MASSIVE AI News : AGI Secret UNLOCKED! o3, GPT5, OpenAI Moves Faster And Stunning Humanoid Robots

00:30:06
https://www.youtube.com/watch?v=LhxGCDgoy2U

摘要

TLDRThe video delves into the latest news regarding AI and robotics, particularly highlighting OpenAI's ongoing journey towards AGI, speculation about upcoming innovative models, and the ramifications of AI on employment and society. Key discussions include the emergence of AI capable of aiding in invention, updates on the capabilities and design of humanoid robots, debates surrounding the economic impacts of AI, and concerns over the use of AI in military contexts. Moreover, advancements like program synthesis are proposed as significant steps towards more generalizable AI learning.

心得

  • 🚀 The potential for AI to aid in inventions is seen as a game-changer.
  • 🤖 Humanoid robots are increasingly advanced and human-like.
  • 💼 AI may replace numerous jobs but could also create new roles.
  • 🔍 Program synthesis allows AI to learn with less data.
  • ⚔️ The use of AI in warfare raises significant ethical concerns.

时间轴

  • 00:00:00 - 00:05:00

    The discussion begins with speculation about AI advancements, particularly focusing on OpenAI and the notion of AGI (Artificial General Intelligence). The narrator expresses cautious excitement about the potential for AI to aid in invention and innovation, referencing the internal progress at OpenAI and the timeline of AI developments that might lead to significant changes in the industry.

  • 00:05:00 - 00:10:00

    The conversation shifts to the recent advances in AI image generation and the mention of new models that outperform existing ones. The narrator introduces 'Recraft,' an image-generation tool with outstanding capabilities, while also hinting at OpenAI's upcoming model, GPT-3.5, which is anticipated to be a substantial update in AI intelligence and capabilities.

  • 00:10:00 - 00:15:00

    Continued speculation surrounds the impact of AI on various sectors, especially in terms of job loss. Major firms like Salesforce hint at reducing hiring due to AI's productivity gains. The narrator discusses how AI tools are enhancing productivity but warns of the potential job displacement in lower-tier roles and emphasizes the necessity for humans to adapt to these changes.

  • 00:15:00 - 00:20:00

    The potential of robotics and humanoid robots is highlighted, showcasing improvements in robot capabilities and efficiency. The narrator shares insights on the rapid evolution of humanoid robots and discusses the implications of having robots that operate almost indistinguishably from humans, leading to ethical and societal considerations of such technology.

  • 00:20:00 - 00:25:00

    Further discussion reveals a focus on creating AI with a true understanding and adaptability, portraying a shift from traditional learning methods to a new way of assimilating knowledge that mimics human learning. The idea of 'program synthesis' emerges as a promising method that allows for greater efficiency in learning and invention.

  • 00:25:00 - 00:30:06

    Finally, the narrator touches on the looming concerns of AI in warfare and national security. The conversations indicate a growing realization that as AI becomes more advanced, it may be seen as a national asset, prompting countries to secure their technology and possibly view AI labs as military targets, thus highlighting the profound implications of AI on global strategy and security.

显示更多

思维导图

视频问答

  • What is the significance of the tweet about AI innovations?

    The tweet suggests potential advancements in AI that could aid in invention, hinting at OpenAI's future goals.

  • What does AGI stand for?

    AGI stands for Artificial General Intelligence, which refers to AI systems that can understand or learn any intellectual task that a human being can.

  • What is the impact of AI on jobs according to the video?

    The video discusses concerns that AI may replace many jobs, especially in software engineering and banking, but also suggests that AI could lead to new job roles and increased efficiency.

  • How does program synthesis relate to AI learning?

    Program synthesis enables AI to learn by understanding rules rather than needing vast amounts of data, allowing it to generalize better.

  • What is the concern about AI in warfare?

    There are discussions about AI becoming a national asset and potentially being used in military applications, leading to data centers being targets for warfare.

查看更多视频摘要

即时访问由人工智能支持的免费 YouTube 视频摘要！
字幕
en
自动滚动:
  • 00:00:00
    so one of the first pieces of AI news
  • 00:00:01
    cuzz I really just want to get stuck
  • 00:00:02
    right into this is essentially this
  • 00:00:04
    piece of news right here now I know that
  • 00:00:06
    this tweet is essentially a large part
  • 00:00:09
    speculation but I can't help but think
  • 00:00:11
    that the timing of this tweet is quite
  • 00:00:13
    specific considering we've heard time
  • 00:00:16
    and time again many many Sparks of ASI
  • 00:00:19
    from open AI employees so basically and
  • 00:00:22
    I'm going to you know link this to
  • 00:00:23
    another point cuz I was supposed to do
  • 00:00:25
    that in another video but I actually did
  • 00:00:26
    manage to forget but it talks about how
  • 00:00:29
    they got you know some readin to what
  • 00:00:31
    internally is happening at opening ey
  • 00:00:33
    and holy mother of God I don't know how
  • 00:00:35
    to express my feelings without sounding
  • 00:00:37
    like hype um I don't know what to say
  • 00:00:39
    but I'll share this the innovators are
  • 00:00:41
    coming the problem is we don't know how
  • 00:00:43
    they got here now that essentially
  • 00:00:44
    refers to the levels to AGI most people
  • 00:00:47
    actually do forget that this is the
  • 00:00:49
    entire road map for open AI in terms of
  • 00:00:51
    what they're setting out to accomplish I
  • 00:00:53
    think it's really important to you know
  • 00:00:55
    know where the company's headed so we
  • 00:00:56
    can actually understand where the
  • 00:00:57
    industry is going as a whole but
  • 00:00:59
    essentially level four is the innovators
  • 00:01:01
    and that's of course AI that can Aid in
  • 00:01:04
    invention so this is going to be
  • 00:01:05
    something that's rather you know
  • 00:01:06
    fascinating because this is of course AI
  • 00:01:09
    that does change the game um and it's
  • 00:01:11
    going to be pretty interesting because
  • 00:01:12
    of course you know right now this is the
  • 00:01:13
    year supposedly of AI agents I've heard
  • 00:01:16
    varying various things from different
  • 00:01:17
    Laboratories but um yeah I mean I mean
  • 00:01:20
    it's super super interesting that now
  • 00:01:21
    this is the innovators now I'm a little
  • 00:01:23
    bit skeptical I do have to be honest
  • 00:01:24
    because AI that can Aid an invention I
  • 00:01:27
    mean you know at the first glance you
  • 00:01:28
    would have to be skeptical but but we
  • 00:01:30
    have seen inklings of this when we
  • 00:01:32
    actually think about how AI systems have
  • 00:01:34
    been developed in the past when we look
  • 00:01:36
    at things like Alpha fold and stuff like
  • 00:01:37
    that when we think about how it manages
  • 00:01:39
    to make stuff I mean I guess you could
  • 00:01:41
    say invention is basically just search
  • 00:01:44
    and just testing stuff out so I mean
  • 00:01:46
    potentially I don't think that would be
  • 00:01:48
    that hard considering the systems before
  • 00:01:50
    this the reasoners were actually built
  • 00:01:52
    on a SE search based architecture so
  • 00:01:54
    maybe that could be there but I don't
  • 00:01:56
    know it's pretty crazy tweet honestly
  • 00:01:57
    now the Tweet goes on to say that this
  • 00:01:59
    is not a it's not 03 it's not 40 it's
  • 00:02:01
    not gbt 5 but that it could be an
  • 00:02:03
    iteration or new version of an existing
  • 00:02:05
    thing and that's you know actually
  • 00:02:06
    making me think that it is somehow
  • 00:02:08
    similar to Alpha fold in a sense that
  • 00:02:10
    it's able to create new things um and
  • 00:02:12
    this says you know being vague and
  • 00:02:13
    sincerely simultaneously is very
  • 00:02:15
    difficult for y'all y y y but I think I
  • 00:02:17
    think you know this is like somewhat
  • 00:02:19
    true because now that I'm thinking about
  • 00:02:20
    something I do remember that samman
  • 00:02:21
    actually said something rather specific
  • 00:02:23
    he actually did say that um one of the
  • 00:02:25
    things that is going to be coming in the
  • 00:02:26
    new year is something that a lot of
  • 00:02:28
    people didn't think about and that is
  • 00:02:30
    something that I do think is along the
  • 00:02:32
    lines of using generative models to
  • 00:02:33
    potentially invent certain things and I
  • 00:02:35
    think that potentially open a ey working
  • 00:02:37
    on that in secret so if I had to guess
  • 00:02:39
    like if I had to put my best guess
  • 00:02:40
    forward I would say that you know this
  • 00:02:42
    is potentially something that is along
  • 00:02:44
    the lines of something that uses a
  • 00:02:45
    generative model to potentially use
  • 00:02:47
    search maybe to come up with new
  • 00:02:49
    Solutions or new existing potential I
  • 00:02:51
    don't know ideas maybe I mean there's a
  • 00:02:54
    lot of stuff going on there is actually
  • 00:02:55
    a new Microsoft model that you know does
  • 00:02:57
    something along those lines we've seen
  • 00:02:59
    the iterations of Al aold actually I'm
  • 00:03:01
    going to be covering something later on
  • 00:03:02
    in the video that is like that so it
  • 00:03:03
    wouldn't be surprising if open AI kind
  • 00:03:05
    of had that kind of model I don't think
  • 00:03:07
    something like that would ever be
  • 00:03:08
    released to the public but it definitely
  • 00:03:10
    is something that is super interesting
  • 00:03:12
    now I think I do have also something
  • 00:03:15
    that I want to share with you guys as
  • 00:03:16
    well there was a really weird tweet that
  • 00:03:18
    I saw but it was you know super
  • 00:03:20
    speculative that I'm going to show you
  • 00:03:21
    guys in a moment but amongst the AI hype
  • 00:03:23
    I did want to talk about the fact that
  • 00:03:25
    you know there is of course some truth
  • 00:03:27
    in certain statements so noan Brown
  • 00:03:29
    person who you know worked on reasoning
  • 00:03:31
    an open AI actually said lots of vague
  • 00:03:34
    AI hype on social media these days there
  • 00:03:36
    are good reasons to be optimistic about
  • 00:03:38
    further progress but plenty of unsolved
  • 00:03:40
    problems you know in research remain so
  • 00:03:43
    this is a good tweet and a lot of people
  • 00:03:45
    were appraising this tweet because
  • 00:03:46
    whilst right now you have pretty much
  • 00:03:48
    not every open AI employee but a vast
  • 00:03:50
    number of you know open AI employees
  • 00:03:52
    talking about how they've maybe achieved
  • 00:03:54
    AGI or how maybe they've you know
  • 00:03:56
    figured out exactly how to achieve ASI
  • 00:03:59
    no I'm the one you know really
  • 00:04:01
    spearheading the reasoning of open AI is
  • 00:04:03
    actually talking about you know there's
  • 00:04:04
    vague AI hype but of course we're
  • 00:04:06
    optimistic but there are still a lot of
  • 00:04:08
    Unsolved research problems that still
  • 00:04:10
    remain in the space so it's definitely a
  • 00:04:13
    sign that whilst yes things are moving
  • 00:04:14
    ahead of course there is still realism
  • 00:04:17
    that people are encountering he says
  • 00:04:19
    that between the 01 announcement the 03
  • 00:04:21
    announcement and various podcast talks I
  • 00:04:24
    think we've said a lot we believe 01
  • 00:04:26
    presents a new scaling Paradigm and
  • 00:04:27
    we're still early in scaling along that
  • 00:04:29
    Di mention it's definitely very true the
  • 00:04:31
    scaling Paradigm that the 01 series
  • 00:04:33
    represent I think it's more significant
  • 00:04:35
    than people are giving credit for
  • 00:04:37
    because I personally think that this is
  • 00:04:38
    going to truly lead to some superhuman
  • 00:04:40
    performance in certain areas quicker
  • 00:04:42
    than most people do realize and that's
  • 00:04:44
    just based on the passings we've seen
  • 00:04:46
    with other prior systems that are pretty
  • 00:04:48
    much neuros symbolic that have achieved
  • 00:04:50
    really amazing things so before we dive
  • 00:04:52
    into any more AI related developments
  • 00:04:54
    let me actually tell you about something
  • 00:04:56
    that is revolutionizing the AI art bace
  • 00:04:58
    so have you heard about recraft it's
  • 00:05:01
    actually the new king of image
  • 00:05:03
    generation and I'm actually not
  • 00:05:04
    exaggerating here their V3 model is
  • 00:05:07
    currently at the number one spot on
  • 00:05:09
    hugging fa text to image Benchmark at
  • 00:05:11
    performing giants like mid journey and
  • 00:05:13
    opening eye with an impressive
  • 00:05:16
    1,139 ELO rating you might be thinking
  • 00:05:18
    well what sets recraft apart well we're
  • 00:05:20
    actually talking about photorealistic
  • 00:05:22
    images that will make you do a double
  • 00:05:24
    tape perfect text generation for all
  • 00:05:27
    your design needs
  • 00:05:30
    and the most anatomically accurate AI
  • 00:05:32
    model out
  • 00:05:33
    there over 2 million creators including
  • 00:05:36
    teams at Netflix and Airbus already
  • 00:05:38
    using it and here's the best part I've
  • 00:05:40
    actually got a special deal for my
  • 00:05:42
    viewers use my promo code to get $12 off
  • 00:05:45
    any plan you can do that by using the
  • 00:05:47
    code grid 12 whether you're a casual
  • 00:05:49
    Creator or a professional designer
  • 00:05:51
    they've got options ranging from 12 to
  • 00:05:53
    $48 per month for unlimited access check
  • 00:05:56
    out the link in the description to try
  • 00:05:57
    the Revolutionary V3 model and of course
  • 00:05:59
    if you don't want to check out the link
  • 00:06:00
    in the description you can simply scan
  • 00:06:01
    the QR code on screen now to get my
  • 00:06:03
    exclusive discount and for all you
  • 00:06:05
    developers out there they've got a
  • 00:06:06
    powerful rest API that lets you
  • 00:06:08
    integrate both roster and Vector outputs
  • 00:06:10
    directly into your projects it's super
  • 00:06:12
    straightforward to set up just grab your
  • 00:06:14
    API key and you can start generating
  • 00:06:16
    images programmatically in minutes now
  • 00:06:18
    Sam outman actually gave us some
  • 00:06:20
    information on 03 this is the highly
  • 00:06:22
    anticipated model that is currently
  • 00:06:24
    stateof thee art across a variety of
  • 00:06:26
    different tasks you can see here that
  • 00:06:28
    one person commented I want a model
  • 00:06:30
    smarter than 01 Pro I'm willing to pay
  • 00:06:32
    I'm using it all the time for coding and
  • 00:06:34
    it would be awesome if it could generate
  • 00:06:36
    tens of thousands of lines of code at a
  • 00:06:37
    time it can do a decent job generating
  • 00:06:39
    only about 1,500 lines now based on
  • 00:06:43
    experience and he does say that 03 is
  • 00:06:45
    now much smarter and we are turning our
  • 00:06:47
    attention to that now and 03 Pro with an
  • 00:06:51
    explosion Emoji so I think this is just
  • 00:06:53
    going to show us you know that 03 is
  • 00:06:55
    probably a really capable model and
  • 00:06:58
    honestly them releasing it around the
  • 00:07:00
    February Mark is going to be super
  • 00:07:01
    interesting because I personally didn't
  • 00:07:03
    think that that was going to be the case
  • 00:07:05
    I thought it would be a much longer time
  • 00:07:07
    considering the fact that 03 is going to
  • 00:07:09
    be really expensive and from what I'm
  • 00:07:11
    hearing that they're still losing money
  • 00:07:13
    on 01 Pro mode so that inference cost at
  • 00:07:17
    the moment however doing it um is
  • 00:07:19
    probably not that efficient so I'm
  • 00:07:21
    surprised they're releasing 03 as
  • 00:07:22
    quickly as they are but who knows maybe
  • 00:07:24
    they're not going to release it straight
  • 00:07:25
    away but there is an 03 API that should
  • 00:07:28
    be coming on that same date now they
  • 00:07:30
    also spoke about GPT 5 I did speak about
  • 00:07:33
    this in another video but since this is
  • 00:07:34
    a news video I'll quickly introduce not
  • 00:07:37
    introduce include what they said but
  • 00:07:38
    they basically talk about how for the
  • 00:07:40
    estimated time of arrival and the
  • 00:07:42
    performance they are still figuring out
  • 00:07:45
    that out but of course they do think we
  • 00:07:47
    will be happy with those results now we
  • 00:07:49
    actually also got an incredible update
  • 00:07:51
    from the unitary robot this was
  • 00:07:53
    something that genuinely surprised me
  • 00:07:55
    because I knew these robots were
  • 00:07:57
    incredibly you know flexible at what
  • 00:07:59
    they able to do but when I saw this
  • 00:08:01
    video it just surprised me because it
  • 00:08:03
    just goes to show that with a few
  • 00:08:05
    upgrades a few software fixes a few you
  • 00:08:07
    know more you know trainings or whatever
  • 00:08:08
    it was you know optimizations that
  • 00:08:09
    were're able to do to you know make this
  • 00:08:11
    robot a lot more effective that with the
  • 00:08:13
    same platform that they released you
  • 00:08:14
    know I think about a year ago now that
  • 00:08:16
    they were able to make a robot that
  • 00:08:17
    looks borderline human like jogging down
  • 00:08:20
    you know the hill right there that was
  • 00:08:21
    very uncanny like I would definitely
  • 00:08:23
    believe that that is CGI if I wasn't you
  • 00:08:25
    know in this space like if I saw that
  • 00:08:27
    robot walking by me I would be like what
  • 00:08:29
    the like what's going on like when did
  • 00:08:31
    these robots just Ascend and become very
  • 00:08:33
    humanlike so I think this is you know
  • 00:08:35
    showing you that the the robot race is
  • 00:08:38
    really really on and I think unry is one
  • 00:08:40
    of those companies that they definitely
  • 00:08:42
    must be working s days a week 24/7
  • 00:08:44
    because I've never seen a company move
  • 00:08:46
    this quickly in terms of their hardware
  • 00:08:48
    and in terms of the iterations to the
  • 00:08:50
    software in terms of how that's making
  • 00:08:52
    the hardware move like the you know the
  • 00:08:54
    gate on this is incredibly humanlike and
  • 00:08:57
    it's really incredible because prior to
  • 00:08:59
    you know the I I I don't even remember
  • 00:09:01
    what it was called but the engine AI
  • 00:09:02
    robot prior to that robot most robots
  • 00:09:04
    just walked in a very strange way so I
  • 00:09:07
    mean you know walking is definitely one
  • 00:09:09
    small thing but when you do have robots
  • 00:09:11
    that are just walking around like that
  • 00:09:12
    it's definitely going to you know maybe
  • 00:09:14
    not feel like a cyberpunk world but
  • 00:09:15
    start to feel like okay maybe this kind
  • 00:09:18
    of technology is actually real and I
  • 00:09:21
    think most people also underestimate how
  • 00:09:23
    difficult it is to get a robot to not
  • 00:09:25
    only run up a hill but also jog down a
  • 00:09:28
    hill in a way that looks humanlike but
  • 00:09:30
    also manage the stability the uneven
  • 00:09:33
    terrain and to do it in an environment
  • 00:09:35
    that it hasn't been PRI you know tested
  • 00:09:37
    on so I think that just goes to show how
  • 00:09:40
    robust these robots are getting and with
  • 00:09:42
    all the kind of policies that they're
  • 00:09:43
    probably you know running there I mean
  • 00:09:45
    it really goes to show that this team
  • 00:09:47
    honestly just doesn't make great
  • 00:09:48
    Hardware but also makes great software
  • 00:09:50
    to you know really boost the performance
  • 00:09:53
    of these robots so this is something
  • 00:09:55
    that's genuinely surprising me like I
  • 00:09:56
    thought robots would take a lot longer
  • 00:09:59
    but it's clear that even on the software
  • 00:10:01
    side like right now we're getting tons
  • 00:10:02
    and tons of investment and there is
  • 00:10:04
    clear incentive for these companies to
  • 00:10:06
    actually speed up the robotics process
  • 00:10:08
    because we know we're at this like you
  • 00:10:10
    know cross-section of where Ai and
  • 00:10:11
    Robotics kind of collide and that's just
  • 00:10:13
    going to you know further push the
  • 00:10:15
    industry in terms of what is potentially
  • 00:10:17
    capable I mean it's definitely really
  • 00:10:18
    going to be an interesting future now if
  • 00:10:20
    we're talking about humanoid robots and
  • 00:10:22
    the unitary you know G1 robot and this
  • 00:10:24
    robot is pretty small but I can imagine
  • 00:10:25
    a huge robot running around I want to
  • 00:10:27
    show you guys the engine AI robot now
  • 00:10:30
    this was one that broke the internet and
  • 00:10:31
    I didn't even include this in the
  • 00:10:32
    previous video cuz I actually forgot but
  • 00:10:34
    it was crazy because I remember when I
  • 00:10:36
    saw this one uh a lot of people were
  • 00:10:38
    basically stating that this was CGI and
  • 00:10:40
    I was like okay people are freaking out
  • 00:10:42
    because I remember when I produced
  • 00:10:44
    another video okay on engine AI on a
  • 00:10:46
    different kind of robot they were also
  • 00:10:48
    thinking that that one was CGI too so I
  • 00:10:51
    think technology is starting to reach
  • 00:10:53
    that point where we're starting to get
  • 00:10:55
    to that point where we actually may have
  • 00:10:57
    a situation on our hands where a lot of
  • 00:11:00
    the technology just doesn't seem real
  • 00:11:02
    like people who are in the space will
  • 00:11:03
    know that yes this is of course
  • 00:11:05
    realistic they know that you know
  • 00:11:07
    reinforcement learning and you know by
  • 00:11:08
    changing the policies of how the robot
  • 00:11:10
    works and updating it through you know
  • 00:11:12
    many different training cycles and that
  • 00:11:13
    kind of stuff you can achieve this
  • 00:11:15
    eventually but people that are outside
  • 00:11:18
    of the community would just think that
  • 00:11:19
    this is 100% computer generated because
  • 00:11:22
    of course they haven't been paying
  • 00:11:22
    attention to the updates and this is you
  • 00:11:24
    know I don't want to say surprising
  • 00:11:26
    because I understand that you know it
  • 00:11:27
    looks super super human but I'm starting
  • 00:11:29
    to think that if people think that is
  • 00:11:30
    CGI what happens on a robot can like run
  • 00:11:32
    around the corner can you know uh you
  • 00:11:34
    know jump up and down and do you know
  • 00:11:36
    other humanlike movements and it's
  • 00:11:37
    completely autonomous and has realtime
  • 00:11:39
    audio like people are going to be I
  • 00:11:41
    don't want to say losing their minds but
  • 00:11:42
    they're definitely going to be more
  • 00:11:43
    surprised than they are now because this
  • 00:11:45
    is just a human-like walking gate so if
  • 00:11:48
    people can't believe this on face value
  • 00:11:49
    when it does seem to be pretty realistic
  • 00:11:52
    it's going to be quite the shock when
  • 00:11:53
    you do have fully autonomous robots that
  • 00:11:55
    are out there doing a variety of
  • 00:11:57
    different tasks now there is is actually
  • 00:11:59
    something super interesting there is a
  • 00:12:02
    new company trying to achieve AGI so we
  • 00:12:05
    can see here that franchis so is joining
  • 00:12:08
    forces with Mike nup to start Nar a new
  • 00:12:11
    AI lab and their focus is deep learning
  • 00:12:14
    guided program synthesis and they're
  • 00:12:15
    betting on a different path to build AI
  • 00:12:17
    capable of true invention adaptation and
  • 00:12:20
    Innovation so if you are familiar with
  • 00:12:23
    these people well if you aren't familiar
  • 00:12:24
    with these people that person is Francis
  • 00:12:26
    soay and this is the guy that created
  • 00:12:28
    the Arc AGI Benchmark that is you know
  • 00:12:32
    highly holded as the key to AGI because
  • 00:12:34
    if you can solve it then potentially you
  • 00:12:36
    could solve AGI with that same kind of
  • 00:12:39
    framework that you use because those
  • 00:12:40
    problems are essentially supposed to be
  • 00:12:42
    resistant to memorization so that is the
  • 00:12:45
    claim there now this is super
  • 00:12:46
    interesting because I always love when
  • 00:12:48
    people are taking different approaches
  • 00:12:50
    to AGI because we've all seen how
  • 00:12:52
    opening I are taking their approach to
  • 00:12:54
    AGI and how Google are taking their
  • 00:12:56
    approach to AGI but I think we're
  • 00:12:57
    actually starting to get many different
  • 00:12:59
    approaches like whilst yes llm search is
  • 00:13:02
    you know putting us into a situation
  • 00:13:04
    where there are some really really
  • 00:13:06
    impressive discoveries about how these
  • 00:13:08
    systems scale differently to traditional
  • 00:13:10
    you know the GPT series of models but I
  • 00:13:13
    think that when we actually take a look
  • 00:13:15
    at other you know companies that are
  • 00:13:17
    looking at different architectures and
  • 00:13:18
    different ways to scale these models I
  • 00:13:20
    think that is where we're going to get a
  • 00:13:21
    lot of innovation because I think it was
  • 00:13:24
    Yan Lun that said You know open ey kind
  • 00:13:26
    of started this offramp where everyone
  • 00:13:28
    is just focusing on the llm architecture
  • 00:13:30
    because there's so much hype there's so
  • 00:13:32
    much money there's so much investment in
  • 00:13:33
    that area um and it kind of I don't want
  • 00:13:36
    to say ruined the space because
  • 00:13:37
    obviously it didn't it just actually
  • 00:13:39
    sped things up quite a bit but he was
  • 00:13:41
    saying that not many companies are
  • 00:13:43
    actually looking at different hybrid
  • 00:13:45
    architectures and that of course you
  • 00:13:47
    know that is something that meta is
  • 00:13:48
    doing but now that these guys are doing
  • 00:13:50
    it and since they created that crazy
  • 00:13:51
    Benchmark I would not be surprised if
  • 00:13:53
    they managed to do something crazy
  • 00:13:54
    that's able to learn in a remarkable way
  • 00:13:56
    now what's crazy about here is that they
  • 00:13:58
    say that you know current deep learning
  • 00:14:01
    based AI while impressive and
  • 00:14:02
    economically valuable is ultimately
  • 00:14:04
    constrained by its inability to
  • 00:14:06
    efficiently learn and adapt it excels at
  • 00:14:08
    known tasks but crumbles when faced with
  • 00:14:10
    open-ended problems it reflects back on
  • 00:14:12
    us only the knowledge programs and
  • 00:14:13
    abstractions found in his training data
  • 00:14:16
    like the man behind the curtain in The
  • 00:14:17
    Wizard of Oz it hides the fact that
  • 00:14:19
    human general intelligence actually
  • 00:14:20
    created that data and if AI cannot
  • 00:14:23
    efficiently adapt it is forever
  • 00:14:24
    constrained by what humans teach it and
  • 00:14:26
    to compress the next 100 Years of
  • 00:14:28
    scientific progress 50 or perhaps 10 we
  • 00:14:31
    need general intelligence not task
  • 00:14:33
    specific skill we need computers that
  • 00:14:35
    can pose problems and explore new
  • 00:14:37
    territory not just apply known Solutions
  • 00:14:39
    we need computers that can innovate now
  • 00:14:42
    the path to AGI is not through
  • 00:14:44
    incremental improvements to existing
  • 00:14:46
    methods the problem with deep learning
  • 00:14:48
    are fundamental and cannot be addressed
  • 00:14:50
    superficially it's time for a new
  • 00:14:52
    paradigm and the good news is that we
  • 00:14:54
    now know what this new paradigm is
  • 00:14:56
    program synthesis so this is where they
  • 00:14:59
    explain that you know this is the key to
  • 00:15:01
    AGI and I might make an entire video on
  • 00:15:03
    this if you guys want me to but uh it's
  • 00:15:06
    actually super interesting okay now I'm
  • 00:15:08
    going to just basically simplify this so
  • 00:15:10
    this is a different focus and it's
  • 00:15:12
    basically how you know most AIS learn by
  • 00:15:15
    you know guessing patterns and lots of
  • 00:15:16
    examples like recognizing cats from
  • 00:15:18
    millions of cat pictures but this
  • 00:15:20
    approach struggles to fully understand
  • 00:15:22
    things and it needs way too much data
  • 00:15:24
    like in order for humans to recognize a
  • 00:15:25
    cat you don't need to look at 50 million
  • 00:15:27
    different pictures of cats you don't
  • 00:15:29
    need to see a cat once or twice and be
  • 00:15:30
    like okay I know what a cat is so
  • 00:15:32
    program synthesis is you know I guess
  • 00:15:34
    you could say it's like human learning
  • 00:15:36
    because it's like teaching the AI to
  • 00:15:37
    write more programs that perfectly
  • 00:15:39
    explain how things work so for example
  • 00:15:42
    instead of showing the AI millions of
  • 00:15:43
    math problems you know it just learns
  • 00:15:45
    the rules of math and then it can solve
  • 00:15:47
    new problems with just a few examples so
  • 00:15:49
    this is pretty crazy because this allows
  • 00:15:52
    for generalization power so program you
  • 00:15:54
    know synthesis helps AI understand the
  • 00:15:56
    big picture better so it can solve you
  • 00:15:58
    know new problem s even with little data
  • 00:16:00
    which is how humans essentially learn so
  • 00:16:02
    this means it doesn't need as much data
  • 00:16:04
    and this mixes program synthesis with
  • 00:16:06
    deep learning which is patent
  • 00:16:07
    recognition which means that it could
  • 00:16:08
    become you know way smarter than humans
  • 00:16:10
    now this is you know a new and growing
  • 00:16:13
    field kind of like where deep learning
  • 00:16:15
    was back in 2012 but researchers are
  • 00:16:17
    starting to realize that this could be a
  • 00:16:19
    big part of the puzzle for building
  • 00:16:20
    smarter AI so it's actually a pretty big
  • 00:16:23
    moment for AI basically saying that look
  • 00:16:25
    we actually need to you know take new
  • 00:16:26
    approaches to AGI and this is going to
  • 00:16:28
    be one of the key ways that they do take
  • 00:16:30
    that so another thing that I forgot to
  • 00:16:32
    add and you know cover in one specific
  • 00:16:35
    video but this is why we have the news
  • 00:16:36
    videos is that open AI shared their
  • 00:16:38
    vision for shared Prosperity so this is
  • 00:16:40
    where they talk about shared Prosperity
  • 00:16:42
    is near and measurable as the new jobs
  • 00:16:45
    and growth that will come from building
  • 00:16:47
    more AI infrastructure from Chip
  • 00:16:49
    manufacturing facilities and power
  • 00:16:51
    plants and as our CEO SM elment has
  • 00:16:53
    written a will soon help our children do
  • 00:16:55
    things that we can't not far off is a
  • 00:16:57
    future in which everyone's lives can be
  • 00:16:59
    better than anyone else's life is now
  • 00:17:01
    and with such prosperity in sight we
  • 00:17:03
    want to work with the policy makers to
  • 00:17:05
    ensure that ai's benefits are shared
  • 00:17:07
    responsibly and equitably and this
  • 00:17:09
    economic brewprint that they were
  • 00:17:10
    outlining was designed to support the
  • 00:17:12
    entrepreneurship and individual freedoms
  • 00:17:14
    that have long been at the heart of the
  • 00:17:16
    American innovation ecosystem so overall
  • 00:17:18
    they're basically saying look they want
  • 00:17:20
    shared prosperity for everyone now I got
  • 00:17:22
    to Mo with you guys this was actually
  • 00:17:24
    met with a lot of backlash from you know
  • 00:17:26
    various different parts of the internet
  • 00:17:28
    um and I probably I'm not using the best
  • 00:17:30
    example of person to you know quote this
  • 00:17:33
    but uh Gary Marcus says uh that's awful
  • 00:17:36
    if they make a lot of money they aren't
  • 00:17:37
    going to share it and they will likely
  • 00:17:39
    destroy more jobs than they create and
  • 00:17:41
    this is you know one of the times where
  • 00:17:43
    I see a lot of people agreeing with Gary
  • 00:17:45
    Marcus because if you aren't familiar
  • 00:17:47
    with his reputation some would say he
  • 00:17:48
    has a reputation for just disagreeing
  • 00:17:50
    with everything in the main AI space
  • 00:17:52
    honestly do think Gary Marcus has a lot
  • 00:17:54
    of valid points about AI but I think
  • 00:17:56
    sometimes his valid points get
  • 00:17:57
    overshadowed by the fact that when there
  • 00:17:59
    is a genuine breakthrough in AI he just
  • 00:18:01
    simply disregards it at another you know
  • 00:18:03
    hype train or another thing that's just
  • 00:18:05
    too hyped up so for me I do think that
  • 00:18:08
    sometimes it just goes a bit too far but
  • 00:18:10
    the point is is that I did see a few
  • 00:18:12
    posts about Reddit saying you know we're
  • 00:18:13
    about to experience a very different
  • 00:18:15
    world in the future but most people
  • 00:18:17
    aren't even prepared for it now if you
  • 00:18:18
    want to talk about AI doing some you
  • 00:18:21
    know things that a lot of people
  • 00:18:22
    wouldn't want it to do which is of
  • 00:18:24
    course no more hires this is where we
  • 00:18:26
    have the Salesforce CEO Mark Bean off
  • 00:18:29
    actually talking about how he probably
  • 00:18:31
    won't hire much more software developers
  • 00:18:33
    this year because of the productivity
  • 00:18:35
    gains they're gaining from AI now at
  • 00:18:37
    first glance I think this is pretty
  • 00:18:39
    crazy because it's one of those
  • 00:18:40
    situations where it's like okay oh my
  • 00:18:42
    God like all the software engineering
  • 00:18:44
    jobs are gone like I said in another
  • 00:18:45
    video I think that is not the case I
  • 00:18:47
    think it's actually quite the opposite
  • 00:18:48
    but I do think that what AI is enabling
  • 00:18:50
    it's enabling for us to create more
  • 00:18:52
    companies at scale and what I mean by
  • 00:18:54
    that is that like what you used to need
  • 00:18:56
    maybe like 200 people to do you now
  • 00:18:58
    probably can do with like a 25 person
  • 00:19:00
    lean team and I think that's what this
  • 00:19:02
    is saying is that maybe we're not going
  • 00:19:04
    to fire people but maybe we're just not
  • 00:19:05
    going to hire any more people we are
  • 00:19:07
    going through a global labor shortage
  • 00:19:09
    where we see all these declining birth
  • 00:19:11
    rates we understand that it is harder to
  • 00:19:14
    hire especially people here right in the
  • 00:19:15
    United States in sales and service you
  • 00:19:17
    can see it and um this is going to give
  • 00:19:20
    us the ability to do more and as an
  • 00:19:22
    example of that look at engineering I
  • 00:19:24
    think in engineering this year at
  • 00:19:26
    Salesforce we're seriously debating
  • 00:19:28
    maybe we aren't going to hire anybody
  • 00:19:30
    this year because we have seen such
  • 00:19:31
    incredible productivity gains because of
  • 00:19:34
    the agents that work side by side with
  • 00:19:37
    our engineers and making them more
  • 00:19:38
    productive and we can all agree that
  • 00:19:41
    software engineering has become a lot
  • 00:19:43
    more productive in the last two years
  • 00:19:46
    with this basically these new models now
  • 00:19:48
    software Engineers aren't the only ones
  • 00:19:50
    getting the short end of the stick we
  • 00:19:52
    actually take a look at the Wall Street
  • 00:19:54
    job losses they actually might replace
  • 00:19:57
    200,000 rolls
  • 00:19:59
    with AI so we can see here that it talks
  • 00:20:01
    about how back middle office roles are
  • 00:20:04
    you know these are the ones that are at
  • 00:20:06
    risk and of course the bank's profit
  • 00:20:08
    could surge due to the improved
  • 00:20:10
    productivity so it says Global Banks
  • 00:20:12
    might cut as many as 200,000 jobs in the
  • 00:20:15
    next 3 to 5 years as AI encroaches on
  • 00:20:17
    task currently carried out by human
  • 00:20:19
    workers according to Bloomberg
  • 00:20:20
    intelligence and Chief Information and
  • 00:20:22
    Technology officers surveyed for
  • 00:20:24
    Business Insider indicated that on
  • 00:20:26
    average they expect a net of 3 % of
  • 00:20:29
    their Workforce to be cut so customer
  • 00:20:31
    services could see changes as botch
  • 00:20:32
    manage you know client functions while
  • 00:20:34
    know your customer duties would also be
  • 00:20:36
    vulnerables and any jobs involving
  • 00:20:38
    routine repetitive tasks are at risk so
  • 00:20:40
    it's going to be a really interesting
  • 00:20:42
    time as the role of you know individuals
  • 00:20:45
    or certain companies change we're going
  • 00:20:47
    to see a new management roles for people
  • 00:20:48
    managing teams we're going to see
  • 00:20:51
    different roles emerge I mean it's going
  • 00:20:52
    to be a a rapid rapid transformation as
  • 00:20:55
    companies try to you know reap the
  • 00:20:56
    benefits as quickly as possible and I do
  • 00:20:58
    personally think that majority of these
  • 00:21:00
    companies probably won't be hiring
  • 00:21:01
    people again they're definitely going to
  • 00:21:03
    be trying to you know make as much money
  • 00:21:04
    as they can because I think people need
  • 00:21:06
    to understand that most companies just
  • 00:21:08
    exist to make a profit I know that some
  • 00:21:10
    companies exist out there to help the
  • 00:21:12
    world but when we're talking about Banks
  • 00:21:14
    their job is literally to make money so
  • 00:21:16
    it wouldn't surprise me if they didn't
  • 00:21:17
    hire much more people but I do think
  • 00:21:19
    there will be some new roles that are
  • 00:21:21
    going to be super interesting and I
  • 00:21:22
    think it will make a lot of the you know
  • 00:21:24
    humano human interactions with clients a
  • 00:21:26
    lot more valuable and interesting and I
  • 00:21:28
    do think that there's going to need to
  • 00:21:29
    be a lot more people working in security
  • 00:21:32
    because with all of this you know AI
  • 00:21:34
    impersonating people's voice it being
  • 00:21:35
    able to talk exactly like you it being
  • 00:21:37
    able to you know take an image of you
  • 00:21:39
    and then immediately start a zoom call I
  • 00:21:40
    mean it's going to be pretty crazy now
  • 00:21:42
    if you're thinking what kinds of you
  • 00:21:43
    know jobs and stuff people are going to
  • 00:21:45
    be doing maybe you might be managing an
  • 00:21:47
    AI agent swarm as sa Adel has said at a
  • 00:21:50
    recent conference the real breakthrough
  • 00:21:52
    for me is can we demystify creation of
  • 00:21:56
    these agents and make it simple just
  • 00:21:58
    like how we say you know create
  • 00:22:00
    documents or we create spreadsheets we
  • 00:22:03
    should be able to create agents just
  • 00:22:05
    like that in fact the way to
  • 00:22:07
    conceptualize the world going forward is
  • 00:22:10
    every one of us doing knowledge work
  • 00:22:13
    will use copilot to do our work and we
  • 00:22:15
    will create a swarm of agents to help us
  • 00:22:19
    with our work and drive the productivity
  • 00:22:21
    of our organization and so to that end
  • 00:22:23
    we're building co-pilot studio so think
  • 00:22:25
    of this as as simple as if you can
  • 00:22:27
    create a spreadsheet or a document you
  • 00:22:29
    should be able to create an agent and
  • 00:22:30
    that's what co-pilot Studio does so here
  • 00:22:32
    I'm building I happen to be a Frontline
  • 00:22:35
    worker I want to create a field service
  • 00:22:37
    agent I have a field service system so
  • 00:22:39
    all I need to do is to be able to give
  • 00:22:42
    it a prompt right that says in
  • 00:22:43
    instructions let's say say I want you to
  • 00:22:45
    behave like a c you know uh a field
  • 00:22:47
    service agent and then I give it
  • 00:22:49
    knowledge sources in this case they
  • 00:22:51
    happen to be a SharePoint site with a
  • 00:22:53
    lot of documents uh as well as my field
  • 00:22:55
    service applications in this case let's
  • 00:22:57
    say it's a Dynamics applic so I give it
  • 00:22:59
    the two sources of knowledge I give it
  • 00:23:01
    the instructions click click click and I
  • 00:23:03
    have an agent that now can dock with
  • 00:23:06
    co-pilot so I can go into my co-pilot
  • 00:23:09
    and say at field service agent I can now
  • 00:23:11
    start talking to my field service agent
  • 00:23:13
    so that's the Simplicity um oh that's
  • 00:23:16
    the ease with which one can start
  • 00:23:19
    creating agents these thousands and
  • 00:23:21
    hundreds of thousands of agents that
  • 00:23:22
    will be there inside of an organization
  • 00:23:24
    just like documents and spreadsheets
  • 00:23:27
    that are there inside now something that
  • 00:23:28
    was also rather interesting was Sam
  • 00:23:30
    Alman actually spoke about how he
  • 00:23:32
    believes that we're actually going to be
  • 00:23:33
    moving even faster and remember how at
  • 00:23:36
    the start of the video I spoke about you
  • 00:23:37
    know opening eye or talking about how
  • 00:23:39
    things are moving pretty quickly in a
  • 00:23:41
    recent interview this was a statement
  • 00:23:43
    that was poured out because it is a
  • 00:23:45
    change in the tone of his voice and I
  • 00:23:48
    don't really usually see that there's
  • 00:23:50
    usually like a consistent I wouldn't say
  • 00:23:52
    hype train but there usually consistent
  • 00:23:54
    theme of you know we're going to do big
  • 00:23:55
    things we know where it's kind of headed
  • 00:23:56
    but recently it's like okay things are
  • 00:23:58
    moving even faster and what's crazy
  • 00:24:00
    about this is like you know when I've
  • 00:24:01
    seen opening eyes track record on the
  • 00:24:03
    things that they've done in the past I'm
  • 00:24:05
    actually you know I don't want to say
  • 00:24:07
    surprised but I'm actually somewhat
  • 00:24:08
    excited because I do know that whatever
  • 00:24:10
    it is it probably isn't as hypy as most
  • 00:24:13
    people think like I know all the time
  • 00:24:14
    people hype up products and stuff but
  • 00:24:16
    this is truly like next level stuff when
  • 00:24:18
    you actually think about what we're
  • 00:24:19
    dealing with they're dealing with you
  • 00:24:20
    know super intelligence AGI and that
  • 00:24:22
    kind of thing so the implications are
  • 00:24:23
    definitely profound and if that stuff
  • 00:24:25
    happens faster the implications are
  • 00:24:27
    going to happen much faster too what's
  • 00:24:28
    something you've rethought recently on
  • 00:24:30
    AI or changed your mind about I think a
  • 00:24:32
    fast takeoff is more possible than I
  • 00:24:34
    thought a couple of years ago how fast
  • 00:24:36
    feels hard to reason about but something
  • 00:24:38
    that's in like a small number of years
  • 00:24:39
    rather than a decade wow what do you
  • 00:24:41
    think is the worst advice people are
  • 00:24:43
    given on adapting to ai ai is hitting a
  • 00:24:45
    wall which I think is the laziest
  • 00:24:47
    [ __ ] way to try to not think about it
  • 00:24:49
    and just you know put it out of sight
  • 00:24:50
    out of mind now there was also this
  • 00:24:52
    thing called evolutionary scale model
  • 00:24:54
    free so this is basically about proteins
  • 00:24:57
    and how proteins ESS tiny machines in
  • 00:24:59
    your body that do super important stuff
  • 00:25:01
    like building muscles fighting germs and
  • 00:25:03
    basically keeping you alive and you know
  • 00:25:06
    scientists study these all the time to
  • 00:25:08
    make new medicines cure diseases and
  • 00:25:10
    even design useful things like glowing
  • 00:25:12
    proteins now there's this fancy new
  • 00:25:14
    model as I'm talking about called esm3
  • 00:25:16
    and this is basically the ultimate
  • 00:25:19
    protein expert and it's trained to
  • 00:25:20
    understand what they're made of how they
  • 00:25:22
    fall into shapes and what they actually
  • 00:25:23
    do and it can actually create new
  • 00:25:26
    proteins and invent you know brand new
  • 00:25:28
    new ones and it's designed to you know
  • 00:25:29
    invent a glowing protein that is totally
  • 00:25:32
    different to any that exist in nature
  • 00:25:33
    and it basically simulated 500 million
  • 00:25:36
    years of evolution in a computer and you
  • 00:25:39
    know they've essentially decided to open
  • 00:25:41
    source this technology so it's pretty
  • 00:25:44
    pretty crazy so I do think that this is
  • 00:25:47
    going to be the kind of thing that
  • 00:25:48
    opening I release sometime soon I'm not
  • 00:25:50
    entirely sure what kind of you know
  • 00:25:51
    Avenue they're going to go down because
  • 00:25:53
    there's just so many different vertical
  • 00:25:55
    that they could Challenge and they could
  • 00:25:57
    take on but it's going to be super
  • 00:25:58
    interesting to see where this stuff does
  • 00:26:00
    go we also got to see the crazy
  • 00:26:02
    effectiveness of AI in learning so there
  • 00:26:05
    was a new randomized control trial of
  • 00:26:07
    students using GPT 4 as a tutor in
  • 00:26:09
    Nigeria and 6 weeks after school
  • 00:26:11
    tutoring AI tutoring equaled 2 years of
  • 00:26:14
    typical learning gains outperforming 80%
  • 00:26:17
    of other educational interventions and
  • 00:26:19
    it helped all students especially girls
  • 00:26:21
    who were initially behind so I think
  • 00:26:23
    this is one of the things that like a
  • 00:26:24
    lot of people are caught up on llms and
  • 00:26:26
    large language models and AI for this AI
  • 00:26:28
    for that but there are so many different
  • 00:26:29
    vertical where this is truly helping a
  • 00:26:32
    lot of people and I think a lot of the
  • 00:26:34
    times like a lot of the stories I Do
  • 00:26:35
    cover just cover like really really
  • 00:26:37
    intense AI news but we're seeing that AI
  • 00:26:39
    has a wide range of use cases and there
  • 00:26:41
    is actually a video I do need to include
  • 00:26:43
    where someone is essentially blind but
  • 00:26:45
    they are now using the metag glasses and
  • 00:26:47
    they can simply say with a prompt
  • 00:26:49
    exactly you know hey what's in front of
  • 00:26:51
    me hey is my stake done hey what does
  • 00:26:53
    this sign say and I mean I can't imagine
  • 00:26:56
    being blind and then this technology
  • 00:26:57
    gets invented how much of a lifesaver
  • 00:26:59
    that is going to be so it's just
  • 00:27:01
    something that is just super super
  • 00:27:02
    incredible and it's pretty awesome so
  • 00:27:05
    another thing as well there's also a
  • 00:27:06
    crazy study about 01 and I covered this
  • 00:27:09
    in detail but it basically is not good
  • 00:27:11
    in terms of the reasoning stuff so you
  • 00:27:13
    know maybe that's why they're including
  • 00:27:15
    that new architecture but I think it was
  • 00:27:17
    super interesting because they talk
  • 00:27:18
    about how just like they did with the
  • 00:27:19
    Apple paper there was a 30% reduction in
  • 00:27:21
    accuracy when math problems were
  • 00:27:23
    slightly variated which kind of su su
  • 00:27:25
    guess that maybe the architecture that
  • 00:27:26
    we're using isn't the most efficient
  • 00:27:28
    even if we can get the solutions there's
  • 00:27:30
    a 30% reduction in accuracy when the
  • 00:27:32
    problems are slightly variated which is
  • 00:27:34
    not good so basically saying that look
  • 00:27:36
    if these llms are so reliable and
  • 00:27:37
    they're so good why is there a 30%
  • 00:27:39
    reduction especially for these reasoning
  • 00:27:40
    models on similar questions where we
  • 00:27:43
    only change the numbers and the tech so
  • 00:27:46
    it's definitely a you know a little bit
  • 00:27:48
    of a red flag but it was something that
  • 00:27:49
    I covered that was super interesting now
  • 00:27:51
    something else that was rather
  • 00:27:52
    concerning was the fact that we're now
  • 00:27:54
    starting to get more and more
  • 00:27:55
    conversations about Ai and warfare in
  • 00:27:58
    the sense that you know we're going to
  • 00:27:59
    be using chat gbt to you know power
  • 00:28:01
    killer humanoid robots but in the fact
  • 00:28:03
    that like AI is quickly becoming a
  • 00:28:05
    National Asset like a national thing
  • 00:28:07
    that countries are really starting to
  • 00:28:09
    pay attention to because eventually AI
  • 00:28:11
    will most likely be used to create
  • 00:28:14
    weapons of National Defense in terms of
  • 00:28:16
    the inventions in terms of what it's
  • 00:28:18
    going to be capable of doing and
  • 00:28:19
    managing systems and in this short clip
  • 00:28:21
    Yoshua Benjo actually talks about how
  • 00:28:23
    you know data centers in the future
  • 00:28:25
    might actually become military targets
  • 00:28:27
    for foreign Nations that want to you
  • 00:28:29
    know take out a country or catch up in
  • 00:28:32
    the AI race so it's it's quite
  • 00:28:34
    interesting cuz this is something that
  • 00:28:36
    you know I wasn't really focused on I
  • 00:28:37
    did know that this you know is a thing
  • 00:28:39
    that will exist I remember reading the
  • 00:28:41
    very long- winded document called AI the
  • 00:28:43
    decade ahead and I was thinking about
  • 00:28:45
    how like the role of AI is going to
  • 00:28:47
    quickly change and this is going to be
  • 00:28:49
    something that you know countries have
  • 00:28:50
    to face and of course things like these
  • 00:28:53
    AI Labs like currently right now they're
  • 00:28:54
    just research organizations you know
  • 00:28:56
    it's just some really smart people at
  • 00:28:58
    those labs but eventually those labs are
  • 00:29:00
    going to have to be really really secure
  • 00:29:02
    like you're going to have to have some
  • 00:29:03
    Securities clearance to be able to go in
  • 00:29:05
    there there's probably going to be you
  • 00:29:06
    know 10ft walls with Bobb dwire on the
  • 00:29:09
    outside I mean that is going to
  • 00:29:10
    completely change once we do get to AGI
  • 00:29:12
    or ASI and I think that change is
  • 00:29:14
    probably coming a lot closer than people
  • 00:29:16
    think because this technology has the
  • 00:29:18
    potential to change everything provided
  • 00:29:20
    it's applied in the right way imagine a
  • 00:29:24
    country that is not leading in AI
  • 00:29:31
    and has
  • 00:29:33
    nukes you can guess which
  • 00:29:36
    one and they don't want to see say
  • 00:29:40
    us develop Weaponry that would be way
  • 00:29:42
    above what they can defend against so
  • 00:29:46
    what's their
  • 00:29:48
    option yeah press the button destroy
  • 00:29:51
    destroy our data centers yes so data
  • 00:29:54
    centers are going to become a
  • 00:29:56
    military uh
  • 00:29:59
    asset when they can run AGI
  • 00:30:03
    [Music]
标签
  • AI
  • OpenAI
  • AGI
  • robotics
  • employment
  • program synthesis
  • innovation
  • military
  • technology
  • education