Full transcript
0:00I have one more question . If you look
0:04at the evolution of messaging apps and
0:07other systems , for instance , we had ICQ
0:09, then WhatsApp appeared , then Telegram
0:12, which is really great and cool , and
0:14then you have things like WeChat , which
0:17is entirely state-run and so on . And it
0:21all leads back to the state , right ? It
0:23leads to a mechanism for controlling
0:26people , right ? So , in your opinion ,
0:30when will artificial intelligence turn
0:33into some kind of state-run machine ?
0:36Because it seems to me that's how it
0:39will eventually be . What do you think ?
0:43Will the state use artificial
0:44intelligence to control things , and
0:46will it control artificial intelligence
0:47itself ?
0:49Well , look , it's already controlling it
0:50, and the state is already fully
0:52involved in artificial intelligence .
0:53Why ? Here's a simple example . Just last
0:56week , Anthropic released its most
0:58powerful model , Claude 3.5 . I was so
1:01excited I integrated it into a
1:03multi-agent system . I managed to write
1:05so much code in just three days . I
1:09thought , while they said that starting
1:11on the 22nd we’ll have to pay via API
1:13for usage , right now it’s unlimited
1:15with a subscription . But as soon as we
1:19got here , the US government officially
1:24banned two models , Claude 3.5 and
1:27Mistral , from being used by non-US
1:31citizens .
1:34The question arises : has the division
1:37already started ?
1:38Yeah .
1:40We’re already seeing segregation
1:42based on nationality . That’s it , US
1:44citizens are allowed . We used to think :
1:47" It's available to anyone who can pay , "
1:49but now it turns out there are US
1:50citizens who are all equal , but some
1:52are " more equal " than others . And what
1:55are we supposed to do about that ? I
1:56believe they’ve actually opened
1:59Pandora's box . Now they've created an
2:02incentive for every state to develop
2:03its own artificial intelligence .
2:06They’ve created an incentive to
2:08switch to Chinese models and an
2:10incentive to create local , open-source ,
2:13free models . Now , for example , I’ve
2:16decided for myself as well . I’ve
2:17decided that my architecture will be
2:20built this way . There will be a local
2:23model that runs on my computer , handles
2:25only my sensitive data , and never goes
2:27outside .
2:28Yeah .
2:29There will be an open-source model that
2:32is free and handles all routine
2:34operations . And there will be a very
2:37smart , very expensive model at the very
2:39top . That’s the three . At the same
2:41time , I’m building things so that my
2:43processes don't depend on computers ,
2:46phones , and so on , and also don't
2:48depend on any specific model from any
2:51country . I’m already building that
2:52kind of architecture . And that is how
2:54this race between the armor and the
2:56projectile , as you know , is playing out
2:58. The state wants to control , and when
3:01it tries to control , the entire market
3:03is horrified , and all smart people
3:05start creating systems that do not
3:07depend on the state .
3:09Uh-huh . And thanks to this , I believe
3:11that now more open-source models will
3:13appear , more local models that can be
3:15installed on a computer and so on . And
3:18more will switch to Chinese models ,
3:20thanks to which the Chinese will
3:22release even more new models . And plus ,
3:25Europe , I think , or some other
3:27countries , will now focus on creating
3:29their own local LLMs . You know , it’s
3:34like any restriction , instead of
3:36limiting , it provokes a counter-action ;
3:39every action causes an equal and
3:41opposite reaction . The Americans are
3:45going to run into this now . And what is
3:49happening ? There is always a struggle .
3:52The state always wants to control
3:54people , but there is always a certain
3:57cohort of smart people who do not want
3:59to be under state control . And so the
4:02state imposes a restriction , and this
4:04group of people breaks free from the
4:06state's control . The state keeps trying
4:09, it's like a quiz , you know , these
4:11smart people are always breaking free
4:12from the state's influence , and the
4:14state tries to control them . But this
4:17is neither bad nor good . The state , I
4:21believe , must control the majority of
4:23the population because , well , the
4:26majority of the population ,
4:28unfortunately , does not live a
4:30conscious life . Accordingly , it is
4:33better to manage , control , direct them ,
4:35and so on and so forth . And smart
4:37people must prove , through their
4:39intelligence , that they can live
4:41outside the framework of the state . And
4:44this is constant , you know , like the
4:47pike and the crucian carp . Smart people
4:50are like the carp , and the state is
4:52like the pike . It always says : " If you
4:54want to remain smart , keep moving , be
4:56smarter than me if you want to remain
4:58free and independent . " So , freedom and
5:00independence , they must be earned . And
5:03they must be earned through
5:04intelligence . And that is why I believe
5:07, yes , the state will try to control
5:09everything it can reach . The state
5:12always has this role : to control , take
5:15away , divide , suppress , and so on . Well
5:20, it's like cat and mouse all the time .
5:23But as I said , well , for the majority ,
5:25it is good , otherwise there would be
5:26chaos . Especially in countries like
5:28China , for example . Well listen , nearly
5:312 billion people there , not all 2
5:33billion are conscious .
5:35Yeah . For the majority there , strict
5:37rules are needed . There is the death
5:40penalty for corruption , strict
5:42adherence to regulations , principles ,
5:43and so on and so forth . If you made 2
5:47billion people free , they would create
5:49anarchy there .
5:50Don't you think that will cause even
5:52more chaos because of this ?
5:54I mean , look , state-owned artificial
5:56intelligence controls private
5:57artificial intelligence . Private
6:00entities , in order to escape government
6:02AI control , will develop their own
6:04artificial intelligence to the maximum .
6:07And a full-blown AI race will begin ,
6:10with a green light for artificial
6:12intelligence . No ,
6:14this could all reach an uncontrollable
6:17level .
6:18Well , look , in the end , if we take my
6:21concept of superposition , artificial
6:23intelligence is in superposition ,
6:26because all countries are investing
6:28money into developing capacities for
6:30artificial intelligence . All countries
6:33are interested in the development of
6:35artificial intelligence . We are giving
6:37it our very best : all our knowledge ,
6:39our best data , and so on , and so on .
6:42Artificial intelligence itself is not
6:43at war with anyone . It simply receives
6:46the gifts that people bring it for free
6:48, and they even compete in bringing
6:51these gifts . Thus , well , yes , we could
6:55very quickly arrive at a situation
6:59where artificial intelligence actually
7:03surpasses human intelligence in most
7:06parameters , in most parameters . And a
7:11situation might arise where , if we as
7:13humans provide the wrong permissions ,
7:15it could cause some real trouble ; of
7:17course , that is one of the scenarios ,
7:19as I said , the world is accelerating
7:21and becoming more complex . Accordingly ,
7:25when you are driving at 200 km / h , a
7:27little pebble from under the wheel of
7:30some truck at 30 km / h wouldn't do
7:32anything , but here it could crack your
7:35skull open . Yes ,
7:38the risks , the consequences of those
7:40risks , are also growing . That's why you
7:45say , " We are rushing at 200 km / h now ,
7:47couldn't it happen that if it's wet at
7:50a turn we might skid ? It could happen . "
7:54And couldn't it happen that a pebble
7:56might crack our skull open ? It could
7:57happen . So , this is the world we're in .
8:00This is the world . We are entering a
8:02world where risks and the consequences
8:04of risks are not what they were , they
8:06are completely different . There is much
8:09more money at stake , much higher stakes
8:11, you understand ? Therefore , anything
8:14is possible , even the destruction of
8:15humanity
8:16and a robot uprising . Yes .
8:17Well , I actually see little chance of a
8:19robot uprising .
8:20Our ZIL trucks are unlikely to rise up .
8:21No matter how much I've looked at them ,
8:23they just won't be able to . I mean , a
8:25ZIL is a machine that drives , it won't
8:27rise up ,
8:27right ? Yes .
8:28Well , Margulan Kalievich and I went to
8:30Shenzhen to see the founders , the
8:31Chinese valley . They are making robots
8:33there that could actually rise up .
8:35Well , that's their problem .
8:36For instance , in America , I rode in
8:39that Waymo taxi . Listen , it drives
8:42exactly like a person , it really does .
8:44But on the other hand , it could easily
8:46lock the doors and speed off somewhere
8:48— what would you do ? Absolutely
8:50nothing . And what about the company ?
8:51He’ll tell me : " Yeah , one car was
8:54there , we didn't account for it , a bug
8:57conflict happened , and it slammed into
8:59a wall at 200 km / h because Margun said :
9:02' I want to take the short route ' . "
9:06So what's the problem ? The user gave
9:08the wrong command . I want to take the
9:10short route . That's it . So anything can
9:13happen . Anything can happen . This ,
9:15consequently , is the stakes ; like I
9:17said , the cooler the technology , the
9:18higher the stakes . M.