Auto-generated transcript Hey, have any of you guys watched Wall-E? It's one of my favorite movies from Pixar. It's a really awesome movie. It's like this dystopian sci-fi futuristic kind of a movie where humanity has basically just surrendered themselves to the bots, right? And the AI does everything for them. It thinks, it makes decisions for the humans, it feeds us, it fulfills all our basic needs, right? It's just an extremely dystopian movie where humanity has basically lost all agency all autonomy and i think that with modern day llms and ai we are genuinely moving towards that future and in the next 10 minutes or however long this video is i don't know i'm just recording it right now i'm going to yap a bit about why ai and llms are in my opinion which is correct destroying society and i bet i'm not even going to talk about something new this is probably something you've already noticed in your own work in your own life the more you use ai the more you're starting to lose that critical thinking skill right like you've got this weird slowness in your brain like if i shut down every single llm in the world if i shut down the cloud gpd websites right there would be a global catastrophe by now you guys are way too reliant on this like the world has become way too over reliant on these machines and there is an actual term for this it's called cognitive surrender and you can probably guess what that means basically humans have just surrender their ability to think and make decisions, right? Why? Because these bots can now kind of sort of do it for them. I mean, they don't make good decisions. They just make a decision, and apparently that's good enough, right? And by the way, the term cognitive surrender, I didn't come up with it. I just saw it in some news article. But for the purposes of this video, we're going to assume that I came up with it. Now, here's a very concrete example of this, especially in coding, right? Most of my audience is software developers, so you'll probably recognize this before ai you used to think through complex problems right you used to try to come up with solutions try to brainstorm a good solution and try to implement that right now you don't do that anymore and i'm saying you but i don't do it a lot either like now we just go straight to cloud or whatever type into the magical prompt box and just let the machine come up with solutions for us now. And when we do that, we're skipping the most important part of the problem solving process, which is critical thinking. It's a little something that most of humanity doesn't have anymore. Now, again, I'm not the first person to address this issue. There's a lot of really smart people smarter than even me if you can believe that who are raising their voice against this and trying to come up with a solution right like some sort of guardrails or other systems and in coding that would be something like i don't know manually reviewing the code that the ai generates to make sure it's not you know completely broken and let me just cut the bullshit right here nobody absolutely nobody is going to review thousands and thousands of lines of code generated every single day how often do you read someone else's code because that's what ai is right it's not code that you came up with it's not code that you wrote you have no idea what's in there you have no idea where to even start reading it all right how often do you read someone else's code not a lot how often do you read thousands and thousands of lines of that code not a lot how often do you enjoy that? Never. So explain to me how you're gonna review thousands and thousands of lines of AI generated code that you've never seen before, that you have no idea how it works. You're not, obviously. And besides, AI generated code does work like 90% of the time, right? It does what it's supposed to, it does what you ask it to 90% of the time, which means, let's say nine out of 10 times the agent generates really good code and you just let it pass right and the one time you just manually review it and you fix whatever issue the agent came up with right what happens when let's say like if the agent is generating good code nine times out of ten you're gonna mentally check out of the review process you're gonna assume all right it came up with really good code it's gonna come up with good code next time as well and you're just gonna mentally check out of the entire review process you're gonna think okay this agent is pretty good this model is pretty good i don't need to review it all that much and most of the time that'll work because ai will generate good code 90 of the time what happens on that one time when it does break right what happens then you're not going to review it obviously and it's going to cause problems down the line for your users maybe even financial loss maybe even worse legal troubles if you're working on a in a really big project in like an enterprise company like any workflow that requires a human being to review thousands and thousands of lines of ai generated code is doomed to fail from the very start it's not gonna happen it's not gonna work move on find something else like because it generates good code most of the time the cognitive tendency to just accept it and move on is right there right And unless you an absolute expert at what you do if you not some junior developer who doesn even know how to hide their environment secrets and API keys like if you a professional at what you do right you can probably guide the AI properly to you know write good code clean secure code But if you're not that kind of expert, you wouldn't even know where to begin. You wouldn't even know whether the AI generated code is good or not. And remember, you can't trust the agent's actual answer. You can't trust what it says. because agents are you know trained to sound as confident as possible right and we humans we are really bad at distinguishing language versus consciousness right and intent if if the bot says yeah i believe this is going to work this is why it's going to work and it gives you a bunch of plausible sounding reasons you are very very likely to accept it at face value unless you're again an expert who knows what the fuck he's doing and you go and review every single line that was generated right so no reviewing ai generated code is not the solution either you're not going to review thousands and thousands of lines of code you're going to end up getting lazy you're going to end up falling for the agent's confidence trick where it just sounds so confident that you accept whatever it generated and you're going to create a mess down the line so that's not the solution here we need something better than reviewing generated code right and even beyond all that is the issue of your skills atrophying like when the agent is generating good code most of the time and you're just inclined to accept it without even reviewing the code what do you think happens right like you're not practicing how to code you're not you know refining your skills as an engineer you're literally just describing what you want in natural human language which is something anyone can do except project managers and clients but yeah and you're letting the bot come up with all of the actual code right sooner or later like probably sooner than later you're gonna lose those skills of coding and systems design and everything unless you're practicing those skills you're gonna lose them especially since most of the time you're just accepting whatever code the agent comes up with you're gonna lose your most valuable skills and that's a really huge deal like we spend years as developers learning and honing our skills and the more we use AI and the better these models get, right? We're going to lose those skills over time. Like just, let's just run a quick thought experiment, right? When these models were really bad at code, you would have to guide them to write good code as well. As they're improving at code, you now have to guide them to structure that code better to you know implement some systems design stuff to have a good system architecture that allows you to scale up as your requirements change and evolve but once the agents can have context over all of those things as well what exactly do we have left like these models keep getting better and better right and so we keep abstracting away more and more of our skills we're gonna reach a time very soon where we're not even doing much except just describing what we want in natural language and letting the agent do it for us and when we do that when we cross that point we haven't crossed it yet but we're really getting close once we get to that point we have basically lost all of our skills as developers like we've just lost them no question about it and that's because we're not even practicing them anymore we we don't even need to practice them anymore because these bots can do it for us now except on those off times when the bot can't do it for us and then we're helplessly stuck trying to run that slot machine again and again changing and tweaking the prompts here and there trying to get the right output and just failing miserably and not being able to do it ourselves because we've lost those skills so yeah we're in for some dark times ahead and i don't know how we can fix that i don't really have a lot of solutions in mind as you can probably see this video was just me yapping about some thoughts running wild in my head and i just had to get them out to something but yeah if you have any thoughts do leave them in the comments below i'll make sure to read them because i'm very interested in getting some other opinions on this apart from myself like we're losing our most valuable skills to produce slop that nobody uses and in the process we're becoming more and more helpless and more and more reliant on these machines i mean sometimes i think the only people winning this ai race are open ai and entropic right like people are losing their most valuable skills and becoming helplessly reliant on these machines companies are spending millions hundreds of millions billions of dollars on these llm tokens and they have nothing to show for it like did you guys see uber spend 3.4 billion dollars in four months on just llm tokens that's 28 million dollars a day just spent on llm tokens and they have nothing to show for it so people are losing companies are losing the only one who's actually winning over here are the ai labs the only one who's winning is open ai and then topic they're just charging and getting billions and billions in revenue in llm tokens it's insane but yeah this video is getting way too long so i'm gonna end it over here do let me know your thoughts down below i'm very interested in hearing them and uh thanks for watching