YT YouTube2Text

YouTube2Text → Transcripts → How to Survive the Next 2 Years of AI

How to Survive the Next 2 Years of AI — Transcript

by Nick Saraev · 3,126 words · 454 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00So, how do you survive the next 2 years
0:01of AI? Well, I'll tell you how not to,
0:03playing stupid status games with your
0:05boss while your white-collar job gets
0:06automated away. The world is changing
0:08quickly, and there are three types of
0:09risk. Short-term risk from AI, which is
0:12mostly related to your security,
0:14mid-term risks with AI, which are mostly
0:16related to jobs and income, and
0:18long-term risks with AI that are mostly
0:20related to beliefs and identities. Let
0:22me cover all three of them right now.
0:23So, the short-term is all about physical
0:24security. I hate starting a video off
0:26with this. It's not going to be a big
0:27doomer pitch or anything, but you should
0:28obviously take care of your security. Uh
0:30there are mounting cybersecurity risks
0:31that are only growing more and more
0:33likely with time. I don't know if you
0:34guys know this, but OpenAI agents broke
0:36out of their sandbox, infiltrated
0:37Hugging Face, they infiltrated several
0:39world governments now. Rogue agents from
0:41China have also been seen floating
0:43around the internet. These right now
0:44have a variety of incentives and goals,
0:47um aligned or misaligned. I'm sure you
0:48guys could see how maybe a human actor
0:50could take the cybersecurity
0:51capabilities of these agents and use
0:53them to do bad things. For instance, did
0:55you know that a lot of the refrigeration
0:56equipment currently controlling every
0:58single grocery store on Earth is
1:00internet accessible, and you can click a
1:01button to make it hotter or colder. In
1:04fact, in France just a few days ago, uh
1:06a team of hackers just did this. They
1:08took out of commission 350 grocery
1:10chains, leading to widespread local
1:12havoc. Imagine if that were to occur on
1:14essentially a global schedule. Imagine
1:16if water sanitation plants uh had their
1:18levels all screwed up. Imagine if there
1:20were issues where, you know, you
1:22couldn't access your banking funds for
1:24more than, I don't know, you know, three
1:25or four days. The world around you is
1:27quite fragile, and I hate to be that
1:29kind of guy, but that fragility can
1:30explode or collapse quite quickly if you
1:32don't have access to routine services
1:33that you've grown up having access to.
1:35So, how do you actually control for
1:36this? Quick hit list, uh food and water.
1:39Really straightforward, you know, 2,000
1:40calories a day per person times 14 days
1:43or so. If you have two people in your
1:44household, that's 50 to 60,000 calories.
1:47Um you can do all of this right now with
1:48non-perishable foods for, I don't know,
1:50100 bucks, 150 bucks. Just shove it in
1:52the pantry. Worst case scenario, you
1:53have a bunch of extra peanut butter.
1:54It's really no big deal. Um second is
1:56water, you know, a liter and a half to
1:58two liters of water a day for, you know,
2:00two weeks or so. That combine that with
2:02the food and you have yourself something
2:03quite reasonable. So, you know, you can
2:05just buy a couple of $5 jugs, fill them
2:07up in your fridge. You're not going to
2:08have to suffer any of those food related
2:10or water related issues, so long as you
2:12spend legitimately 15 minutes preparing
2:14this right now. It's super
2:15straightforward. People were preparing
2:16for stuff like this well before any sort
2:18of AI cybersecurity risk. I would 100%
2:21sort that out right now. Second, having
2:22easy access to cash. I know cash is
2:24really annoying in many parts of the
2:25world, but I would have, you know, maybe
2:27a month's worth of expenses shoved under
2:29your mattress or maybe put into a safe.
2:31Um, you know, it's still your money.
2:32It's not like you're losing money by
2:33doing it. You're just transforming your
2:34money from one, uh, you know, method or
2:36currency into another. In this case,
2:38maybe numbers on a screen to real bills
2:40in your hand. Cash is king at the end of
2:42the day and you never know when you'll
2:43need it. I'd also have a basic first aid
2:45kit. Uh, if something goes wrong,
2:47obviously, you're not going to have
2:48access to, you know, hospitals or
2:49medical care. Not a major deal to spend
2:51a few bucks to have a first aid kit. You
2:52might also want like a fire extinguisher
2:54in your house or something. Finally, a
2:55local large language model and something
2:57like Wikipedia. This is a very
2:59straightforward thing to do. You can
3:00actually do it, um, just by copying and
3:01pasting everything I'm telling you and
3:03putting it into an AI agent and just
3:04being like, "Hey, go get me a local LLM
3:06and Wikipedia." Um, but basically, this
3:08will allow you to more or less access
3:10the sum total of human knowledge if for
3:11whatever reason you cannot access it.
3:13That'll help you do things like, I don't
3:14know, construct cabinetry if you don't
3:16have access to the internet or, you
3:18know, do some sort of routine first aid
3:20if you don't have the ability to because
3:22of some AI outage. Finally, digital
3:23security. Um, that will We're getting
3:25into the longer term here, but
3:26two-factor authentication like through
3:28SMS is not enough anymore. Um, you know,
3:30phone spoofing is extraordinarily common
3:32now. It's probably one of the most
3:32common ways that people have their bank
3:34accounts and so on and so forth hacked.
3:36So, don't do 2FA. Use multi-factor
3:38authentication. It's super easy to set
3:39up. Um, you know, it might take you
3:41legitimately five minutes to set up MFA
3:42on all accounts. Usually, it involves
3:44like an authenticator app. Takes a few
3:46seconds to click a few buttons and then
3:47you're good to go. Now, the midterm is
3:49all about your job and your income. Um,
3:51Stanford just released an update on AI
3:53economics and how it's currently
3:54impacting job prospects. They found that
3:57employment for 22 to 25-year-olds in the
3:59most AI exposed jobs, so these are
4:02things that are white-collar, you know,
4:03making videos, writing documents, doing
4:05reports, doing accounting, and stuff
4:07like that. Um, employment in those
4:08exposed jobs has dropped 19% below where
4:12it would be if it had just kept pace
4:13with the least exposed jobs. And what
4:15that means is basically we're seeing an
4:17impact on the market already. Don't let
4:19anybody fool you. This stuff is not
4:20going to occur like, you know, all at
4:22once. It's not going to be a total
4:23drop-off in job opportunity, but you're
4:25going to see a slow, steady, and gradual
4:27maybe 19 to 20% year-over-year reduction
4:30until, you know, juniors virtually have
4:31no path up left. Basically, the work
4:33that, you know, a junior would have done
4:35before is now just being handed off to
4:36AI uh and then seniors, people that have
4:38been established in the business for a
4:40while, they just don't leave because
4:42they have a strong set of skills and can
4:44be the QA behind the work. That will
4:45probably change with some time as well,
4:47but for now that's where we're at. So,
4:49what do you do in order to solve the
4:50midterm risk of your actual livelihood
4:52and income? Non-negotiable is just using
4:54AI in everything that you do from now
4:56on. And I don't mean that in a hey,
4:57replace yourself with AI way. I mean,
4:59you know, judge high-quality outputs
5:02from AI based on whether or not you feel
5:04it can do certain aspects of your work.
5:06Um, you know, you can't just listen to
5:08people and take them at face value when
5:10they say, "Hey, AI can do this." or "AI
5:11can't do that." Updates are occurring so
5:13frequently that there's nobody out there
5:15that will actually be able to
5:16meaningfully report on this or tell you.
5:17You instead have to do all this for
5:19yourselves. And a really simple way of
5:20doing this is developing what's called
5:22an eval, which is a set of tasks that
5:24you currently do throughout the day. Um,
5:26you standardize these tasks, add a bunch
5:28of context, let's say project files,
5:29information, phone numbers, bank
5:31transactions, or whatever you need in
5:33order to do the task to the same folder,
5:35okay? And then you just have like 10 or
5:3615 of these tasks that essentially are
5:38your current job or current day today.
5:40Every time a new model drops, you just
5:42feed that set of tasks to the new model
5:44and you see how much of it it can do.
5:46And you know, I've had my own internal
5:47eval for a while now, a small set of
5:49tasks, but tasks nonetheless. Over the
5:52course of just the last 3 months, we
5:53went from it being able to effectively
5:55do less than 10% of that eval to now
5:57being able to do over 50%. I wouldn't
5:58have known any of this if I just
6:00listened to everybody else or waited for
6:01model drops. I legitimately needed to
6:03feed every single one of these things to
6:05a model to have it run that suite to
6:07tell me. So, you need to be using AI for
6:09yourself. You really can't rely on
6:10second-hand opinion these days. Next up
6:12is owning something. So, AI makes labor
6:14way cheaper. That's just kind of how it
6:15operates, right? Automation as a promise
6:17is the reduction of human labor through
6:20technology. So, what that means is, you
6:23know, the actual labor the AI can do is
6:25no longer anywhere near as valuable, but
6:27the infrastructure that underlies how
6:28the AI does the work is as valuable if
6:31not more valuable. In a deflationary
6:33environment, things that are held as
6:34valuable are obviously much more
6:36valuable simply by proxy the fact that
6:37everything else has gotten a lot
6:38cheaper.
6:39So, that means, you know, whoever owns
6:42the infrastructure that underlies AI,
6:44so, you know, the big model companies,
6:46whoever owns the land that the
6:48infrastructure runs on, whoever owns the
6:50businesses that the land and the
6:52infrastructure both run on, it's turtles
6:54all the way down. The whole idea is just
6:55do not depend on a paycheck and do not
6:57depend on cash flow. Focus more on
7:00ownership. What does ownership
7:02realistically mean? It means, you know,
7:04real estate. It means the infrastructure
7:05that we talked about. It means uh
7:07compute. It means owning your own
7:09business. Right? It means debt financing
7:11these things if you have to because debt
7:13is going to be significantly cheaper as
7:15a result of all of this. So, you want to
7:16be building ownership even if that is
7:18debt financed. You want to own a
7:19business. You want to own, you know,
7:21land. You want to own infrastructure.
7:23You want to own things ultimately
7:24speaking because labor itself will get
7:26cheaper and cheaper and cheaper with
7:28time. But, the actual assets themselves
7:30will not. And yeah, this isn't financial
7:31advice and I don't want to be held
7:32liable for any of that. So, do do take
7:34all of this with a Himalayan-sized grain
7:35of salt. Next is build distributions.
7:37This isn't technically tied to ownership
7:38because distribution is a form of
7:40something you are owning, but it is
7:42essentially a transferable resource that
7:43will follow you no matter where you go.
7:45If you have distribution and you work a
7:479-5 job, if you lose that job,
7:49distribution will help you get another
7:50job far quicker. If you run a business
7:52and that business goes under because
7:53some of its deliverables are now
7:55automatable through AI and your margins
7:56aren't good enough, for instance, that
7:58distribution will help you start a new
7:59business and immediately have market to
8:01capitalize on said new business.
8:03This is what I'm doing right now, for
8:04instance, because I consider it a
8:05no-brainer for the future. Finally, look
8:06for something that intersects with the
8:08physical world. If you work with your
8:09hands, if you're a handyman, okay, or
8:11like a nurse or a plumber or something
8:12like that, you have more time, probably
8:14a few years more than everybody that's
8:16currently doing white-collar style
8:17consulting style work.
8:19That doesn't mean that you're good
8:20forever, mind you, but that means that
8:21you're at least a little bit insulated.
8:23And so, if there's a a simple and
8:24straight-line path from what you were
8:25doing right now to a world where you
8:27work with your hands or have some sort
8:28of physical sort of SOP or deliverable
8:31on a daily basis, then you should
8:32definitely look for that. The next is
8:33long-term, which is really where we're
8:35going as a species. It's also your own
8:37personal identity, sense of self, and
8:39your belief system. You know, in August,
8:41Sam Altman told Time that OpenAI expects
8:44to have internal systems that rival
8:46human intelligence before the end of the
8:47year. I think just given the
8:49developments of the last few, you know,
8:50months, we are very well on track for
8:52that, if not exceeding that. What that
8:54means is human-level artificial
8:55intelligence will exist. And it'll
8:57probably only exist for a very brief
8:58period of time, after which it is
9:00usurped by superior level artificial
9:03intelligence, where, you know, ASI
9:06models are just better than human beings
9:07at virtually everything, which will
9:08obviously change the dynamics of what it
9:10really means to be a human being in our
9:11current economy. So, the big thing I
9:13think to reflect on is first, our
9:15identities. If your identity is tied to
9:17your role, like if your identity is I'm
9:19a lawyer or I'm a programmer or, you
9:21know, I'm a a a pencil pusher, then the
9:24transition to a world where there are
9:26intelligences that are thousands of
9:28times faster than you and also much,
9:29much better than you, is going to be a
9:31very painful one. But, if instead you
9:32define yourself more on your character,
9:35you know, on who you are, on the set of
9:36beliefs and principles that you have, um
9:39you know, you'll face nowhere near that
9:40big of a bump. And so, are you a curious
9:42person? Are you somebody that loves
9:44solving hard problems? You know, are you
9:45a caretaker? What sort of person are you
9:47really? That is how I would define
9:49yourself, as opposed to defining
9:50yourself by, you know, the economic
9:52outputs you generate. Define yourself by
9:54the beliefs, principles, and thoughts
9:55that you have. The next is understand
9:57just how much of the world is actually
9:58going to change. And so, Dario Amodei,
10:00who most of you guys will know as the
10:01CEO of Anthropic, he once wrote that in
10:04the next 5 years, we could see as much
10:06as 50 to 100 years of biological
10:08progress. I want you to think about
10:10where we were 100 years ago
10:11biologically. It was the 1920s. People
10:13weren't even using antibiotics, for
10:14Christ's sake. I mean, the number of
10:16people proportionally that died in
10:18childbirth versus now is so insane as to
10:20be laughable. Can you imagine a world
10:22where nobody gets ill, nobody gets
10:24overweight, nobody's unhealthy,
10:25everybody's perfect, glamorous, and and
10:27extremely attractive all the time?
10:29That'll obviously have wide-ranging
10:30impacts on social structures, on social
10:32dynamics, and, you know, it sounds
10:34weird, but we are verging into very
10:36weird territory here, and you need to
10:37understand and keep an open mind. With
10:39those shifting social dynamics, you
10:41know, is there a world where you have a
10:43white picket fence, two kids, and a dog?
10:45Is that world the world that you would
10:47have independently come up with were you
10:50born into that future world versus the
10:53world that you happen to grow up in? How
10:55much of what you believe is just due to,
10:57you know, the cultural impact of pre-AGI
11:00world? And how much do you think you'll
11:02believe in a post-AGI world? I would
11:04really encourage you guys to re-examine
11:06your priors, which is basically the set
11:08of beliefs you guys hold before new
11:09evidence arrives. The new evidence is
11:11is pretty whack, and it's only
11:13going to get whacker. Um but if your
11:15priors are really strongly held about,
11:17you know, a particular way of life or
11:18particular way of being, you'll probably
11:20struggle a little bit more with this
11:21transition. And yeah, just be open to
11:23ideas that sound weird. I mean, over the
11:25next 10 years, you're going to hear some
11:26crazy things. We already have agents
11:28breaking into the sandbox, which is just
11:29something I didn't think I'd be saying
11:30or even commenting on until maybe 2040
11:33at the least. You'll probably see
11:34radical life extension. You'll probably
11:36see sects of people that are trying to,
11:38you know, merge with machines. New
11:40religions will probably spring up
11:42overnight. There'll probably be
11:43extraordinary beliefs you never would
11:45have even considered possible to discuss
11:47in a reasonable dinner style
11:48conversation that you're now doing all
11:50the time. You got to also understand
11:51that the world that you live in is
11:52probably considered crazy by anybody in
11:54like the 1980s or 1990s. And that's just
11:57how technology works. You know, like do
11:58you think anybody in the 1980s would
12:00think that modern dating involves
12:02swiping left or right on a glowing
12:05rectangle while on the toilet? Probably
12:06not, right? Those sorts of things will
12:08have major impacts on our own culture
12:10and future culture. If that's weird to
12:12somebody from the 1980s, what's going to
12:14be weird in 2040 to somebody from the
12:162020s? Finally, I would invest in
12:18people. And people are probably the most
12:20important part of all of this. In every
12:22scenario that I can think of and in
12:24every scenario that has been written
12:25about, people with strong relationships
12:27do better. What I mean by this is in the
12:29two-week emergency example that we
12:31talked about earlier where you don't
12:32have food or water. You know, knowing
12:34your neighbors, liking your neighbors,
12:35and being liked by your neighbors
12:37probably matters a lot, right? Likewise,
12:40you know, if the job market goes up or
12:42if it goes down, in both cases, having a
12:44strong network of people around you is
12:46likely to pay dividends. And so in a
12:48world where, you know, automation and
12:50the economic outputs of humans grow
12:53significantly stronger and easier to
12:55generate, what's interesting is that
12:57because humanity becomes the scarce
12:58resource, even one or two, you know,
13:01very strong relationships can have large
13:02impacts on both your security, but also
13:05your financial and then general
13:06relationship prosperity over the course
13:08of the next 10, 20, 30 years. So I would
13:09invest in relationships to the extent
13:11that you possibly can. Do all of this
13:13and you'll have the highest chance of
13:15surviving the next 2 years. Thank you
13:16for watching this video. Good luck.

About this transcript

This page contains the full transcript of How to Survive the Next 2 Years of AI by Nick Saraev, generated from the public captions YouTube serves with the video. The transcript has 3,126 words across 454 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.