Transcription
Think back to February 2020, right about 6 years ago to the month. The stock market was doing great. Your kids were in school. If someone told you they were stockpiling toilet paper, you would have thought they were crazy. But then, over the course of just about 3 weeks, the entire world changed.
Well, this article called "Something Big Is Happening" just went viral with over 47 million views in under 24 hours. It was written by Matt Schumer, the CEO of Hyperight, and his argument is pretty simple. He thinks we're in that exact same "these things are kind of overblown" phase with AI. But in this article, Matt actually argues that this time, it's actually bigger than COVID.
Now, this video is actually very heavily inspired by this article, and I'm going to link to it in the description because I highly recommend that you read the whole thing. It is definitely worth your time.
Now, there's also some solid counterarguments to the sort of COVID analogy. Like John Kugan from TVPN here, pretty accurately explains that the number of people who caught COVID didn't follow an exponential curve. It was a logistic curve. It was growing exponentially from 2020 to about 2022 and then started decaying. Eventually, the number of cases plateaued. Mathematically, that was inevitable because to keep growing indefinitely, you need an infinite supply of humans that can get infected, and there's only 8 billion people on Earth.
I also kind of agree that the COVID analogy is a little bit dramatic and not totally accurate, but I think the point still stands that a lot of people are not prepared for what's about to happen. I also don't think AI is going to like sneak up on people like COVID kind of did, where people were dismissing it for a couple months and then all of a sudden, boom, it was here. No, AI has been bubbling up and bubbling up and bubbling up for years. So, I don't totally appreciate the analogy, but I do think that a lot of people are not ready for what's coming. So, let's get into it.
The core argument of this article gets into something that I think is really, really important to understand. Matt says that the reason so many people in the industry are sounding the alarm right now is because this has already happened to them. They're not making predictions. They're telling you what already occurred in their own jobs and warning you that you're next. He says, "I'm no longer needed for the actual technical work of my job. I describe what I want built in plain English, and it just appears. Not a rough draft I need to fix, the finished thing. I tell the AI what I want, walk away from my computer for 4 hours, and come back to find the work done, done well, done better than I would have done myself with no corrections needed. A couple months ago, I was going back and forth with the AI, guiding it, making edits. Now, I just describe the outcome and leave."
And personally, I've been finding the same sort of thing for myself. I built this mattwolf.com site with basically one prompt. It's very simple, but it automatically pulls in the subscribers from my YouTube channel, automatically pulls in anytime I make a new video, pulls in my latest podcast from The Next Wave, pulls in all of my writing, shares my portfolio, and even this down here where it asks you to subscribe to the newsletter. Well, this number right here is actually pulled in automatically from my newsletter subscribers. And this whole site was built in 45 minutes, maybe.
I'm also in the process of completely overhauling and rebuilding Future Tools to be more focused on the news. It's still going to have the tools database, but the focus is going to be more on news. And the point that this site is at right now, I mean, we're talking maybe an hour or so of telling AI what I want. I mean, for this YouTube channel that you're watching, I built a tool that generates thumbnails for me. I gave it a whole bunch of headshots and facial expressions for it to pull in, a handful of logos that I often use, and some of my past thumbnails as examples of what I'm looking for. And I just walk through this step-by-step process, and it generates thumbnails for me. Now, I built this in a couple hours.
I built a tool called Video Titler. When I'm done recording this video, I'm going to take the video file, drag and drop it right here, and it's going to suggest a whole bunch of titles for this video and then rank it on which one it thinks is going to perform the best. And then I'll probably use that title for this video. Again, this was probably done with one prompt. Actually, I think it was two prompts because after it made the upload video, I also wanted to give it the option to upload a transcript if I already had it transcribed. So, two prompts to make this thing. Doesn't look pretty, but it solves a real problem for me and handles it very quickly.
Almost everything I just showed you was made in the past like week since GPT-4.5 CodeX came out and Claude Opus 4.6 came out. And here's what Matt has to say about those models. "They weren't just executing my instructions. They were making intelligent decisions. It had something that felt, for the first time, like judgment, like taste. The inexplicable sense of knowing what the right call is that people always said AI would never have."
Now, I want to be transparent. Matt Schumer does run an AI company, so he does have skin in the game. But I will say this, I've been using these tools, specifically GPT-4.5 CodeX and Claude Opus 4.6, for the past week, like nonstop, and they definitely feel different. Like, we definitely crossed a threshold with these models. And we're not the only ones feeling this way either. The vibes on Reddit have completely shifted since these new models came out. We're seeing posts like, "GPT-4.5 CodeX is a massive improvement. I used to spam prompt cues with 'please fix,' 'check for bugs,' but now 5.3 CodeX seems to do this for me already." Here's another one: "GPT-4.5 CodeX just dropped, and it's scary good. We can play with multiple external tools to get things done, testing, taking screenshots, etc." Here's a comparison somebody did between 5.3 and Opus 4.6: "Codex optimizes for momentum, not elegance. Opus optimizes for coherence, not speed. Codex assumes you'll iterate anyway. Opus assumes you care about getting it right the first time." Both these models are really good, and the people that are actually using them are agreeing.
Now, I already know what some of you are probably already thinking, because I hear it often and see it in the comments constantly. "But I tried AI and it wasn't that good. It made stuff up. It hallucinates. It wrote crappy code. It just wasn't impressive." And this article nails that. If you tried ChatGPT in 2023 or 2024 and thought, "This makes stuff up or this isn't that impressive." Well, back then, you were right. The early versions were genuinely limited. But that was 2 years ago. In AI time, that is ancient history. The models that are available today are unrecognizable from what existed even six months ago. Anyone still making that argument either hasn't used the current models, has an incentive to downplay what's happening, or is evaluating based on an experience from 2024 that is no longer relevant. It's like if you used a flip phone back in 2005 and somebody told you that phones are about to change the world forever, you'd probably be like, "No, they're not." But you weren't evaluating smartphones. You were evaluating the flip phone that came before it.
Part of the problem is that most people are using the free version of AI tools. And well, the free versions are over a year behind what paying users have access to. The people paying for the best tools and actually using them daily for real work know what's coming. So if you're somebody that's been on the fence and just refuses to use AI 'cause you used it a couple years ago and it wasn't for you. Well, the thing you tried a year ago or two years ago or even 6 months ago is not what we have access to today. It is so much better and so much different now.
Next, let's talk about how fast this stuff is actually moving. If you haven't been watching super closely, this is the part that I think is very difficult for people to comprehend. In 2022, AI couldn't do basic arithmetic reliably. It would confidently tell you that 7 * 8 was 54. By 2023, it could pass the bar exam. By 2024, it could write working software and explain graduate-level science. By late 2025, some of the best engineers in the world said they had handed over most of their coding work to AI. And then on February 5th, 2026, new models arrived that made everything before them feel like we're in a different era. We just saw people talking about that on Reddit.
Here's where it gets even more interesting. There's this company called Meter, MER. They do evaluations and research on AI models. They track how long a task takes by a human expert and compare it to how long it takes for an AI to do the same task. So let's take a look at this chart here. When GPT-3.5 came out back in March 2022, this was what you used when you used ChatGPT. If you used it when it first launched, that was GPT-3.5. It can do tasks that took humans about 36 seconds to do. Not super impressive. Fast forward to one year later, March 2023, GPT-4 comes out, and it could handle tasks that took human experts about 4 minutes to do. Then in May of 2024, 6-minute tasks. June 2024, 11-minute tasks. September 2024, 19-minute tasks. October 2024, 20-minute tasks. You know, fairly slow rate of improvement. Then we got 01. This is when the model started using chain of thought, where it would think a little bit longer. You can see it's thinking, and then it would come to its conclusion. 01 would accomplish 38-minute tasks. Then we got Claude 3.7 Sonnet. It could do 60-minute tasks. We're talking from December to February. 3 months, we jumped from 38-minute tasks to 60-minute tasks. Then we got 03. Now we're doing 2-hour and 1-minute tasks. February, March, April, 3 months later, it doubled. Then we got GPT-5 in August, and now it's doing tasks that would have taken a human expert 3 hours and 34 minutes. Then Claude Opus 4.5 in November, 5-hour and 20-minute tasks. And now GPT-5.2 is doing 6-hour and 34-minute tasks that a human expert would do and can do upwards of 17-hour tasks. And we don't even have the newest models on this chart. The length of tasks AI can do is doubling every 7 months.
And this isn't just happening in coding either. Meter actually published a follow-up study about how this time horizon translates across various other domains. We're seeing similar growth curves in math, in full self-driving vehicles, in Google proof question and answers, in the ability to use browsers, simulated robotics, video Q&A, scientific Q&A. So, it's not just coding where we're seeing this acceleration happen. It's across everything. One researcher called this, and I quote, "probably the most important single piece of evidence about AGI timelines right now."
Now, here's where it gets really crazy. If you extend this trend that's held for about 6 years now, without any signs of slowing down, we're looking at AIs that can work independently and autonomously for days at a time, if not weeks at a time, within a year. And within 3 years, it'll be working on projects that would normally take humans months to do completely autonomously. Dario Amodei, the CEO of Anthropic, the company that makes Claude, has been on record saying, "It is my guess that by 2026 or 2027, we will have AI systems that are broadly better than all humans at almost all things." I'm recording this in 2026. So, think about it. If an AI is smarter than like all PhD students, don't you think it's probably going to be capable of most office jobs as well?
"The whole thing about exponentials is, you know, it looks like it's going very, very slowly. It speeds up a little bit, and then it just zooms past you. And I I I think we're I think we're on the precipice. I think we're a year or two away from it really zooming past us."
Okay. So, you might look at that graph and say, "Well, this has to slow down eventually, right?" And well, in actuality, most researchers actually think it's going to speed up. And the reason being, AI is now building its own next AI. On February 5th, OpenAI launched GPT-4.5 CodeX. That was the model that everybody was saying, "This is starting to feel different." But there's a line in this announcement that really had people sort of perking up a bit. "GPT-4.5 CodeX is our first model that was instrumental in creating itself. The CodeX team used early versions to debug its own training, manage its own deployment, and diagnose test results and evaluations. Our team was blown away by how much Codex was able to accelerate its own development. CodeX goes from an agent that can write and review code to an agent that can do nearly anything developers and professionals can do on a computer."
This is absolutely insane, and I really, really want to try to bring home this point. The AI helped build itself. This isn't a prediction about what might happen someday. This is OpenAI telling you today in a press release that the AIs are now creating the next AIs. And they're not the only ones saying this either. Dario from Anthropic put out this blog post on his website called "The Adolescence of Technology" in January of 2026. In this blog post, he says, "AI is now writing much of the code at Anthropic. It is already substantially accelerating the rate of our progress in building the next generation of AI systems. This feedback loop is gathering steam month by month and may be only 1 to 2 years away from a point where the current generation of AI autonomously builds the next. The loop has already started and will accelerate rapidly in the coming months and years."
Here's another blog post from Anthropic's website. "Anthropic engineers and researchers use Claude most often for fixing code errors and learning about the codebase." This is Anthropic, the company that made Claude, saying that their engineers and researchers are using Claude to make their models better. "Claude is handling increasingly complex tasks more autonomously. 6 months ago, Claude Code would complete about 10 actions on its own before needing human input. Now, it generally handles around 20, needing less frequent human steering to complete more complex workflows." And this is from December of 2025, before these newer models even came out.
So, just think about how this loop works. Smarter AIs write better code. Better code then writes smarter AIs. Then, that smarter AI writes even better code. And that better code writes a smarter AI. And this cycle is continuing to turn and pick up speed and just pick up more and more momentum. This flywheel is spinning like crazy right now. Researchers are literally calling this the intelligence explosion. That's why this Meter graph might continue to accelerate. The thing getting smarter is making it easier to make the next thing even smarter.
Okay, so knowing all this, what does it actually mean for real-world jobs? So far, I've been really focused on coding. And you're probably thinking, "I'm not a coder. What do I care?" And it matters to your job and what you do, no matter what it is, probably more than you realize. Last year, Dario did a series of interviews where he predicted that 50% of entry-level jobs will disappear and that we'd see a 10 to 20% unemployment rate because of AI. This would be the highest joblessness levels since the Great Depression. In fact, this is an exact quote from Dario: "We, as the producers of this technology, have a duty and obligation to be honest about what is coming. I don't think this is on people's radar."
And Dario is not the only one sounding the alarm right now. Two researchers from Anthropic, Schulto Douglas and Trenton Bricken, went on the Daresk podcast and predicted widespread automation of white-collar work could happen within just a few years, going on to say it could be a "pretty terrible decade" as AI automates more white-collar work. Heck, just this week, Marium Charma left his job at Anthropic and did a mini-essay as to why he left. And in that essay, he says, "I continuously find myself reckoning with our situation. The world is in peril. We appear to be approaching a threshold where our wisdom must grow in equal measure to our capacity to affect the world, lest we face the consequences." Jensen Huang, the CEO of Nvidia, said that "every single job will be affected." Kai-Fu Lee called predictions that AI will displace 50% of jobs "uncannily accurate."
So yes, code was sort of the first thing that these AI models tackled, but Matt Schumer made a great point in this article. The AI labs made a deliberate choice. They focused on making AI great at writing code first because building AI requires a lot of code. If AI can write that code, it can help build the next version of itself. We've already talked about that. A smarter version, which writes better code, which builds even smarter versions. Making AI great at coding was the strategy that unlocks everything else. That's why they did it first. AI isn't replacing one specific skill. It's a substitute for work that requires thinking. It's getting better at everything simultaneously.
Now, when factories started automating, humans had other roles that they could fall into. They were able to take the office, like admin jobs. When the internet disrupted retail jobs, people were able to move into logistics and services and things like that. There was always somewhere to go next. But right now, AI isn't really leaving much of a convenient gap to fall into. Whatever you might be retraining for, AI is going to be able to do that, too. And specific industries like legal work, financial analysis, software engineering, content writing, customer service, medical analysis, it's not coming for those jobs someday. Those capabilities are arriving right now. It's going to take time for this to ripple through the economy, but the underlying ability, those capabilities, they're already here.
A lot of people find comfort in the idea that certain things are safe. That AI can handle the grunt work but can't replace human judgment, creativity, strategic thinking, empathy. However, the most recent AI models make decisions that feel like judgment. They show something that looked like taste. A year ago, that would have been unthinkable. And well, his rule of thumb from this article: if a model shows even a hint of a capability today, the next generation will be genuinely good at it. These things improve exponentially, not linearly.
Okay, now let's actually zoom out for a second because this actually goes further than jobs. In this giant essay that Dario wrote on his blog, he introduced this thought experiment. Matt Schumer's article references this same exact thought experiment, but I wanted to show you the actual source here. I think the best way to get a handle of the risk of AI is to take the following question. "Suppose a literal country of geniuses were to materialize somewhere in the world in roughly 2027. Imagine say 50 million people, all of whom are much more capable than any Nobel Prize winner, statesman, or technologist. Suppose you were the national security advisor of a major state responsible for assessing and responding to the situation. Imagine further that because AI systems can operate hundreds of times faster than humans, this country is operating with a time advantage relative to all other countries. For every cognitive action we can take, this country can take 10. They don't sleep and they're constantly thinking. So what should we be worried about? Autonomy risks, misuse for destruction, misuse for seizing power, economic disruption, indirect effects, things like radically destabilizing everything." The person in charge of security would probably say that this is the single most serious national security threat we've faced in a century, possibly ever.
And much like Matt Schumer's article, this article from Dario was designed to be a wake-up call. He says, "This essay is an attempt, a possibly futile one, to jolt people awake." And that's also why I'm making this video, because if you're not sort of in the weeds using this stuff every day and seeing the progress, you're probably not realizing what's happening right now.
Now, this portion here is also from Dario's article, but Matt does a good job of summing it up here. "The upside, if we get this right, is staggering. AI could compress a century of medical research into a decade. Cancer, Alzheimer's, infectious disease, aging itself. These researchers genuinely believe these are solvable within our lifetime. However, the downside, if we get it wrong, is equally real. We've already seen attempts at deception, manipulation, and blackmail in controlled tests." Heck, I did a whole video about how these models are cheating on the benchmarks that they're presented with. The people building this technology are simultaneously more excited and more frightened than anyone else on the planet. They believe it's too powerful to stop, but also too important to abandon.
Okay, so this video has been pretty intense, and I know that. Definitely a different vibe from my normal videos, but I felt like this was too important of a topic not to talk about. And not everyone agrees with this exact timeline either. In fact, here was an article on SpyGlass written by M.G. Seigler. He says, "What Matt Schumer is actually describing, if you strip away the apocalyptic framing, is a technology that is very useful, improving quickly, and that will probably change a lot of jobs over the next 5 to 10 years, which is correct and also not a novel observation, and also not COVID. It's closer to the internet, which did in fact transform virtually every industry, but over the course of decades, not months, and in ways that were far more nuanced and surprising than anyone predicted back in 1995."
And heck, there was even this clip on Bloomberg that was claiming that we're seeing a rise in some jobs as other jobs go away.
"I wouldn't see AI as something that that would affect the long-term prospects for employment because we we already are seeing more more adoption of new roles. So now consulting firms are hiring these AI roles, you know, new roles they didn't employ before, and those actually exceed the number of entry-level consultants, which is kind of wild because, you know, we think of consulting firms as this launching pad for, you know, early career hires, and that's less the case now than it has been."
So my honest take, even the skeptics aren't saying this isn't happening. They're saying it might take 10 years instead of two. And honestly, whether it is 2 years or 10 years, the actions you should be doing right now stay the same. You need to start engaging with this stuff.
So, let's talk about the practical stuff. What can you actually do? I don't want you to walk away from this video feeling helpless. That wasn't my intention. The single biggest advantage you can have right now is simply being early. Early to understand it, early to use it, early to adapt. And I could not agree more with that take. That is the single reason that I started this YouTube channel and started talking about AI over 6 years ago now. I wanted to show people what you are capable of now that you weren't capable of prior to that video. And things have escalated much faster than I ever anticipated when I started this channel. My thoughts on AI have completely changed. The reality of what it's doing to the world compared to what I thought it would do have completely changed. And who knows, my thoughts on this a year from now can be completely different as well.
But the advice that Matt Schumer gives here is the exact same advice that I found myself giving over and over again. And I know that's very convenient in the context of this video, but if you've ever seen me speak on stage or go on other podcasts, this is literally the advice I've been giving people for like years now.
Number one, start using AI. Seriously, not just as a search engine. Use the paid versions of Claude, ChatGPT. Use the $20 a month versions. Again, the versions that are on the free plans are like a year behind. They're not as smart. You're not seeing what these AI models are capable of if you are on the free plans.
Second, when you do use it, don't just ask it quick questions. It's not a replacement for Google. Push it to actually do work. Like in my example, I found things I needed. I found bottlenecks in my business. I wanted to rebuild Future Tools. I wanted to build a mattwolf.com site. I needed a faster way to make thumbnails. I wanted help getting title ideas for videos. I just went out and tried to build little tools that would do those, and now I've solved all of those problems for myself. I'm not just asking ChatGPT for a solution. I'm going and asking these tools to build me solutions.
Number three, don't assume it can't do something just because it seems too hard. Try it. If you're a lawyer, don't just use it for quick research questions. Give it an entire contract and ask it to draft a counter-proposal. If you're an accountant, don't just ask it to explain a tax rule. Give it a client's full return and see what it finds. The first attempt probably won't be perfect, but then you can iterate, rephrase what you asked, give it more context, and try again. And you'll probably be shocked that it actually works and gets you closer than you thought it would. As we always say, "This is the worst it's ever going to be." And you'll be surprised by how like good this current worst version is.
And four, be the person who understands what's coming and can show others how to navigate it. The person who walks into a meeting and says, "I used AI to do this analysis in an hour instead of three days," is going to be the most valuable person in the room, not eventually, right now. Spend an hour a day just playing with these tools, at least an hour a day, not reading about it, not watching YouTube videos about it, actually getting in and playing with the tools, trying new tools, discovering new tools. Heck, I put Future Tools together so you can find tools that you didn't know existed. If you do this for a few months, you'll be ahead of 99% of the world. Almost nobody is actually doing this. Most people are just complaining about how they don't like AI slop on social media and use ChatGPT as a replacement for Google. Most people are not using this as a tool in their day-to-day life. You want to be ahead of people, figure out how to use it for more than just what everybody else is doing with it.
And also keep in mind that now is the best time ever to do that thing that you might have always wanted to do. Did you ever have an idea for an app but didn't know how to make it? Now it's easier than ever to build it. Did you ever want to write a book but thought of it as too hard? Now AI can assist you with that. Want to build an e-commerce business but don't know where to start? Yep. AI is going to help you with that as well. There is no better time in history to go test and build something and just try it out. The barrier to entry for almost anything is gone, and you just got the best tutor in the world handed to you.
And again, I didn't do this video 'cause I wanted to scare people. I don't even agree with every single prediction being made. But what I do know is that the people who understand this stuff and use it more than anyone else on the planet, they're telling us to pay attention. They're sounding the alarm bells and telling us, "Hey, this is moving faster than you realize." When the CEO of one of the biggest AI companies on the planet tells you, "Hey, we need to perk up and pay attention to what's going on," and we have a duty to be honest about what's coming, I feel like we should probably pay attention.
Again, the link to the original article is down in the description. I still highly recommend reading it. I did try to cover as much ground around it as possible, but it's definitely worth the read. And I think Matt Schubert deserves credit and should get the eyeballs on it. But I really felt it was important to make this video because I think this is the type of video that should be shared around. Anybody who is sort of like on the sidelines or putting up blinders to what's going on with AI or saying, "I will never touch AI. AI is evil." Well, guess what? You're going to be the people that fall behind. And I stand by that statement. Doesn't mean you have to love AI. It doesn't mean you have to spend tons of money with all the big companies in the world, but it does mean you should be aware and you should be paying attention to it. And if your ego gets in the way and says, "No, I'm ignoring AI on principle." Well, I hate to break it to you, but that's not going to be a good strategy in the long run.
So, again, I wanted to put this video out so that people know how fast this is coming and what's really steamrolling towards us right now because I do feel like there's a lot of videos that are just hyping up the next tool or showing you the latest breakdown of how to use a specific tool. Heck, I make a lot of those videos myself, and I'm going to go back to making a lot of those videos 'cause that's what this channel's going to be. But I felt like this video was an important one to get out, and I wanted to get it out.
And if you like staying in the loop with AI and you want to know about all the latest news, that's what this channel's for. Every week I'm breaking down the latest news so that you are keeping up with the pace of AI as much as I'm keeping up with the pace of AI. I'm showing off tools and how to use them so that you can stay looped in and that you can actually be ready, be that person that has your finger on the pulse, that is on the cutting edge of how to use this stuff because it's going to be such an important skill in the very, very near future. So, if you're that person who knows that's important, maybe consider liking this video and subscribing to this channel, and I will do my best to make sure that you stay as tapped in as humanly possible. That's what I got for you today. I know this was a heavy topic. I did a lot of research into this one, and I just felt it was important to get out. So, thanks for tuning in. Thanks for nerding out. Hopefully, I'll see you in the next one.