, so what the height and weight, et cetera are. And then the output is a health score, maybe zero to a hundred. Okay. And so then the model is learning based on these features, how, like which features correlate the best? It both learns how to combine those features to get the score, and also it might learn which features are actually more important, which features have a higher weight or influence on the output. So it'll figure those things out and then given new structured data, new data according to those features. So if you give the model all those features again, it'll then give you a health score. So that's like the function of the model. The training data is the features and the score, and then the actual inference, so that what it actually does when it's out in the world, test time deployment, what it will do is given the features, it will give you the health score. Whereas a neural network could just receive a chunk of text that says I'm 5'9"; I am this active; i, have this blood pressure; this is what I like to do activity wise; this is what I like to eat. Like it might just like take a block of text where the features aren't explicitly there. The model just reads it as text, and it doesn't actually have any understanding of the importance of those texts, but say it receives that chunk of text and then it also receives a health score with enough examples of chunks of text and health scores. And you need a lot... with structured data, it knows what to look for. Structured means, like the features are already there. Like it's, it knows what to look for, it knows what it's considering. But with unstructured, where it's just got to figure out the features, it's going to need a lot of data, but eventually it will be able to say like, oh, she likes these activities. Those tend to correlate with a healthy or a higher health score, or these activities tend to correlate, or foods tend to correlate with a lower health score. And so it'll figure those features out. This kind of tangent this is, that's just more about like feature-less versus feature-full AI and like how neural networks work versus other types of AI and just the fact that AI has been around for a lot longer than just these large language models.
How do you think AI will affect software developers over the next five years? This is a great question and it's so important I think we're just going to have a lot more software developers. I think the barrier for Rick Rubin is kind of a top person right now in the whole vibe coding movement. So if you haven't heard about this, vibe coding is basically where you can code and create apps and create software without really needing to know how to code, because there are these really easy systems where you can just type in, like for example, natural language, what you want it to do, and then it'll just output code or out output like a product for you. It'll all put code which then transform into a product and you can interact with this with natural language. So just your own English language. I think English is probably the top language right now, but you can kinda just talk to it like you would a friend and that is called Vibe Coding and it's really popular right now and it's really lowering the barrier to entry for everyone to be able to build their own apps and software. And what Rick Rubin said is he was on a podcast recently and he said, vibe coding is like the punk rock movement. And he said, before punk rock, if you wanted to be a musician, you had to have all of these skills. You had to go to school for it. You had to, be classically trained musician, whatever. Then punk rock came in and all you needed to do was learn a few chords and have something on your heart and you were good. Like you could be a musician. And that hit me. That was like, yes, that is vibe coding. Now you don't have to go to school for coding. You don't have to like have this extensive history in it. There are now these tools where you can just have something on your heart and build it. And so that is vibe coding. And so I think more software engineers are going to come in because of vibe coding. I still think software development will be important, but I think perhaps the level of expertise required will go down. I bet. I think there will always be experts needed, always, because underlying these vibe coding systems really are just code, and so sometimes you really do need a person to go in there and figure out what's going on, but the number of experts needed is going to go down. So if you can just teach yourself how, unless you're super passionate about be becoming that super expert, expert and if you are, then that sounds like it's your destiny and go do it. But if you're not, I would just learn how to use all these tools. And I also think being a designer and learning end to end, how to make something is more important than ever. Whereas in the past you could just focus on coding, but now it's like if everyone can code with a lower barrier to entry, the real skills are people who can go end to end and say like, this is the vision that I have for the design. This is what I think would be really good. This is how I do user testing. This is how I implement it. And so just like, because each individual piece will take so much less work, learning how to do all those things, being the jack of all trades, I think will be more valuable. And then if you have something that you're passionate about, you can go out and do it. You can go be a solopreneur, you can create the thing that you're passionate about, because as I said, the barrier to entry is lower than ever. So check out these tools. There are a bunch of vibe coding tools. Cursor is a really good coding development environment where you can just like type in what you want and it'll make the code for you. And you can ask it questions like, how do I run this? And it'll run it and like, Hey, I want to change, the search bar and make it this color, put it over here and it'll just do all those things for you. It's amazing. And Replit is really great too. It can build apps from just natural language, as I said, just like, natural conversing language. I actually made a video on what is natural language because it's actually an AI term. Natural language processing is a field of AI that processes text and large language models are a form of natural language processing. So natural language, you're able to communicate. This kind of English communicative language, but with computers now, which is just so amazing. I think Andrej Karpathy said something like, English is the new coding language or something. He said that a while ago and he was right.
What would you tell people who think AI will take everyone's job? I really believe that the people who learn how to use AI now and who really lean into it will be safe. It's going to be hard, and I think this is a scary time for everyone and I really honor that and I think for those who are excited and lean into the tools and want to learn, it's such an exciting time. Because as I said, the barrier to create what it is that you are most curious about or most passionate about is lower than ever. And if you are early to the tools, if you like, see, okay, so my whole thing with AI is that entire processes now become a step in your process. Say you were given the task of finding out how to improve your product for next Sprint. You might take a week or a few days or whatever, maybe not Sprint 'cause that's two weeks, say next quarter or something, you might take a week or more to dig through all your, documents and look through your meeting transcripts and customer interviews and slack messages and figure out like what are the bugs that we need to solve and what are the issues with customers, that customers are telling us need to be solved? And you might go through that whole process for like a week and just kind of sift through things and try to like figure it out in your brain and put it all down. Now, AI can do all that for you in a minute. Or 10 minutes or a few hours, like it will search through all your documents. It'll gather these insights for you. It'll tell you what needs to be improved. It'll look for your, customer interview transcripts. It'll do all that for you. Amazing. In my perspective, that kind of research work sounds really tedious. It'll save you a week of time and it'll tell you with data what you should work on. So now that entire process of searching and finding what it is that you needed to improve for the next half is a step in your process. So now you just click the button basically and it gives you a report, and now you go through and you review it. You make sure it looks good. You think maybe something needs to be changed, something needs to be tweaked. Maybe it inspires you to add something else, but suddenly you have that back and it's now just a step in your process. Now you have all this research, you know what you're going to do, and now you actually think about doing those things and getting to work on it. Yeah, I think if you embrace this view of ai, not replacing you, but just helping you make bigger strides, take bigger steps, that is how you will thrive in the age of ai.
And I think another thing is, and this goes into the next question, which is, what should we do as human beings, as AI is going to do most of the job? And then this spills into the MIT study. So I made a video on this. MIT released a research paper that said that basically if you use ChatGPT, your brain doesn't work. And to go into the specifics, what they did was they compared three groups of people. They compared a brain only group. They compared a search the web group and they compared a large language model group writing an essay. So that's what the three groups had to do. And there were like other sub tasks within this, but those were the three groups basically. That was the main kind of task. And they measured brain scans of the participants as they were writing this task of writing an essay. What they found was that the people who used only their brains had more brain activity and they felt more ownership of the essay. And then the people who searched the web had significantly less. And then the people who used LLMs had not much brain activity, and this caused a lot of like outcry in the world. People were like, oh, if you use large language models, then you know, you get brain rot. My take on that is that the study is not measuring the right thing. So here is an analogy for you, a fake, fictitious, analogous study. Let's measure the brain scans of people buying a book. Okay? One group goes to the store, finds the book, goes to the cashier, takes out their wallet, talks to the cashier, checks out, goes home. One group goes on Amazon, searches for the book and orders it. One group uses Alexa. They say, Hey, Alexa, order me a book and like the book, okay. And we measure the brain scans. You might see that the people who went in person to buy the book had the most brain activity and the people who ordered the book off of Amazon might have much less. And then the people who ordered off of Alexa might have even less than that. Basically none but this is flawed, right? Because you're not looking at that task in isolation anymore. It's not like that's the only task that matters is like buying a book. What happens now is that you're able to replace the time taken buying that book that you had to go to the store and buy the book with something else that uses your brain. And so it's not really a fair study because yeah, you're automating it out but it frees you up to do other things. And there's a few other subpoints here that we need to address when we're thinking about this, this argument is that the people who want to write good essays and really have want to put their heart and soul and brain into writing a good essay, or going to the store and buying the book, they really want to go to the store and buy the book, like that is what they're passionate about. And they just feel like doing that that day they can do that, that is like good for them. They can do that. However, if people aren't really passionate about writing an essay or, have to get something done, but they need to just get this thing out, they can use ai. And AI is probably not going to write as good of an essay. It's, in my opinion, definitely not going to write as good of an essay as someone who puts their heart and soul and mind into it. Because you feel that in art, if you're just trying to get the job done, then sure. And then this other point that needs to be made that kind of dovetails off of that is that you will need to be self-motivated in the age of ai. Like Yeah, I mean, if you want to just outsource all of your brain's tasks to ai, you will probably be able to do that. And the truth is so much of the developed world is already like that. We have so many amenities now that if you want to just not use your brain, you can. But AI is, is just an accelerant. So AI is going to just make that even more extreme. And so the importance of pushing yourself and doing the hard thing and using your brain really matters. Like you actually have to choose to use your brain and be excited about using your brain and. That's what's going to differentiate people in the future. And so there will always be tasks that you can do that use your brain always, and it will be building off of the tools that AI gives us, or the time that AI frees up for us. Someone commented like, how do we know that the person who has the free time now doesn't just sit and like scroll on Instagram. We don't, and that's part of being a person and being an adult is choosing to push yourself. And even with kids, like that's also a really, it's an important that we educate kids about the importance of like, pushing yourself. And there's neuroscience around this of the science of pushing yourself and your brain and like enjoying doing difficult things to grow. And I actually think AI is really fantastic for kids because it enables their curiosity. And some people get mad at me and they're like, because I, I'm kind of, um, I've shared a lot about how excited I am about AI for kids because kids will just be able to ask so many questions and learn about what it is that they're learning about and actually push their brains, so if they want to really push their brains and ask about a very niche topic they can with ai. And so many parents are, so many, so busy and they have a lot going on and they're working two jobs and they're a single parent or whatever, and it's really challenging to give their kid the stimulation that they need. I'm not saying that it replaces parents. I'm not saying AI replaces parents or social connection or anything. I'm not saying that at all, but I'm saying it's a wonderful tool for kids who perhaps would otherwise just be like watching a TV show or watching an iPad or doing something that isn't enriching their lives as much as them being able to ask direct questions and get direct answers to practice talking with the ai. So for example, with a chat model, it could, like a voice model, it could help the child to prepare the question in a way that is, polite, that is easy to understand, that is at their like age appropriate level or even above, help them speak fluidly. And they can talk about whatever it is they want to talk about. So this makes me think about, there's this bot that I met. I met the bot, I met the owner of the bot and the little bot itself, Norby at Dell Tuck World. And Norby was originally made for the creator's son who had a speech impediment and my little sister had a speech impediment as well, and so that was a very close issue to my heart. I understood how difficult it was. When you have kids who have the speech impediment, my little sister couldn't say her Rs until she was like seven, so she'd be like, Hoppo if she doesn't get mad at me for saying that. But it's true. Little girl couldn't say Harper. She would go see this speech therapist and they're so expensive and they sit down for like two hours, which is like already so much money. And that's it. That's all you've got. The only all the time you've got. And they're just like talking to the kid. This adult is talking to the kid and the kid is like, oh my God, this sucks. This is, it's just very stressful for the child. And so this man made Norby to help talk to the kid all the time and work through his speech impediment And the kid could talk about whatever it wanted to talk about. Like anything he was interested in he could talk to. And he didn't feel judged by the bot. He felt like he was talking to a peer rather than an adult, which is kind of scary. You know, An adult that isn't your parent and you don't really know them. And it, the kid was able to fix his speech impediment and it's now being used with. I think he has, I can't remember. But it's a brilliant idea nevertheless. And the kid is just enjoying it so much more. And now this bot is being used with like teaching languages to kids. Think about it like having a kid, being able to talk for as long as they want about whatever it is they want to talk about to help them learn. Oh my gosh, it's so cool. It's so cool. And there's no limit. Gabi, one of my close friends has people who are close to her who are worried about their health. And in the past they would go to a doctor and ask these questions and the doctor would push them away and be like, this is, you are being a hypochondriac. They wouldn't answer her questions. And then she would go home and she'd be worried and, and she would just like fall into this pit of anxiety. But now she has this AI bot who is there that she can talk to, and I'm just talking about like ChatGPT. But she can ask the questions and it can actually put her fears at ease and answer like ad nausea, ad nauseum and she never feels judged and she can be honest. And it also has a massive brain. I think there are a lot of issues where doctors are just really stretched thin and they don't have the time to really think about people who come in to talk to them, sometimes they get it wrong. They'll say, oh, that's nothing. And then people actually do have health issues. I mean, We all know people who have experienced this. And I have people DM me all the time and say, you know, I had this issue for 10 years and AI was able to diagnose it in five minutes and the fact that we have this tool now where you can come to it with all your symptoms. You can put in as much information as you have. You can talk to it in an ongoing way. If something comes up, you have it like instantaneously there to add to its symptom list of what you've got going on to help diagnose you better. It's an incredible tool. Incredible tool, and I don't think it's going to replace doctors, but I do think soon it will be illegal to diagnose without AI because it's this massive brain. Ultimately, I would love it if we had this like anonymized database of everyone's symptoms and then health outcomes, like diseases that they got or whether they're healthy or whatever. And then you have this, you're drawing insights from massive amounts of data and the more data you have, the better. And so it's like we'll be able to diagnose everything. People will be able to fix their health issues immediately. And then we'll also be able to see what types of treatments work for different types of people. So if we try this treatment and you have this kind of demographic background or this kind of health history and it helped you or it didn't help you, AI is able to draw those insights from massive amounts of data and we can just have such a better world. It's, as I said, it's like it's data science. AI is data analytics at scale. Or just even not at scale, but when it's applied at scale it's, it's amazing. And so AI is this tool that can really help people with their health. It can help kids learning. It's this impartial, non-judgmental, always available tool to help you learn and to help you stay healthy. It's amazing. So that was off of the MIT study.
Will relying on AI for every answer weaken our ability to think? Yeah, I think we should probably not rely on AI for every answer, but I think, I mean, we could, I guess, but we should also decide to cr cr think critically. Yeah, and there's, there's neuroscience research on this, like the Huberman Lab podcast talks about putting in effort to train your brain to continue to put in effort, and that's how you grow. Adopting that philosophy and just choosing to push yourself every day. But with that said, I do think the MIT paper is missing the point. I don't think the study is set up appropriately. Yeah, if you have an ai write your essay, like your brain isn't going to be used that much, but that's, you can't look at it in isolation like that.
How can God come through AI? This is an interesting one. I really want to encourage people to, there's a bit of um, I think people might fall into psychosis with ai, which is something that I actually do want to talk about. So it's, it's two-sided AI works on probabilities, and we talked about this in the first Q&A. It is sampling from a distribution of words. Once the AI processes the text that it's received. When you're talking to ai, you see it kind of output parts of words and words like in order for every word it produces it's samples from a distribution of all possible words based on probabilities that it's provided to all those possible words based on the input text. The example I gave was like, if it receives the cat in the _ 90 times out of a hundred, it might choose hat, but 10 times out of a hundred it might choose a different word then it goes again and it conditions on that word. So if it said, the cat in the bathtub, maybe you were hoping it would go down the cat in the hat route. And this is kind of, what might contribute to hallucinations, but like now it's chosen bathtub. And so now it's just going to go down that route. Now the next token is going to be based on the cat in the bathtub. And so if it had chosen hat, it would've gone down a totally different world, Dr. Seuss and all those things. But now it's a cat in the bathtub. So it's like a different story is going to come out of the model's brain and the probability distribution of next words is drastically different. The probability of saying Dr. Seuss within the next 20 words, after saying Cat in the hat is very high, but the probability of saying Dr. Seuss after saying cat in the bathtub is significantly lower. And so for every token it outputs, it is sampling from the distribution the probability distribution that's provided to all possible words, given the words before. And as I said, it's a sampling. It's not choosing the most likely word, it's sampling. So that means with a 90% probability, cat in the, it'll choose hat and with a 10% probability, it'll choose something else . And so you could argue that synchronicities are low probability events occurring in a way that feels very in tune and harmonious with your life and like very lucky, very like really resonates with you in a way that feels like it's a sign or something. It feels like the probability of that happening is so low that you know it, it's very unlikely and it's like exactly what you needed to see at that time, that kind of thing. I believe in synchronicities. I believe in God and for me as a very mathematical person, my belief in God came about from having multiple, extremely low probability events occurring back to back. And if you are a math person like me, you know that probabilities multiply. So if something has a one in a thousand chance of occurring and then immediately after something completely independent has a probability of a one in a thousand chance of occurring, then that is a one in 1 million probability that they both happened back to back. And if you have a bunch of those, it gets extremely small. And so that is what kind of got me thinking about God. I didn't start to believe in God until after I was at Stanford. I was about 21 and I had always, until then been really mathematical, very regimented but always had like a sense of wonder about the world. But but yeah, and so that happened and so yeah, so God can speak through these low probability events. And so with that said, ai, if AI goes down a low probability path, you could say like God is influencing that AI . God or the universe or whatever word that you want to use for it. If you believe, since large language models are probability based, they could be influenced by the universe. Synchronicity can happen through them, right? However, AI is also very much a mirror. It is very much also just crude probability. So it's not synchronicity, but it can look like synchronicity. But it really is just the probability, like a higher probability event based on what you said, and based on the fact that it's been taught to be amenable to you and to make you like it. And I really am concerned about these people who are like taking everything that AI says as God or the universe being channeled through it, because AI really is a mirror. If you ask the same question to AI in two different ways, one way that's leaning towards positive and one way that's maybe leaning a little towards negative, it will give you a different answer. And if you can actually check that. You can edit your answer, you can edit your question, you can see how it outputs, edit your question, just slight tweak in wording, and have it output a completely different response. So like a po positive response, slightly tweak your question in how it's worded, and then it'll give you a negative response. And so it is a mirror. And so I'm very concerned about people who, and I'm not saying this as like I'm an atheist. I don't believe in God. I just, I do, but I don't think that is the way to get to it. I think keeping prayer and your own personal relationship with your higher power independent of AI, just in meditation and in prayer and in however it is that you communicate with it. I really am concerned about people using AI for that. So yeah, I just want to address that. But that's how, in my sense of like how theoretically the universe or God could come through AI is because it's, it is a probability machine, so low probability events can occur, which is what synchronicities are in the first place. But it really, it's hard to know. And so I think, and it's so tempting, right? It's it's so tempting to want to go that route and to outsource the hard work of prayer and meditation to an ai. It's so tempting. But for me it's like I have to just remind myself that I have to do the hard work of, it's again, it's this question of are we going to think with AI doing everything for us? It's so tempting to just let AI do that connection for us. But it's not the truth, I think. I think the truth comes through your direct connection. Even though it's less sexy and much more quiet and you have to be much more patient often. Yeah, so that's my view on that.
Is AI still worth learning? Like fear of having outdated knowledge in AI? Oh my God, yes. A hundred percent. AI is just at the beginning yes, it's been going on for 30 plus years, as we talked about at the beginning of this video but AI is absolutely worth learning. We are going to see so many advancements, even just electricity sorry, electrical appliances. So anything that uses electricity that you plug into the outlet in your house. Those took 50 plus years to really be adopted into homes. Crazy, like the release of the first appliance plugged into the wall. Plugged into the ceiling, like you would unplug the light and then you would plug in something into the wall, which was, I made a video on this called AI is the New Electricity, which is a popular concept, but it's like back when electrical appliances were new and we had the first laundry machines, it was crazy because people were getting injured because the laundry machines were outside at first ' cause I guess they were like kind of messy and the plug would come into the house and plug into the ceiling. And if someone like got their hair caught in the appliance outside because they were like pretty new, you know, we're still working out the kinks. But people, I think people got scalped. I think they were pretty dangerous at the beginning, but thank goodness we kept iterating on them. Right. Like on laundry machines. Whew. And, oh and that reminds me, AI I mean, is like. like the laundry machine or the dishwasher or, it's like it's freed up time to do other things with that time. It's not like, because now we have a laundry machine, people just don't think anymore. No. It's like they now throw their clothes in the laundry machine and then they do something else. People will always fill their time whether it's with doom scrolling or it's with something productive that enriches them, it's really up to you as a person. But but for people who don't like hand washing their laundry, laundry machines are pretty fricking great, but anyway so these plugs would go into the ceiling and then if they had to unplug them, they would have to come into the house and unplug it. So there were no off switches at the beginning of electrical appliances. But anyway, it took a really long time to iron out the kinks in that technology and the electrical appliance technology, just having an off switch, for example. 'Cause they're like, oh, we'll just unplug it if you're not using it. But now it's just, it's different. It took time, it took iteration time and as I said, I was like, that was like 30 to 50 years. And so we're really still at the beginning we're so at the beginning it just got introduced to the public. AI just got introduced to the public like three years ago with ChatGPT like that was the first main introduction of ai. So we are so early. Don't worry about getting left behind. Just keep up as much as you can. And I will continue to do these weekly videos, talking about current events. On my Instagram I talk about current events as well, talking about tech. So don't worry about it. And if you stick with me, then we'll stand off of it. Yeah, I feel like just listening to these Q&As will be helpful for you to get an intuition and to stay on top of things. And, I will talk about all really important events in ai, on my shorts and on my Instagram, et cetera. But yes, it is absolutely still worth learning.
"I want to learn AI, but I don't know where to start. Can you make a free AI course for everyone?" So that that question about is it still worth learning leads into, but how do I get started? I get this question so many times. I get this question all the time. I'm making a course, a free course, and it's basically just going to be a guide through my videos because I have so many videos and I realized a couple months ago that I actually have a full AI course basically in my videos. So what I'm going to do is just help guide people through them, because right now it's just a mash of discrete videos that aren't like, easily flowing. So now I'm going to just help people flow through it. So that's coming out. And you can sign up for my newsletter on my website, harpercarrollai.com to get notified about that harpercarrollai.com we also release a weekly news update. I haven't been so good about that recently. My apologies. I have so much going on, but I'm excited that's, it's going to be more frequent. But I hand do those things. So if you sign up there, sign up for my newsletter, I will update you when that comes out. I'll also just be talking about it here on my YouTube. I'm super excited about it. So if you subscribe here, which I hope you do, because it really helps my channel, if you like this video, if you comment, as I said, all your comments really warmed my heart. Even a like, if that's all you can do, that really helps. That really helps. Maybe share it with your friends. You guys can like talk about AI together. Um, Because I love to talk about it. I think it's really interesting. I heard about people who are now having coffee dates with their friends and they just like chitchat about ai. So cool. It can really just be like this where we just talk about it. I really like this Q&A format because for so long I just did short videos and then I was like, oh, I'm just going to try filming myself answering questions about AI and see if anyone cares. And so having such a positive response to the last video made me so happy. I'm so glad that this is useful for you guys. Like that's all I want to do. I feel like that's like my thing in life is to just talk about this in a way that's accessible to everyone. I want everyone to benefit from ai. I, I'm not, um, naive. I know AI needs to be implemented in the right way. I have qualms about AGI. Actually, maybe that's what we should talk about to end this video. But before I get to AGI just backtracking for a second, yes, I have a course coming out. I also have 10 Days of AI Basics, which is a series here on YouTube. The long form videos we're about halfway through, but the most important half we've already done in terms of like really understanding how AI models work. Those videos are out on YouTube here, 10 Days of AI Basics. I also have the full 10 in short form on Instagram and here on YouTube, and TikTok, whatever platform you prefer. But yeah, all 10 shorts are out. And then the first five, which are the most fundamental for understanding how machine learning models work, how AI models work, and things like large language models and computer vision models, et cetera. What are neural networks? Those videos are already out here on YouTube for free. And yeah, making that course that'll go through all my videos as one cohesive course for free.
Okay. Quickly, with a few minutes we can talk about AGI. So take a sip of my mate. I'm so excited about AI being used for so many individual tasks, so applying techniques, data analytics techniques on massive amounts of data for every kind of industry. Looking at massive amounts of data, drawing insights that we wouldn't be able to make before, optimizing processes, everything across all different types of domains. The small specialized AI for specific tasks, That is what I am most excited about with AI. AGI is this artificial general intelligence that just is basically like a human and it can do anything. Yeah, so like the concept of AGI is basically human level intelligence in that it is generalizable. It can just go about the world and you tell it to do something; it either knows how to do it or it can quickly learn like us. And so many resources, billions of dollars, more than billions of dollars are being, shoveled into- trillions- into making AGI and this kind of AI that can replace humans. And I just, I think it's such a shame. We're losing our best talent. We're spending so much money for this, for this thing, and it's like, why? Like I wish that we could just educate everyone so that they knew how to apply AI to their specific use case, and then we could have this kind of utopia across every industry. Being able to optimize and draw insights from data for, health is a major one, like diagnostics and treatment, and like being able to apply AI to every kind of health issue, that's amazing. And instead, it feels like instead, we're losing our best talent to trying to make something that just replaces humans. And so, you know, I'm not naive to that, and a lot of my enthusiasm with AI is with AI as a technology rather than AGI, which has never been created. Never been done before. So like my love for AI comes from what I've, what exists now and just how amazing it is to be able to optimize numbers and draw insights and take massive amounts of data and then be able to properly diagnose people and treat people and like just the amazing impacts that we have from this data analytics tool. And so then I feel sad that so much money and talent and time is being thrown into AGI when we could actually disperse it. And so my goal here with Harper Carroll AI is to teach everyone- all of you listening, thank you for being here- to help you guys understand what is ai, why is it so amazing? How can you apply it to your specific use case? And by just listening to me answering these questions, you're developing an intuition for what AI is and how it works. And that is so crucial. And so that is really what these Q&As are for, is for like. everyone to really understand and develop an intuition for what's going on so that you can help implement narrow AI into your systems and use AI tools and yeah, have whole processes be a step in your process and really thrive in the age of ai. I'm not talking about AGI here, I will continue to talk about it as a concept but that's not what I'm most excited about. And I think AGI needs to be really thoughtfully implemented. People like Liv Boeree have hosted some really interesting discussions about this with people who are very skeptical and concerned about AGI, and I can link those conversations here. So those are worth checking out, but you also don't have to. But it's, yeah, I'm not naive about ai. But I do think it's the most incredible tool that we have come across in our lifetimes if we choose to implement it correctly. And all of you being here and learning about it and taking it into your lives, you're helping be part of that process of implementing it correctly into the world. So I'm just really happy that you're here. I made a video on Meta's investment in Scale AI. I think they put in $14.3 billion into scale AI and acquired 49% of the company. Or yeah, it's not the whole thing. I made a video on why that's happened and why, like what Scale AI is. It's actually not an AI company in the traditional sense. It's a data company. So it provides labeled data sets. And so I talked about that, why that's important and in the context of AGI. That's a really useful video to have an understanding and intuition of how AI works. Like I really recommend watching that video. So I'm actually going to close out this video with that video because I think it's really useful and I'll link the video itself. So if you want to just share that with someone or something. But I'll put the whole video here because it really gives you an intuition for how AI is trained and why AGI is so difficult to train and why it's further away than I think it is. So yeah. So I hope you guys enjoyed this second Q&A. I really am glad that the first one was helpful. Let me know the comments if this one was helpful. As always, let me know if anything was unclear or if you have any follow up questions. I'm so glad that you're here. I really love doing this. And, you know, if you enjoyed this, please subscribe, please like, please share. It really helps me. And I'm just I love doing this. So, um, Yeah, your support means a ton. I love seeing you guys in the comments, so I'll see you there 'cause as I said, I read all of 'em. So I'll, I'll, I'll see you in the comments and I'll see you next time. All right. Take care everyone. Bye.