"1,000 days left" Anthropic founder

本站添加: 2026-05-12

[00:00:00]So, we will in our lifetimes live through one of the wildest transitions that humanity has ever seen. We're kind of approaching the endgame here, folks.

[00:00:11]So, you know that whole thing about the frog in boiling water. As long as you raise the temperature slowly, you can boil the frog, it won't notice. Just as long as you slowly increase the heat, everything's fine. It won't register the point at which the water becomes too hot for its own survival. I love AI. I love where kind of things are heading. I'm very excited about the future. There are a few things, I've said it many times before the in kind of medium term that I am extremely extremely worried about.

[00:00:42]Here's a recent post by Jack Clark, the co-founder of Anthropic, and he actually talks about both of those things. So, first and foremost, here's what he said.

[00:00:52]And if you miss this, it's important that you kind of register this cuz we keep hearing these crazy headlines and after a while, just like the frog, we kind of become immune to it. But here's again Jack Clark. He was the head of policy at OpenAI. He was the co-founder of Anthropic. He is an insider at a frontier AI lab. He's very close to policy makers, kind of the government circles. He's at the intersection. You know that ven diagram. He's like that one person in both those circles. He's locked in. And what he thinks is going to happen is this. He's saying that now he has this viewpoint. He reluctantly came to the view there's a likely chance 60 plus% that no human involved AI research and development. Meaning it's an AI system powerful enough that it builds its own successor fully autonomously. Right? So if we build version one and version two, version three let's say or whatever you know version five right version five builds version six and then version six builds version seven etc. We've been talking about exactly that on this channel and we've been tracking how ever so slowly we kind of were getting to that. we haven't yet crossed the line, but he's saying that it's possible, it's likely even that this will happen by the end of 2028. He's saying this is a big deal.

[00:02:17]Again, not a hype guy. He's saying pay attention to this. He's saying I don't know how to wrap my head around it.

[00:02:23]We've never had anything like this in the history of humanity, in the history of the world. This is a world's first.

[00:02:31]Saying we understand it would be like monkeys saying they will understand all of the sort of consequences of, you know, the homo sapiens being born out there in the world and understanding all of the things that that will cause. If you get nothing else out of this video, just get this. You will be alive in a time, most likely, God willing, all of us will be alive at a time where RSI kicks in. Recursive self-improvement or automated AI research and development, the intelligence explosion, whatever you want to call it, it's coming online, you know, soon. Coming soon. Remember this chart from wait but why kind of this is the highest intelligence on Earth. And then here we create ASI and boom, it just goes straight up. The recent chart from meter research that tracks agent capabilities. It it does kind of look like that with the Claude Mythos preview. This was three months ago. A scary looking chart to begin with and then boom, that's Claude Mythos. Also, you might recall we talked quite a bit about Google Deep Mine trying to hire a kind of a new role at the company.

[00:03:40]Probably a job description that have never been seen before. Well, they finally hired somebody. Here it is. Alex Emos this week. He's saying he's starting at Google deepmind as director of AGI economics working with Shane Le that was the person initially talking about this co-founder of Google DeepMind Shane Le I think put it excellently when he said that you know how we have this system of contributing our manual andor cognitive labor in exchange for access to resources right that's kind of what our economy is but if you think about it it's very general that was kind of our system going back to the hunter and gatherer days everybody contributed what they could whether it was their knowledge their skills weaving baskets collecting berries or hunting whatever and in return they got access to resources food a place around the campfire etc well Shane leg was saying that that system is going to get disrupted by AGI and we really need to think about a new system so at Google deepmind one of the higherups they're thinking and hiring for right in a very real way they're preparing for how Frontier your AI can and will reshape the economy. What happens to wealth?

[00:04:52]What happens to labor? What happens to, you know, jobs? How do institutions adapt? How AI agents shape the market?

[00:04:59]Right? So, the the top echelon over there at Google Deep Mind, that's what they're thinking about. Jack Clark, who's also at the higher echelons of these frontier AI labs, you know what he's thinking about also some of those things. How are jobs going to continue?

[00:05:15]What is the human economy? What is the AGI economy? So, we will in our lifetimes live through one of the wildest transitions that humanity has ever seen. And I wanted to read this Jack Clark post, at least some of the highlights from it, because I think it's important that people understand what's coming. We're kind of approaching the endgame here, folks. So, please do me a favor, hit subscribe, hit thumbs up. I know you hear that a lot, but it does help a lot and it really helps me get this message out there because stuff's getting serious and I've been talking about for a few years and if you kind of talk about enough and discuss it enough for a while, you just you can't be in that state of hyper excitement all the time, it becomes more mundane. But it's important every once in a while to kind of like rejolt ourselves, realvanize ourselves and understand like this is happening. It's not a joke. It's not a marketing campaign. It's not just hype.

[00:06:11]Like, it's coming. Like, for real. Like, for real, for real, for real, for real, for real. All right. So, here's the thing. AI systems are about to start building themselves. What does that mean? As Jack Clark is saying here, this is a big deal. I don't know how to wrap my head around it. He's saying, "It's a reluctant view because the implications are so large that I feel dwarfed by them." And I'm not sure society is ready for the kinds of changes implied by achieving automated AI R&D. If that happens, we will cross a Rubicon into a nearly impossible to forecast future.

[00:06:42]The Rubicon, if you're not familiar with the entire history of Julius Caesar and the Roman Empire, it just basically means the point of no return. Like, if you've ever screamed yolo and then did the thing, then then you've crossed the Rubicon, then you're committed. So, a big part of this essay is him talking about why he thinks the takeoff towards fully automated AI R&D is happening.

[00:07:07]Now, we'll cover some of the big points here. A lot of the stuff I do on this channel goes over various research and stuff like that that shows why we're approaching that. So, hopefully we don't need to dig too deep deeply because we cover this in maybe not every video, but quite a few of them. So, he's going to discuss the consequences of this, but he mostly expects to spend the majority of the essay discussing the evidence for this belief. And here's where it gets super interesting for me. He's saying he will spend most of 2026 working through the implications. This is what I sort of cheer and applaud for. This is why I really was excited about Shane Le and the team at Google DeepMind hiring somebody to be an AGI economist. We need more very smart, very aware people thinking about what happens because it's nearly to me seemingly impossible to predict what happens since we don't have a model for it. We we have never seen anything like this happen before. If you're a writer, an author, you know, there's certain things that are very hard to write about. One of the hardest is, at least according to people like Larry Nan and Arson Scott Card, who are excellent sci-fi writers, they both, I believe, wrote about how difficult it is to write about a character that is much smarter than the that is much smarter than they are. Scott Adams was another person that actually wrote about this at length, I believe. So, if you're writing a book in which a character is super intelligent, much smarter than you are, it's going to be difficult to fully understand what he's thinking, how he's planning, because again, you don't have access to those cognitive capabilities.

[00:08:45]So, you're kind of just guessing. You're trying to model, you're trying to understand it, and forecasting the future after we have something like this automated AI researcher, ASI, whatever you want to call it. The terminology here is a little bit muddy, but what happens then? It's extremely difficult to predict. like we don't know if it's going to be the best thing that ever happened to humanity, the worst thing, somewhere in between. We have no clue.

[00:09:08]So, very much looking forward to reading more about his takes as to the implications of this thing kind of floating through the portal into this world. Just in case this is the first time you're hearing about this idea of a fully automated AI researcher, the important thing to understand is that there's been tons of papers and other sort of demonstrations published that are showing where we are in the process, how to do this. So, this isn't some mystical mythical thing that we're talking about. A lot of AI research is actually pretty simple on the surface. A lot of it is you come up with a hypothesis, then you create some code to run that hypothesis, then you run the hypothesis of your code, then you get the results, and you know, that's it. A lot of this isn't like if you were in microbiology or something like that where you have to sit there with a pipet and you know take your little colonies of bacteria and whatever substrate they're in and actually do that like the whole physical thing is more or less removed. A lot of AI progress is writing words and writing code and and then some math. I know I'm greatly simplifying it but it's important to understand that this is literally a lot of this is in the domain of the large language models.

[00:10:20]This is what they do. One of the most notable achievements in this field of an AI model actually improving itself or future generations of it. Alpha Evolve was a very notable case. Not by any stretch the only one. Keep that in mind.

[00:10:34]We've covered dozens of these, but this one's probably the biggest lab, the most well-known one. Alphve uses Gemini, Google's large language model, that family of models with harness. So similar to open claw or claw code or any of the other stuff you can think of it as an agent of sorts or a large language model with a harness around it. In this case, we have various prompt samplers.

[00:10:56]We have evaluator pools. We have program database all searching for the best program. Whatever that is could be a new algorithm or a new design, but at the center of it is the LM either one or multiple. Here they use an LM's ensemble, right? So, multiple depending on what it's used for. Could be several LM working together. Really, all you need is something that's verifiable. You need to be able to evaluate the outputs and be able to say is it better or not.

[00:11:26]So you can't do this or it's hard to do it for something like a poem. Who's to say if one poem is better than another?

[00:11:32]But for a lot of other stuff you can improve. If you're moving and you're able to shove 20 boxes in the back of a U-Haul and then your friend comes over and he's able to reorganize them and shove 23 boxes, that means, you know, his way of shoving those things in is better. That's a simple example, but with alpha evolve, they're using it for fairly complicated stuff. In genomics, they've improved how well they're able to, you know, correct these DNA sequencing errors. They're working with pack bio to analyze genetic data more accurately at a lower cost, right? Both those things are easily verifiable. Is it more accurate? If yes, then it it's better. That means Alpha Valve found a better way to analyze it. They also improved the power flow with various electricity grids. A lot of the quantum stuff we've been hearing, the quantum correction especially that was driven by AI as far as we can tell. Alpha cubit specifically was able to reduce the number of errors in these quantum calculations. It's also been able to prove a lot of these airish problems considered very very difficult and Trent Tao we've talked about this before one of the most sort of acknowledged and revered mathematicians probably the number one pure mathematician in the world today. He's saying that sometime at the end of 2025, these AI models that there was this inflection point where they became truly useful for mathematical discovery. But most importantly, or at least most importantly to this video, Alpha Evolve is improving AI infrastructure. It's improving itself. Alpha has been used as a regular tool to optimize the design of next generation of TPUs. Right? So TPUs are those chips on which they train the model, on which they run inference. the hardware that runs and builds AI is improved by this AI and improved various cache replacement policies and in the previous publications that they did about alpha evolved they've also mentioned that it improved the training process of Gemini of itself right so kind of the future models that will be trained the efficiency of how they're trained is improved by at the core Gemini itself so we're definitely beginning to see recursive self-improvement or at least something like at the beginning stages of it. As Sam Alman once said, the laral stages of recursive self-improvement, we're beginning to see it. That flywheel is slowly picking up steam. It's not going full speed ahead yet, but it's beginning to move. We're somewhere in that inflection curve. So, back to Jack Clark's piece, he points to a lot of the reasons why he thinks this is beginning to happen and why it will likely happen by the end of 2028. Number one is the coding singularity. The capabilities of these models to code is getting pretty insane. As somebody that's been using it pretty much every single day since last 2 3 years since they started coding, I am kind of blown away. This is the clearest and fastest progress that we've seen with these AI models. The two things that exemplify this trend are Sweet Bench and the meter time horizon plot. This is the meter plot. Now, this is the, you know, not the log view. This is the the regular view and so it definitely shows that exponential increase but just notice this is for various software engineering task that's what sui is a software engineering right so it went from you know maybe being able to fix a bug in a small python library you know somewhere towards the end of 2025 to now just this massive jump upward bench is realworld software engineering problems cloud mythos preview gets 93.9 effectively saturating the benchmark Tons of banks and global banking institutions are warning about specifically claude mythos preview but also models of that capability as they are coming online in the future. They're saying that this could be a massive cyber security issue for banks as well as many many other industries and just various banks and financial organizations might be a particular target because well because that's where the money is. The important thing to understand about models like Mythos is that while it is true that the coding aspect of these models was a big focus for various Frontier Labs like they did focus on that ability more so than others. The important thing to understand is that for example Claude Mythos cyber security abilities they're not there because Anthropic really sat down was like let's really make it good at this particular thing. This was an emergent property of it just getting better across the board at coding and also the model getting larger. Cloud mythos is probably one of the largest models by parameter there is. It's much bigger than opus. It's sort of like that fourth class above opus as far as the size of anthropic models go. So it's important to understand as these models are getting smarter and bigger. They're getting smarter across the board.

[00:16:14]There's more and more emergent capabilities. His next point is that these models are getting more skilled and better able to work at these long horizon tasks without human input.

[00:16:26]They're able to work independently of people for longer and longer periods before needing recalibration. AI is also getting good at core science skills that are essential to AI research and development. If you think about modern science, a huge amount of it is specifying a direction where you want to generate some empirical information, running experiments to generate that information, then sanity checking the results of the experiment. I would argue that these models are pretty good at all of this. The sanity checking, sometimes you have to have a human step in and do the final sanity checks because every once in a while, these models still derp out fantastically. But interestingly with research, if it's running a lot and contributing, you know, useful information, the mistakes aren't like a huge problem. Like if an engineer builds a train or a bridge and there's a mistake and the whole thing falls down, that's a huge problem. If you leave one of these models running overnight and you come back in the morning and it went astray, yeah, that kind of sucks. But as long as most mornings you wake up to there being valuable data or some new software being built, something that's valuable, that's fine. the coding I do with these models, every once in a while it does something wrong. But I wouldn't ever dream of it completely not using it because it's a net positive and it's much much more helpful than the minor inconvenience of checking its work and maybe sometimes telling it to redo it again kind of specifying what it did wrong saying no do it this way. A big part of AI research is you know looking at the various already published scientific papers and seeing if you can replicate the results. So somebody did an experiment and said these were my results. You replicate that experiment and you see do I also get those results?

[00:18:03]Does that also make sense in my setup?

[00:18:06]If you're able to replicate it, well I mean that's that's science. That means we found something that's replicatable and usable etc. So that's the whole point of core bench like can these models get one of these machine learning papers reproduce the results of a research paper given the repository right? So the agent must install the libraries packages and dependencies to run the code. If the code runs successfully, the agent needs to search through all outputs to answer the task questions. So, this is literally a big chunk of AI research. If you had a new hire an AI research lab and they did this incredibly well, you'd be like, "All right, this is probably going to be a good person. They can do all this stuff. They they've got the tech skills.

[00:18:45]They're able to understand what they're doing." It's not the same as novel discoveries, but still huge. In September 2024, the best model we had available then scored 21%. By a year later, in December 2025, that core bench, that benchmark, they they said, "Yep, it's solved." Opus 4.5 got 95.5%.

[00:19:05]So, we went in one year from hopeless, you know, not able to do it to just done, solved. That's it. Kaggle competitions are machine learning competitions online. We've covered some of the stuff on there before. So these models are able to build entire machine learning systems to take part in various gaggle competitions. Our number on this came out early on and 17% was the highest as of a few months ago, February 2026. We now have a 65% let's call it 64.4% score on sort of a a medley of those competitions. So progress is massive.

[00:19:40]Kernel design right. So the one of the hardest tasks in development is kernel optimization. So basically writing code at the foundational level to make it work properly for for what you're trying to do like matrix multiplication or working nicely with specific hardware and these little changes at the foundational level can have these massive multipliers right because if you make something that's a split second faster uses 1% less compute but it's such a foundational thing that it gets used millions or billions of times that can have a massive massive impact. you could effectively double triple your compute your effective compute if you're able to make something you know 100 200% more efficient. So some of the recent works we've seen is for example using deepseeks models to try to build better GPU kernels I mean the conversion of PyTorch modules to CUDA code. So PyTorch was initially it was for various machine learning tasks uh initially developed by Meta. CUDA is Nvidia's code for specifically for their hardware, but it's one of the most kind of common one that that a lot of people know how to use. And we have using LMS to help write kernels for non-standard hardware like Huawei's Ascend chips for example. And he gives many many more. I don't want to go through this because as interesting as they are, I think for people that want to learn more, definitely read this article. I'll link it down below. His point is in a lot of different categories that we have to do as part of AI development, these AI systems are getting very very good at doing it seemingly all of them including AI alignment research. They're also good at these meta skills like management. So managing armies of agents can be done very well by an agent. And here he asks kind of a big question. Is it more like discovering general relativity or is it more like Legos? Is it something where just brute force and putting one foot in front of the other kind of gets you there? Or are these big brilliant breakthroughs needed? Or as Thomas Edison put it, genius is 1% inspiration and 99% perspiration. Maybe these AI systems aren't ready for that kind of inspiration yet, but they can definitely handle the perspiration part of it.

[00:21:50]Meaning they can automate the brick by brick portion of AI research, which is a big big portion of it. So that's an important thing to understand. Even without those creative novel ideas, AI research can still be greatly automated.

[00:22:04]But he's saying if you look at the public data here too, there are tentalizing signs that AI systems may be able to be creative in a way that lets them advance themselves in more impressive ways. By the way, throughout this document, he's talking about the publicly available data, which might be a little bit weird because isn't he one of the main guys over there at a one of the more impressive Frontier AI labs? I think the reason that he went through so much trouble to publish all of the publicly available examples of what he's talking about is because he can't talk about the classified non-publicly available examples that he sees every day. Meaning he's saying this is happening. He probably can't mention all of the evidence that he has for why it's happening. He's saying here's what is publicly available. This is what we can discuss. But if you read between the lines, and this is just my interpretation, but I think that what he's saying is uh he's not uncertain. If you watch that show Billions, they had this like little expression when they had insider information that would say, "I am not uncertain." Right? It meant like I know this because I've got some insider info on this thing. So, we can go ahead and bet big on this. And this is kind of how I'm reading what he's saying. He's saying, "I am not uncertain." he just likely can't say all the stuff that he knows about that that we don't. So, there's a lot more here. I recommend everybody read this, but let's just look at why this matters. What is the implications? What is the point?

[00:23:25]Right? So, up to this point, we we were talking about why this is likely to happen. What is our evidence that this is going to happen? But again, if you've been watching this channel, hopefully, if I've done my job, I've convinced you that this is happening and it's coming soon. And it's uh if you're not a little bit scared, then either you're crazy or you don't fully understand what's happening. It's okay to be excited or or or terrified or whatever, but you got to have like at least a little bit of trepidation and awe for this time in history that we're entering. So he's saying why this matters. The implications of this are profound and under discussed in popular media coverage of AI R&D. Number one is we have to get alignment right. alignment techniques that work today may break under recursive self-improvement as AI systems become much smarter than the people or the systems that supervise them. We've discussed a lot of the research, a lot of it out of anthropic about how to understand and interpret how these models think, how to get them to not lie, not to cheat, not to deceive us. It's still an open problem. We we haven't solved it. AI capabilities are progressing at a much faster rate than our progression of AI alignment, our our ability to understand and control these systems. We might figure it out and I think we will, but it's the question is, you know, what's going to be first massive improvements, RSI, the intelligence explosion or or are sort of being able to quote unquote solve AI alignment? We also talked about how these AI systems might be able to fake alignment. A lot of the research recently we've seen is that these systems have a high degree of situational awareness. When we test them on on certain things where they might be tempted to cheat if we're able to look inside of what they're thinking through their activations or or chain of thoughts, we realize that often times they are aware that it's a test and they're like, well, I better, you know, mark this answer so people don't get suspicious. Similar to how you probably drive extremely well when there's a police officer in your rear view mirror, right? you're in your best behavior.

[00:25:30]That doesn't mean that you're a good and cautious and courteous driver. It just means in that moment you know you're being observed, therefore you do your best. And of course, as AI systems start to contribute more of the foundational research agenda for their own training, we might end up substantially changing the way AI systems get trained and we might not have good intuitions or intellectual foundations for understanding what this means. There's a great quote that somebody posted on Twitter and actually Andre Garpathy talked about in his latest interview.

[00:25:59]Point is like you can outsource your work. You can outsource your research.

[00:26:03]You can outsource your data collection.

[00:26:05]There's one thing you're never or at least right now maybe never will be able to outsource. You will never be able to outsource understand it. In order to understand something you have to look at it and read it and grock it and understand it. You can outsource a lot.

[00:26:22]You can outsource, you know, the the research, the information being brought to you. It can be summarized. It can be explained in a way that's easy to understand. Like you can outsource all of that. You can automate all of that.

[00:26:33]You can't automate understanding. So the fear here, I think, is if these systems are incredibly good at doing all this stuff. We've automated everything, but we can't automate understanding. At some point, we will kind of lose the plot, if you will. We will not be able to understand how AI progress is continuing. And of course everything that AI touches gets a massive productivity multiplier. So there's a lot of things that we have to contend with inequality of access. We are very much limited by compute. And in fact we've seen sort of recently in the news where the government restricted access to mythos because Anthropic wanted to give access to it to a larger amount of firms for them to test it out. the government stepped in citing security concerns but also saying they didn't want the limited compute to affect their ability to interact with mythos and tested. So we're already seeing kind of this from the government priority as to who is able to get access to compute when it's crunch time and this might continue forever like how do we distribute the compute access across everybody that wants it cuz the more useful it gets the more people are going to want access to these models and the compute needed to run them and also he talks about the AMD doll's law for the economy saying that as AI flows into the economy we'll discover places where things break or slow under the increased volume and will have to figure out how to fix those weak links in the chain.

[00:27:57]This may be especially pronounced in areas where you have to reconcile the fastmoving digital world with the slowmoving physical world like drug trials for new medical therapies. If AI is rapidly able to simulate the effects of these drugs, it might discover drugs that could save thousands of lives very very quickly. And we have to answer the question is like do we take the time for humans to review it knowing that some people will die in the meantime or do we accelerate it because it was AI discovered which of course might work great for a while but then on the 10th or 100th time cause a a massive catastrophe and of course the third big point is the formation of capital heavy human light economy. All the above evidence points to the increasing capability of AI systems to autonomy to autonomously run businesses as well.

[00:28:45]Right? So Sam Halman friends have a bet going on as to when they're going to see the first company that's run either with just one person or maybe even zero people. So meaning mainly AI run that's worth a billion plus. Let's say I think that was the number that might be coming sooner than we anticipate. At that point what is needed to make more money? You need some starting capital. You need resources, you need AI, you don't really need humans as much. He's saying this means we should expect for an increasing chunk of the economy to get colonized by new generation of companies which are either capital heavy because they own a lot of computers or OPEX heavy because they spend a lot of money on AI services which they build value on top of and relatively light on labor compared to today's corporations because the marginal value of spending more on AI versus human labor will be constantly growing as a consequence of the sustained capability expansion of the AI systems meaning AI will be better, cheaper, faster than humans. And so dump the humans, get the AI in. That's the way to win and and build your business to to stay competitive. In practice, this will look like the emergence of a machine economy that grows within the larger human economy. By the way, Google, Coinbase, lots of others, they're already building the infrastructure for this machine economy or as they call it the agent economy.

[00:30:02]Though we might expect that over time the machine economy will interact more and more with itself as AI run corporations begin to trade with one another. This will do profoundly weird things to the economy invite all sorts of questions around inequality and redistribution. It may be possible to see the emergence of fully autonomous corporations that are run by AI systems themselves which would exacerbate all the above while also posing many novel governance challenges. Okay, so he's saying that this is coming. He's saying what his chances are. It doesn't matter if he's exactly right or slightly off.

[00:30:36]Who cares, right? I mean, obviously it's going to make a difference, but whether it's 2027 or 2028 or 2029, like even if it's a year or two or three later, then great. We get a little bit more time to prepare, but it's still coming, right?

[00:30:51]If it's coming a little bit slower, it doesn't change the fact that this these are things that we need to think about and prepare for. And as I said at the beginning of this video, I am in the long term very optimistic about AI.

[00:31:03]Novel scientific discoveries, improvements of the human health. We might even find solutions to a lot of or most of the suffering that people go through, whether it's physical or mental or from scarcity of resources. A lot of the reasons why people suffer could I believe in the long term be solved. I am very optimistic in the long run. You know the medium to short term. I am not as optimistic. Humans don't handle change very well, especially when we're in large groups. Often times, it kind of comes down to the lowest common denominator, especially when people think that their safety is being threatened. So as people start losing jobs, as as our ability to rely on, you know, providing our labor in exchange for resources as that's whole thing starts kind of collapsing, how do we make sure that there isn't an increase in violence and crime in people freaking out? We've already seen multiple attacks on Sam Alman. Is it possible in the future anyone that's perceived as pro- AAI has to be very worried about their own safety? I personally would be willing to bet any amount of money that there will be some politicians out there that use this to try to gain more political capital or or more power.

[00:32:17]They'll use it even if it destabilizes things and causes violence is just too good an opportunity to waste, right?

[00:32:25]Saying like AI is going to take all your jobs and kill everybody and if you vote for me, I will protect you and save you from it. It's too good of a line for a politician to pass up. It's also very easy to explain. the best possible solution. It's a lot more nuanced. You have to get people to really understand more nuanced things and and tell them, hey, it might be difficult. There's this transition. Things have to change, but this is the plan and we're kind of going into unknown territory, so we have to all be cool and don't panic and just like work together. And that doesn't often work out. It's kind of like when you're on an airplane and they say, "Buckle in cuz there's going to be turbulence." That's kind of how I perceive it. I think we're still going to land the plane and it'll be fine, but it's going to get bumpy. But whatever the case is, you know, don't panic.

[00:33:10]Don't be afraid. We are very fortunate to be at this moment in time to be alive, to witness this thing coming into the world. We are living through very interesting times. We're not sure yet if that's a curse or a blessing, but yeah, let me know what you all are thinking in the comments. I' I'd love to hear from you. Thank you so much for watching this. Please hit like, subscribe. It helps quite a bit and hopefully it'll help me get this message out to more people to, you know, be prepared to not fall for false narratives. We really need to be at our kind of best, most rational selves throughout the process just to make sure we don't do anything silly. Because if we figure it out and human progress hits this massive upward inflection point, a lot of stuff can be better than we've ever dreamed of. Or, you know, it could go the other way.

[00:34:02]Sometimes you got to roll the hard six, as they say. My name is Wes Roth. Thank you for tuning in. I'll see you in the next