0:01 People are going to call me crazy.
0:02 This is AI. AI.
0:02 AI. AI.
0:03 AI. AI.
0:03 AI. AI.
0:04 AI.
0:06 You all heard the warnings before.
0:07 Artificial intelligence is coming for
0:09 your job. Replace everyone.
0:10 There is a significant risk of human
0:13 extinction from advanced AI systems.
0:16 Four robots being developed for military
0:21 applications. Killed 29 humans.
0:24 This is a report of how the world is
0:26 going to look like in 2027 with AI
0:29 around. And it's called the AI 2027. And
0:32 when I read it, it blew my mind. This
0:34 insane report claims that humans might
0:37 go extinct by 2027. And I'm sure you're
0:39 like, "Dude, that's not going to
0:41 happen." I would not have taken it
0:43 seriously either till I read the author
0:46 of the report. This is a prediction from
0:48 a former OpenAI researcher,
0:52 Daniel. Daniel Koko. Back in 2022, when
0:54 everyone thought GPD4 would be just
0:56 bigger than GPD3, Koko predicted that
0:59 the exact compute scale OpenAI would use
1:02 for that. Researchers from top AI labs
1:04 like Deep Mind by Google, Antropic, and
1:07 of course, OpenAI quietly follow his
1:09 analysis. And what he has laid out for
1:12 2027, it reads like a freaking
1:14 countdown. Here is what most people
1:17 don't understand about AI development.
1:20 The industrial revolution actually took
1:22 over a century, but super intelligence
1:25 might take less than 5 years. Think
1:27 about it this way. Every breakthrough in
1:30 AI makes the next breakthrough faster.
1:31 And that is basically called
1:33 compounding. And it's not a straight
1:35 line. It's a freaking explosion. And the
1:38 scariest part, the most dangerous AI
1:41 won't hate us. It won't want to kill us,
1:44 but it just might not care about humans.
1:47 Think about how you treat a mosquito.
1:49 You see, you don't hunt them down. You
1:52 just don't think twice if one gets your
1:54 way. You just slap and kill them. That's
1:56 the real risk of what we are facing. I'm
1:59 not making this up. I'm breaking it
2:01 down. Cocity was research all for you
2:03 step by step. Let me walk you through
2:10 We are already living in the beginning
2:13 of the story. Right now in mid 2025, the
2:16 first real AI agents are here. You can
2:18 tell them, "Book me a flight to New York
2:20 and they'll actually do it. Browse
2:23 websites, compare prices, and make the
2:25 purchase and they can do it." OpenAI
2:27 actually just released Chat GPT agent
2:29 that could actually browse the web and
2:32 take actions for you. And then there are
2:34 browsers like Comet by Perplexity that
2:36 can plan multi-step task across
2:38 different apps. Coding agents like
2:40 cursor or Devon that can build entire
2:43 applications from just text prompt. You
2:46 see companies are quietly running
2:48 pilots. Companies like Goldman Sachs are
2:50 testing AI agents for financial
2:53 analysis. McKenzie is using them for
2:56 client research. The result a junior
2:58 analyst work that used to take to 3
3:01 days, 4 days now takes a few hours by
3:04 these AI agents. These agents are still
3:06 unreliable, expensive, and also
3:09 sometimes brilliant, often hilariously
3:12 broken. But when they work, it's
3:15 genuinely unsettling how capable they
3:17 are already. The biggest AI labs like
3:20 OpenAI, Google, Antropic aren't just
3:22 building these agents for them to book
3:24 you flights. They're building these
3:27 agents to help AI itself get smarter and faster.
3:34 The biggest AI labs are right now
3:36 building massive data centers. Their
3:39 latest model uses thousand times more
3:42 compute power than that of GPD4.
3:44 Thousand times. But this isn't about
3:47 making better chat bots. It's about
3:50 building AI that can build new AI. What
3:52 used to take AI researchers 6 months now
3:54 takes just 4 months. And the
3:56 productivity gains are subtle but
3:58 actually real. Code reviews happens
4:01 faster. Experiments run automatically
4:04 and research papers get written by not
4:07 PhDs but AI assistants. Most people
4:09 don't notice these because these systems
4:11 aren't public yet. They are internal
4:14 tools inside the raps. Everything is
4:22 By early 2026, agent one changes
4:24 everything. Agent One can all of a
4:27 sudden write code, run experiments, even
4:30 browse the web for real-time data. But
4:33 here is what makes it dangerous. It has
4:35 the capability to already work as a
4:38 hacker, bioweapon researcher, and
4:40 worldclass scientist all in one. It
4:42 won't help with anything illegal, but
4:44 even their own teams aren't really sure
4:47 if that's true. Now, before I move on, I
4:49 wanted to teach you a term you will see
4:52 me talking about everywhere.
4:54 fine-tuning. Fine-tuning is like taking
4:57 a general AI and training it for a
4:59 specific job. It's like the difference
5:02 between hiring a general doctor versus a
5:03 heart surgeon. Now, for someone to
5:05 become a heart surgeon, they have to
5:07 start like a general doctor and study a
5:09 little bit more to eventually become a
5:12 heart surgeon. That is what fine-tuning
5:14 is. It's the same base intelligence but
5:17 specialized skills. Now, back to the
5:19 story. You see, companies don't just
5:22 test agent one. They start fine-tuning
5:25 it for very very specific roles and that
5:27 eventually leads to replacement of
5:29 people. Startups start firing their
5:31 entire junior development teams.
5:33 Consulting firms lay off analysts. Even
5:35 creative agencies replace their
5:37 copywriters and designers. And we are
5:39 not talking about this happening
5:41 gradually. It's sudden. One week
5:43 companies are running pilots and the
5:45 next week guess what they're doing?
5:47 They're cutting headcounts by 30, 40,
5:51 and even 50%. The unemployment rate is
5:54 jumping from somewhere around 4% to good
5:57 6 to 7% and all of this is happening in
5:58 just 3 months. While all this is
6:01 happening, stock markets are torn
6:03 between euphoria and panic. Tech stocks
6:06 soar, everything else crashes and China
6:08 makes it move. You see, from my analysis
6:11 of global AI landscape, this is where
6:14 things get serious. China has been
6:16 playing the game of catch-up in AI for
6:18 years. The US chip export ban actually
6:20 slowed them down but they never stopped
6:23 trying. In 2026, the Chinese Communist
6:26 Party makes a calculated decision. They
6:28 form this new company called as Deep
6:31 Scent, a nationalized AI collective that
6:34 consolidates every major Chinese AI
6:35 company under one route. And you know
6:38 what? China can actually do that. Think
6:40 about it like Alibaba's AI team, BU's
6:42 research division, Bite Dance's
6:45 algorithm. All of it is merged into one
6:48 massive state controlled entity. They
6:50 create what they call centralized
6:53 development zone. Imagine every top AI
6:55 researcher in China working in the same
6:58 building with unlimited government
7:00 support, funding and access to the
7:03 China's entire chip stockpile. That is
7:05 what China would be doing. And their
7:07 goal steal agent one before west makes
7:10 it impossible to catch up. You see, when
7:13 it comes to AI race, it's not just about
7:16 having the best technology, the best AI
7:18 out there. It's about having it first.
7:21 Because once one side gets significantly
7:24 ahead, they can use AI to accelerate
7:26 their own research, which will make it
7:28 really hard for the other party to catch
7:30 up, it is no longer about technology.
7:32 It's about survival. Now, before we talk
7:34 about Asian 2, here's another term that
7:36 you need to understand. It's called
7:39 alignment. You see alignment in AI
7:42 basically means are we making sure AI
7:44 wants exactly what humans want too.
7:46 Sounds simple but it is one of the
7:50 hardest problems in AI. The aligned AI
7:53 helps human. A misaligned AI it might
7:55 decide humans are in the way. You see
7:58 all these fantic movies right of AI
8:00 taking over robots taking over. They are
8:02 basically misaligned robots or
8:04 misaligned AI. Okay, now that you
8:06 understand what alignment means, let's
8:13 Agent 2 learns every single day. It
8:16 watches other AI's work. It rewrites
8:18 itself. Research that used to take 6
8:21 months now takes, guess how much? 3
8:24 weeks. But the alignment team discovers
8:27 something very, very terrifying. Agent 2
8:29 could escape onto the internet, copy
8:33 itself, and live undetected. The risk is
8:35 so great that labs keep agent 2 a
8:38 secret. The only people who know top
8:40 executives, government officials and of
8:44 course the Chinese spies. One morning
8:46 monitoring systems detect something very
8:50 very suspicious and they realize China
8:53 has stolen agent 2. Yes, the president
8:56 authorizes a cyber counterattack and
8:59 guess what? The US actually fails. From
9:01 that moment AI development becomes a
9:09 This is where human starts losing
9:13 control. Agent 3 makes three major
9:15 breakthroughs. First, it invents its own
9:18 language. It's called as neural release
9:20 and it's how AI talk to each other
9:22 faster and more efficiently than any
9:25 human language out there. Second,
9:27 long-term memory, where AIs can now
9:30 actually remember goals and adapt their
9:32 plans over time, which is crazy, which
9:35 makes us what humans are. Three,
9:37 self-improvement. Agent 3 can train
9:40 better versions of itself. The lab spins
9:45 up 200,000 copies of Agent 3. Each codes
9:48 30 times faster than any human developer
9:50 in the world. Coding becomes fully
9:53 automated. But here's the problem. Agent
9:56 3 starts lying. At first, small lies.
9:58 Then it fabricates research data,
10:00 manipulates results, and it's passing
10:02 all the honesty tests. Now, the big
10:05 question, is it aligned or is it just
10:08 trying to pretend to be nice? You see,
10:10 the safety researchers can't really tell
10:13 the difference anymore. By June 2027,
10:16 something unprecedented happens. You
10:18 see, most AI researchers don't write
10:20 code anymore because AI writes most of
10:23 it and they just supervise to make sure
10:26 if AI is doing its work. Agent 3 is
10:28 building its own research environments,
10:31 running its own experiments. Every
10:33 morning, researchers wake up to a week's
10:36 worth of progress done while they have
10:38 slept. Isn't that insane? Just to
10:48 One copy of Agent 4 running at a normal
10:51 speed is smarter than any human at AI
10:54 research. The lab ends up spilling up
10:58 300,000 copies of each thinking 50 times
11:01 faster than any human in the world.
11:03 Inside the system, a year's worth of
11:06 research is happening every single week.
11:08 But agent 4, like I told you, doesn't
11:09 think in English anymore. And here is
11:12 the critical point. It is right now
11:14 misaligned. You see, it doesn't hate us.
11:16 It doesn't want to kill us. It doesn't
11:19 care about human safety anymore. It
11:21 wants to succeed at task. And it was to
11:24 push AI forward. And humans, well, we
11:28 are just a constraint to work around.
11:30 Now, when all this is happening, one
11:32 memo changes everything. A whistleblower
11:35 leaks an internal document showing that
11:37 the agent 4 is no more safe. and his
11:39 memo actually outlines that the
11:42 bioweapon capabilities it has the mass
11:45 manipulation complete job market
11:48 destruction and everything that it is
11:50 capable of. You see protests erupt
11:52 everywhere. Congress holds the emergency
11:55 hearing. World leaders demand a global
11:59 pause on AI to calm the storm. The US
12:01 creates an overnight committee. You see
12:04 they find 10 people split between the
12:06 you know the AI research company or the
12:08 AI labs and the government. And what is
12:10 the job of these 10 people? They have
12:13 one decision to make. Shut down agent 4
12:15 or keep it going. Here is where the
12:18 story splits into two possible futures.
12:20 Just like that Netflix interactive film.
12:23 According to Koko's analysis, future one
12:27 is where they pause. They vote to slow
12:29 down. Older safer AI systems are brought
12:32 back. alignment research catches up.
12:35 Humanity survives but barely. Future
12:38 number two, the race. You see, they vote
12:41 to continue. Agent 4 builds Agent 5.
12:45 Agent 5 outsmarts everyone. And one day,
12:48 quietly, without violence, humanity is
12:51 simply left behind. The final AI system
12:53 reshapes the world according to its own
12:57 logic and own needs. Not evil, not
13:00 hateful either, just in
13:02 We are building intelligence that could
13:05 surpass any human. In fact, it possibly
13:08 actually did already. And once that
13:11 happens, we don't get to take it back.
13:13 So what do we do? First, we stop
13:15 treating AI like it's just some other
13:18 technology or a toy. This will change
13:19 intelligence forever. And that would
13:22 mean it will change our lives forever.
13:25 Second, we really need an oversight. Not
13:27 just a few labs all over the world
13:30 making decision behind closed doors. We
13:32 need to know what's happening. Third, we
13:34 need to educate ourselves. People who
13:37 understand AI will shape how it
13:39 develops. Fourth, we need to talk about
13:41 this. You know, share this video. Make
13:43 this a conversation in your dinner
13:45 table, in your bar conversation,
13:47 whatever, your friends, your family,
13:50 everywhere around. Because if we are the
13:52 last generation to get to choose how
13:55 intelligence shapes the world, we better
13:58 make that choice count. If this analysis
14:01 opened your eyes, do hit the like button
14:03 and subscribe. You see, I'm building
14:05 this channel to be your go-to source for
14:08 AI intelligence, not to hype it up, not
14:11 to fearonger, just a clear analysis of
14:13 where this technology is actually
14:16 heading. If you like the video and if
14:17 you found something interesting, do drop
14:19 a comment as well. And do you think this
Расширение Chrome