This video will change your mind about the AI hype

0:00 it seems like it's just a 1% Improvement this is actually a 10x Improvement to go there's a new virus running around have you heard about it it's called hype it's as old as human history and some of the symptoms include rabidly mentioning the term ai ai first generative AI third aiying AI finally rigorously tested AI with AI and making verifiably false claims Google's best Gemini demo was faked that's their headline I want to try to address some of your anxieties but at the same time I'm not an expert I'm not following every Cutting Edge research paper but I would argue two things sometimes when people spend all their time focusing on every tiny little detail it prevents them from zooming out

0:46 and looking at the bigger picture and also most of the conclusions I'm going to make in this video are actually in line with the leading AI researcher at meta and many other intelligent folks who do happen to be experts so perhaps they is some truth to what I'm about to say now I can't predict the future I can't tell you exactly where the tech job market is going but I'll try to provide some historical context give a lot of facts and to reach my conclusions I'll make a series of thought experiments some of the things on my list to address are fake demos why is every big company investing in AI it must means something right a lot about human behavior actually and also with so many unknowns in a changing world how do

1:32 us humans make big life decisions like which college major to pursue or which career path to transition into pretty much everything I'll talk about today centers around one thing technology changes a lot but there's one thing throughout all of human history that doesn't change and that's human nature obviously all of this started in 2022 when Chach PT was released at that point I think two things were immediately clear to anyone who is paying attention one this is a revolutionary technology completely unlike the web 3 hype two this technology was going to be slow moving it wasn't going to change the world overnight and I publicly made these claims the day after GPT 4 was

2:18 released make no mistake this technology will revolutionize the world it's already doing that but I don't think it's quite as powerful as the hype is making it out to be I even question questioned Sam Alman and open ai's ethics and do you know what people said to me they said I was coping they even said I was a Google fan boy I don't know where they got that idea but more recently people are starting to agree with me it's been revealed that open AI has made employees sign illegal non-disclosure agreements they're promoting a culture of lying what data was used to train Sora we used publicly available data and licensed data so videos on YouTube I'm actually not sure about that

3:05 and this weird thing where Sam Alman is scanning people's eyes in exchange for some useless cryptocurrency For Better or For Worse I've been pretty good at avoiding this sort of herd mentality so honestly it didn't really bother me when people said I was wrong now who knows all of my predictions could turn out to be wrong everything could change overnight but I'm going to explain why that's unlikely these big tech companies can't all be stupid why would they be pouring billions of dollars into Ai and chips firstly I agree they're not stupid but that definitely doesn't mean they haven't been wrong before and second the question of why is kind of hard to answer so let's do a thought experiment to answer the question let me ask you a question if you were Google what would

3:51 you do you have the choice of invest in AI or skip it in the case of investing it's possible that one it pays off AI hype was actually well worth it the second case is that Google loses money they lose their investment it wasn't worthwhile believe it or not from Google's perspective either of these two outcomes doesn't matter because if the investment pays off Google will make a lot of money if it doesn't pay off Google will make a lot of money the last thing Google would want is to skip the investment and then somebody else entirely disrupts Google's business Google is a monopoly they have so much money they don't know

4:38 what to do with it and they sure as hell aren't going to give it to their employees so the case that every company is trying to avoid right now is the case of Microsoft in the late 2000s Steve let me ask you about uh the iPhone and the zoom if if I may $500 fully subsidized with a plan I said that is the most expensive phone in the world and it doesn't appeal to business customers because it doesn't have a keyboard which makes it not a very good email machine now with hindsight are there things I'd do differently of course like what come on charie I probably would have started as doing Hardware earlier so that we could have been more effective in the phone business I mean this is kind of like an IQ test Microsoft missed out on smartphones and missed out on a trillion

5:24 dollar industry that is the case that Google and every big tech company is trying to avoid Maybe these investments will pay off eventually but right now the only company that's actually making money from AI is NVIDIA they're the ones making the profit it's not surprising because they're the ones selling the shovels it's not proven yet that those shovels can actually find gold and it might never be proven history repeats itself this has all happened before in different ways please tell me you don't have amnesia please tell me you didn't forget the Euphoria of 2021 when every company was just hiring as much as they possibly could why do you think they did it because we're in a new world we will need as many Engineers technology is

6:11 everything now and if we don't hire a bunch of people our competitors will and then we'll fall behind and what happened after that not just Google not just Amazon not just meta not just Microsoft all these big tech companies were dead wrong and they laid off thousands of people you have to understand that hype is a force of nature it's the most powerful marketing tool that's ever existed let's talk Devon AI for a second I made a video talking about it right after the announcement I thought the founders were super smart but also that I'm not worried about Devon because these people are clearly fighting an uphill battle I don't know a lot about it the benchmarks say that this is more

6:57 effective at software engineering tasks than the other llms I wonder if that's because it just puts a few things together like it has its own capabilities to like do research like go on Google browse stack Overflow and run code and execute code and they just put those pieces together more more cohesively than GPT obviously in a short amount of time they didn't create their own complex llm I'm pretty sure they're using one of the existing llms I didn't closely examine the evidence or anything like that it seemed kind of obvious to me a few months later people people are realizing they were fooled by the hype Devon is extremely overrated at least for now if you don't believe me maybe you'll believe this guy he basically proved that at this point Devon AI is more useless than a freshman CS student

7:44 who's one week into their first programming course more recently they've kind of pulled back their claims okay for now we're just going to start with some code migrations here and maybe some code refactoring over here now can you think of another company who heavily exaggerated their claims and still turned out to be pretty successful well I can think of a few dozen but the most notable example would probably be Tesla I've patiently been waiting for the day that I can just sit in a car and have it drive me somewhere Tesla has been selling full self-driving software for a long time and I'll say this gently but I think I have very different definitions of full self-driving but it doesn't matter does it people buy the hype it's not going to be long long before Tesla

8:30 cars actually are a good investment your Tesla will appreciate in value you'll be able to use it as a robo taxi when you're not driving it again it's not true but it probably worked in raising the hype for Tesla it's less likely that Devon will replace software engineers and more likely that it'll make the founders and investors a lot of money and there are many times where people blurred the lines between hype and fraud notably examples include thonos and FTX very recently there's been a bunch of fake demos the Sora video was fake it was created largely by a video Studio the Google Gemini demo was fake so why did companies do this because it works

9:17 Devon AI is valued at $2 billion that's a lot for something that's currently just an open AI rapper especially when open AI itself is valued at $80 billion but to go back to answer the question of why build the hype let's think about it from a numerical standpoint if I have a company that's valued at $2 billion that's incredibly valuable because if I need to hire somebody I'll give them a salary and all also just throw in hey why not 100K in equity in my company you can't sell it yet it's just paper money for now but hey bro it's 100K don't pay attention to the fact fact that my company is not profitable now and it might never be my

10:03 company is worth 2 billion this especially applies to startups and non-public companies but it also obviously applies to publicly traded companies short-term stock gains are good for them because they can then issue shares at a higher valuation it's basic math now when it comes to the rate of improvement of these llms throwing better chips at them will help throwing better highquality data at them will help and definitely another research breakthrough like the transformer architecture will also help but it could also be possible that computers are just incompatible for the level of intelligence that many people are expecting them to have isn't it possible that a paradigm shift might be necessary who knows what it could be it could be

10:49 quantum computers or it could be some kind of organic material so here's how I think about this it's impossible to predict what the rate of improvement of AI is going to be but we do have a few data points already it's been almost 2 years since chat GPT was released and quite frankly from GPT 3.5 to GPT 40 the rate of improvement has obviously been very slow now who knows when GPT 5 releases it could be a dramatic Improvement just like how GPT 3.5 was and of course the bigger question isn't what's going to happen in 2 years it's what's going to happen in 10 or 20 years is the rate of improvement going to look like this or

11:35 is it going to look like this well we obviously don't know but if you're good at anything you know that as you get better and better it becomes harder and harder to improve this is analogous to so many things an example I think is really interesting is distributed systems if you have a system with 99% availability and now you you want to push that system to 99.9% availability it seems like it's just a 1% Improvement how hard could it be but that's a very common misconception it's not a 1% Improvement this is actually a 10x Improvement to go from 99 to

12:21 99.9 because the failure rate with a a system that's up 99% of the time is just 1% of the time with this system would be 0.1% there's 10 times fewer failures and of course we know with AI when it works people are fine like they're impressed but when it fails that really matters when it comes to Tesla if your car fails even one time it could be game over this needs to go down as small as possible so in fact we're not trying to go to 99.9% but I want it to see it I want my Tesla to go as far as it could possibly go it'll never be a 100 even humans aren't that good but I think it's

13:07 becoming clear that the rate of improvement eventually is going to dramatically slow down it's just not quite clear what the bottleneck is just yet I have so many examples I could share with you I'll share a couple very quickly one Tesla faked a self-driving demo in 2016 why would they do such a thing companies want you to think that they're better than they are they're improving more quickly than they are because if you buy at the bottom well eventually it's going to be up here and even if it never gets there well you already bought the product if I was a taxi driver and I quit my job over something Tesla or Elon Musk said I would feel kind of like a jackass fool me once shame on you fool me twice shame

13:52 on me so when Elon says we're going to have artificial general intelligence within 2 years I'm a little hesitant and I'm more inclined to listen to The Experts it's not like Elon makes claims purely from an emotional perspective it's I'm sure he's always driven by facts and logic and research I have a relatively simple way of making tough decisions given so many unknown variables and the first thing is it's not proven right now that software developer jobs are being automated who knows that could change overnight and even if it were to start happening it still wouldn't be proven and that we don't know if like the code quality will suffer so much and there will be so many outages that it wasn't worth it it

14:38 wasn't worth saving a little bit of money in the first place and second is more of a human aspect how can you live with regret if you were to drop out of your major right now or transition into a completely different career because you're predicting developers will be automated within a few years and it turns out that that doesn't happen happen can you live with that how would that make you feel obviously consider the alternative you change paths right now and you get a step ahead well what exactly are you getting a step ahead on I don't know what exactly you're planning on doing becoming an accountant a lawyer a physician because if developers get automated which jobs aren't going to be automated there is some aspect of at least we're all in

15:23 this together I would not make any drastic decisions especially when it comes to education guys guys we have calculators why are we still teaching kids math we have voice to text engines now nobody needs to know how to read and write I'm sure all the cuttingedge researchers right now have very poor math skills and very poor reading and writing skills the logic that goes into programming can be applied in so many areas especially business that's why some of the best Business Leaders today are technical Founders honest to God I think I owe most of my success in the education I received early in my life teaching me math physics and eventually programming now of course some things are really complex like architectural details but some things are simple it's

16:11 obvious that human brains are fundamentally different than the most advanced AI today humans only need at most a couple dozen hours to fully learn how to drive medium medium Li well at least most of us small children only need to see and an animal a few times like a cat to be able to identify it in the future this is just not true with AI today and I'm also saying that AI is going to make some improvements it's going to automate many things it's not in my opinion going to live up to this dramatic hype that we've seen but it doesn't need to live up to the hype to make some people a lot of money if you wish to make an apple pie

16:57 from scratch you must first invent the universe thank you very much