• 4 minutes ago
The tech reporter talks with David Remnick about the the rise of Anthropic, the fall of Jensen Huang, and why not to freak out—too much—about the new age enveloping us all.
Transcript
00:00I do want to start with the subject of Anthropic, a company that's just gone public.
00:04In their filing paperwork for the first half of 2026, they reported revenue of $16.3 billion
00:12and a net loss of $80.7 billion. The company is also on the hook for hundreds of billions
00:19in future obligations for chips and data center leases. How do they make the math make sense?
00:30Well, there's a lot to take in. But basically, I think that the bet that investors are making
00:36on this company is that it will just continue to grow at something close to the rate that it has
00:41been growing for the last several years. I mean, this is a company that I think has been growing
00:46faster than any company at its scale in the history of capitalism. So four or five years ago,
00:52they had essentially no revenue. Three years ago, they had about $10 million in revenue. And then
00:58it's been 10xing every year for the last three or four years, not 10%, 10 times. So that is due
01:09in
01:09large part to the popularity of their models for coding and other white collar work. And the bet is
01:16that that just continues that as these tools get better and more helpful, that more white collar
01:21knowledge workers will incorporate them, developers will be paying much more, because the usage just
01:29continues to go up. So the math doesn't make sense. If you think that this is about as popular as
01:36AI
01:37will ever be, and as widely adopted as it'll ever be, it starts to make more sense if you anticipate
01:43that what has been happening for the last few years will keep happening, which is that the models get
01:48better, more people use them. They're paying, you know, companies out here in Silicon Valley, some of
01:54them are spending more on AI tokens than they are on payroll now. So that is the bet long term
02:00for how
02:00this math pencils out is that the economy essentially switches over from primarily running on human labor and
02:08paying human salaries to running on AI labor and paying token prices.
02:13Right. Let's break that down. How will the economy, if that in fact is the right bet to make,
02:19give me a picture of the economy on an AI basis as opposed to human labor. What's it going to
02:25look
02:25like? I think it looks a lot like my company. So I'm starting a startup right now, Machine Gods Media.
02:32We are a six person company right now. And so in our Slack, where we all communicate all day,
02:40we have six humans and probably a dozen agents. And the agents are doing things that we don't know how
02:50to do ourselves. They're writing code. I would just got back from I jokingly call it our engineering
02:56department, but it's just a bunch of clods running around. And they're building tools. They are,
03:03helping with scheduling and coordination with all of the pieces that go into making a podcast that
03:10aren't the actual talking on the microphone. And what would those be? Is it so complex that you need
03:17Claude to do it? It turns out to be pretty complex. So for example, just one, I don't know,
03:24PQ an example is like we have all these deliverables. So when you have a show, it needs to get
03:29delivered in
03:30all these different formats. And Apple needs one format. And Spotify needs one format. And YouTube
03:35needs another format. And it's all got captions and titles and descriptions and metadata. And like
03:40all that has to be used to have to be done by hand. And now we have this tool that
03:44does it all for us.
03:45So tell me how that works for a big law firm as opposed to a six person podcasting outfit.
03:50So it might not be a big law firm. It might actually be a small law firm in terms of
03:55the human
03:57payroll and footprint. It might be a law firm that only has 10 people. But using AI, they have sort
04:04of
04:04access to essentially unlimited amounts of paralegal, junior associate quality labor. And so they can do
04:14with 10 people what a more traditional firm might need 1000 people for. Now, that may or may not be
04:22possible. There might be gaps in what these models can do that make them basically unsuited to that level of
04:29integration. But that's the bet that investors are making is not that just Anthropic will continue to grow or that
04:36its models will get better, but that like this essentially becomes the foundation of a new kind of company.
04:43that we that we built on top of AI.
04:46How many of these companies can prevail in the new world of AI? We have Anthropic, there's OpenAI, there are
04:51other
04:51companies as well. Who's in the lead? And why?
04:56Right now, there are two clear frontier labs, OpenAI and Anthropic. I would say of the two, Anthropic may be
05:05slightly ahead, its models are considered better by some metrics than OpenAI's. But they're, they're definitely the
05:13front runners, the two of them. And then there's a, you know, drop off of small to medium drop off.
05:19And
05:20then there's Google DeepMind, which is to my mind, the third frontier lab. And then there's a pretty steep
05:26drop after that. And then you get to the metas and the SpaceX, SI, whatever they're calling themselves
05:32today. And the Chinese open model companies, some of the other players. But I think the frontier in
05:41America in the West is OpenAI, Anthropic and Google DeepMind.
05:45What do I get out of using Claude, which is obviously Anthropic's agent, as opposed to ChatGPT from OpenAI?
05:54You know, I use both of them. I'm pretty promiscuous. And I find that some of them are better for
06:00some tasks than
06:01other tasks. For me, Claude is what I use to write code, to help with editorial feedback on my writing.
06:10I don't use AI to
06:11write, but I do use it to help me revise and edit.
06:15Okay, I have to stop you there as an editor. Do you get editorial feedback that is any good, as
06:21opposed to when
06:21you used to write for the New York Times, or when you're writing your book?
06:25No, I gotta say, I mean, I've been blessed and fortunate to have some of the most amazing human
06:30editors in my career, including at the Times and at FSG, where I published my book. So I hope they're
06:37plugging their ears when they hear this next part, which is that AI has gotten really good at editing.
06:41So before I turned in a chunk of my manuscript...
06:44You're killing me here, Kevin.
06:45I know. Look, I am not here to tell reassuring fictions. I'm here to tell uncomfortable truths.
06:54I had a council of Claude's that would look at my chapters before I submitted them to FSG,
07:00and give me feedback from various perspectives. I would sort of assign one to be the, you know,
07:05the dyed-in-the-wool AI skeptic, and one of them to be sort of an AI historian that knew
07:11everything
07:12about the technical details of the history of these labs. And I would get feedback from them
07:17in the form of these long dossiers, and maybe 90% of it was crap. But 10% was great.
07:23And it actually
07:24caught... My council of Claude's caught a few mistakes that would have otherwise made it into...
07:28Whoa, whoa, whoa. 90% of the comments were crap, and you think that's good editorial counsel?
07:33Well, how much of what your editors give you is useful?
07:37100%.
07:39I see.
07:42So, Kevin, are you betting that Anthropic is right, that they will prevail, that they'll keep
07:47going at 10x growth, and that investing in Anthropic is a good idea?
07:52Well, I am not an investor in Anthropic. I do not own any AI stocks, and I cannot give financial
08:01advice. The thing I will say, just from the perspective of someone who has been reporting
08:08on this company for a long time, is they are a very unusual company. And I would not plug them
08:14into any kind of conventional valuation spreadsheet that you would use for any other company. I'll give
08:23an example of that. So, one of the core parts of Anthropic's culture is this thing called the
08:29culture interview, where when you are applying for a job at Anthropic, they bring you in, they give you
08:35all the technical tests and stuff if you're an engineer. Then they have someone interview you,
08:39and it's just about... It's the culture fit. They're assessing whether you kind of belong here,
08:45whether you're going to buy into the mission. And one of the questions they ask you is,
08:51how would you feel or how would you respond if we did something as a company that caused the price
08:56of our stock to go to zero? And the correct answer to that question is, I would absolutely support
09:04that decision to take the stock price to zero if it meant making a move on behalf of AI safety.
09:11So, they are a company full of missionaries and zealots. They are not normal by any stretch of
09:17the imagination. And so, I have no idea how the market will react to this company being public.
09:25It may be rapturous, and they may be one of the most valuable companies in the world,
09:31or it may just be that it's too weird to fit this unusual company in this very typical public market
09:36listing. But this is where I'm confused. The PR that I get, and sometimes the press coverage that
09:42I get is that Anthropic is the good company, as you suggest. In other words, the job interviews
09:49even have this moral dimension, whereas OpenAI, quite a different thing. Is that fair?
09:56I think it's fair that that's the image that has been portrayed.
10:00What's the reality?
10:01Well, the reality is that Anthropic and OpenAI are both locked in this race together.
10:07And to my mind, like, I think we have been focusing on the wrong question with respect
10:14to these companies, which is, you know, I've spent time with Sam Altman, with Dario Amadei,
10:19with lots of leaders and researchers at both of these companies.
10:22Dario Amadei is the head of...
10:24Anthropic.
10:25Anthropic, yes. And OpenAI is obviously Sam Altman.
10:27And I think we in the tech press are used to asking the question of, like, are these good
10:32and trustworthy people running these companies? And to my mind, that's often less useful than
10:39looking at the technology itself, because a good, moral, ethical person could still build
10:45an AI model that goes rogue and kills millions of people. On the other side of the coin, someone
10:51who is totally craven and corrupt could manage to produce an incredibly ethical AI model with
10:59all of the proper guardrails. So I actually don't know how much it makes sense to fit the image of
11:06the founder or the CEO onto the overall impression we have of these companies.
11:10Well, in your book, it's a fascinating book, you write that a surprising number of people
11:15who work in these AI companies join them because they were terrified. Because they were terrified.
11:22They decide that if AI was going to be built, that it would be better to be built by people
11:27who were scared, who were alarmed, who were cautious in order to take the appropriate
11:32measures against disaster. Is there any evidence at all that these people who are terrified are
11:39having any sort of significant impact within these companies to hold them back, to restrain
11:44them, to modify them?
11:46Yes, I think absolutely. I mean, I think it's almost impossible to imagine a counterfactual because
11:54the companies were started by people who were terrified of AI. You know, Sam Altman and Elon Musk
12:00started OpenAI in 2015 because they were scared that Demis Asabis and DeepMind were going to win the
12:08race to AGI, this artificial general intelligence. Just explain once and for all, what is AGI?
12:15Yeah. AGI you can think of as a human level AI that can do essentially anything the human brain can.
12:22And then several years later, Dario Amadei and six of his co-workers at OpenAI leave and start
12:32Anthropic because they are terrified of the technology and they're terrified that Sam Altman is not
12:38trustworthy enough to build it. So this is the motif running through the modern AI industry is that
12:44people who are scared of AI and don't believe their bosses or a rival is taking enough steps to
12:51contain it, break off and start their own thing. And they say, we're going to be the safe AGI company.
12:58Those other guys are going to blow up the world if they do it, but we can be trusted.
13:02Some of this, it seems to me, has to do with personality and the way that personality is projected
13:06out into the greater world and the press. Sam Altman projects, and I'm going to ask you to describe it,
13:12one kind of personality. And Dario Amadei, maybe something else. Tell me a little bit about that
13:18and what effect it has. Well, Sam Altman is a master salesman and fundraiser. That is sort of his
13:29superpower from a very young age. He was a networker. He was very close to big deal makers in Silicon
13:36Valley, extremely talented at raising money. He ran Y Combinator, the startup incubator,
13:43before he left to become the CEO of OpenAI. And his reputation is kind of an extremely ambitious,
13:53high EQ people pleaser, right? He will, his superpower, someone once told me, is if he can,
14:01he can figure out, he can laser in on exactly what you want to hear and give it to you.
14:07And that has
14:08served him well in some instances, and it has served him poorly in others, such as when his board decided
14:13that he was lying to them and tried to fire him. And it has led to his reputation as kind
14:20of two-faced
14:21or kind of duplicitous. So that's Sam. He's sort of the inveterate people pleaser who wants to tell you
14:28what you want to hear. And what about Dario? Dario is a very high IQ, stubborn neurotic is the way
14:41I
14:41would describe it. He is a moral absolutist. He definitely has a sort of streak within him that
14:47believes strongly in the good and the right and the righteous. But he's also a brilliant scientist
14:56and saw many of today's AI trends and dynamics playing out like years before anyone else.
15:05And his superpower and his vice, his kryptonite, are that he cannot lie. This was something that
15:17someone who is close to him told me about him. They said, Dario, this is a guy who cannot tell
15:23a lie
15:23to save his life, which is an asset for us on even-numbered days and a liability for us on
15:29odd-numbered
15:30days. How does that work? Well, sometimes it leads to him standing his ground and standing up for his
15:36principles. And when the Pentagon says, we want to use Claude for whatever we want to do, he's the guy
15:45who
15:45actually has the cojones to stand up to them while everyone else is busy trying to snivel and get on
15:53their good side. And of course, the Pentagon banned the use of Claude for a while.
15:57Yes. And a lot of people at Anthropic saw that and admired Dario for standing his ground, even when it
16:03would have been expedient for him to sort of say whatever it took to get Pete Hegseth happy. But then
16:10it also leads to these incredible moments where, you know, like, remember that press briefing outside
16:17the White House where all this AI CEOs are standing around in a half circle and Dario's there and he's
16:24just like jumping out of his skin, like cannot contain his physical displeasure at having to stand
16:31there next to these cowards. And he can't hide it. It's just all over his face. So I think that's
16:38what
16:38they meant by asset on even numbered days, liability on odd numbered days.
16:43I've had this experience and maybe you have in, you know, many more times. There's a certain,
16:48maybe too easy confidence in the world of long form journalism, for example, that this is very
16:55special. The accumulation of and the analysis of and the presenting of a narrative form information is
17:02something that AI is not going to do. Whereas, you know, give me the 10 best restaurants in Seattle,
17:10that's easy to scrape off the internet and present. So there are a couple of things I'll say about this.
17:15One is, I think there will still be a market for human authored work. There just is something about
17:27AI writing that is hollow, that is generic. And what's interesting, they've run all these
17:35blinded tests where they show people two works of literature, two snippets of writing,
17:43and they ask them which one they prefer. And most of the time people prefer the AI generated one if
17:48they don't know. Really?
17:51But then when they unblind it, when they tell you this one was AI, this one was human written,
17:57the preference swaps. So what does that tell you?
18:01Almost everyone likes the human written one.
18:03But they like the idea of the human written one, but they might not innately like the
18:07human one at all.
18:08So I think the future of at least some writing involves humans sticking their names on works of
18:16AI generated literature, because they don't want to break that connection, the spell with the
18:21audience.
18:22That is not entirely reassuring, Kevin.
18:25I'm sorry, I wish I had a rosier view.
18:29Can I just say, you do seem to have, on the continuum of people who are doomers,
18:34who think that we're all going to die as a result, and some of them work at AI companies,
18:38as we know, to people who are absolutely sanguine about this whole thing, and that it's going to be
18:44great, and we're going to work maybe one day a week, and the rest of the time we're going to
18:48sun ourselves, and I don't know, whistle all day long, that you are pretty sanguine about this
18:56subject, and the opposite end of a doomer. Am I right?
19:02I am an optimist by temperament. I believe that we can and should thread this needle,
19:13but I think we've got a lot of work to do. What we're talking about is not just...
19:20What concerns you the most, Kevin?
19:23Right now, what concerns me is
19:27an intelligence explosion.
19:29This is this term in the AI industry
19:33for what could happen after AI
19:37becomes capable of building
19:41AI. So recursive self-improvement, it's sometimes
19:45called, but this is the concept that you have. An AI model that becomes
19:49superhuman at
19:51development of AI models.
19:53So it's able to do the entire workload of a very senior
19:58researcher, engineer at
20:00OpenAI or Anthropic.
20:02Now, this has not
20:04happened yet, but there are some folks,
20:06including Meter, one of these
20:07credible third-party organizations
20:09that believes this could happen
20:11in the next year,
20:13where you have now the ability
20:16to use
20:18AI in a closed-loop, no-humans
20:22environment
20:22to train
20:24more capable
20:25AI models.
20:26And what are the potential disasters
20:28in concrete terms that could come out of
20:30that manifestation?
20:32Well, if these people are right,
20:34let's just caveat that they might not be,
20:36but if we did end up with, say, something like
20:38a decade's worth of
20:40AI capabilities improvements
20:43in the span of a year,
20:44because all this can now happen at
20:46machine speed rather than
20:47human speed,
20:48then I think you're looking at
20:50systems that
20:51just start to break
20:53a lot of things
20:54all at once.
20:55We're seeing some of this
20:56with the cyber attacks,
20:58with the incident
20:59on Hugging Face,
21:02with now
21:03the, you know,
21:03the U.S. government
21:04scrambling to
21:06contain threats
21:07from these AI models
21:09to critical infrastructure.
21:12I just think it starts
21:13to feel like that,
21:15but like times
21:1610 or times 100,
21:18where like
21:18there's just this kind of
21:20base rate
21:21of
21:23chaos
21:24and instability,
21:26you know.
21:26And at the same time,
21:28a very low
21:29propensity
21:30toward regulation.
21:32For now.
21:33Yeah, I think this could change
21:35very quickly.
21:36In fact,
21:37I believe that
21:38part of the reason
21:40the Democrats,
21:41you know,
21:42won the midterm
21:43so decisively
21:44was because
21:45Anxiety about AI.
21:46People had anxiety
21:47about AI.
21:48So things could change
21:49pretty quickly on this,
21:50but as unpopular
21:52as AI might be
21:53right now,
21:55I expect it
21:56to be more unpopular
21:57as, you know,
21:58right now
21:59we're seeing this
22:00massive backlash
22:01against AI.
22:03Unemployment
22:03is still under 5%.
22:06It was not at all
22:07intuitive to me
22:08that we would start
22:08to see this level
22:09of backlash
22:11before any of
22:12the widespread
22:13job loss
22:14arrived.
22:15Well, because you're
22:16reading about it
22:17every two seconds
22:17and you're seeing,
22:18you know,
22:19reports on television
22:20and on YouTube
22:20and all the rest.
22:22There's an anticipatory
22:23anxiety about it.
22:25But think about
22:25how much more
22:27anxiety there would be
22:28if people were actually
22:29losing their jobs
22:30en masse.
22:31I'm not saying
22:31I'm cheering
22:32for that outcome.
22:33I desperately want
22:34that not to happen,
22:35but I just,
22:36to me,
22:36the timing
22:37of this backlash
22:38has been a little
22:38surprising
22:39because it seems like
22:40there's still
22:41a lot more risk
22:42out there
22:43to be priced in.
22:44Now,
22:45I think you have
22:46to be impressed
22:46that we've gotten
22:47this far
22:47in our conversation
22:48without having
22:48mentioned Jensen Huang.
22:50Not bad.
22:51What do you make
22:51of this?
22:52What did we just see?
22:55My God.
22:56It's like the worst
22:57James Bond plot
22:58you have ever
22:59experienced
23:00in your world.
23:00It really is.
23:01It's dime store
23:02stuff.
23:03And I mean,
23:05look,
23:05this is a guy
23:06who has been
23:07so central
23:07to the AI boom.
23:11You know,
23:11NVIDIA's GPUs
23:14are the reason
23:15we have
23:15any of this
23:17today.
23:18And Jensen...
23:18Tell us what a GPU is.
23:19GPU is a graphics
23:20processing unit.
23:21Basically,
23:21a modern AI chip
23:24is descended
23:25from these
23:26things that gamers
23:27used to stick
23:27in their PCs
23:28to be able
23:29to like
23:29render backgrounds
23:31and polygons
23:32better.
23:33And Jensen Huang
23:33is the reason
23:34that those are now,
23:35you know,
23:35that those are now
23:36the most valued
23:37commodity in the world.
23:39So valued
23:40that we have had
23:41to pass
23:42special export controls
23:43to prevent China
23:44from getting access
23:45to them
23:45because
23:47if compute
23:48is the
23:49determining factor
23:50between success
23:51and failure,
23:51we don't want them
23:52to get our most
23:53powerful chips.
23:53You figure they have
23:54it anyway?
23:55Well,
23:56it sure seems like
23:57they've gotten
23:58a bunch of them anyway.
23:59Thanks to Alice Wong?
24:01Thanks to Alice Wong.
24:02Well,
24:02there's this whole
24:03sort of gray market
24:04of these cutouts
24:05in Malaysia
24:07or Singapore
24:07or other countries
24:08where they basically
24:10will buy a bunch
24:10of NVIDIA chips
24:11because they're allowed
24:12to buy them
24:13because they're not China
24:14and then they will
24:15either set up
24:15a data center
24:16and give Chinese companies
24:18remote access
24:18to the data center
24:19or in some cases
24:21they will just
24:21smuggle the chips
24:22to China.
24:22And imitate them,
24:23no?
24:23Can't they be replicated
24:24in some way
24:25once you analyze
24:26what's in front of you?
24:27So there is this process
24:29called distillation
24:30which is something
24:31slightly different.
24:32That's basically
24:32where you take
24:33a Claude
24:34or a ChatGPT
24:35and by just
24:36running essentially
24:37a coordinated
24:40siphoning operation
24:41you can kind of
24:42distill
24:43or embody
24:46the intelligence
24:47of the model
24:48into a smaller model.
24:50Steel would be
24:50another way
24:50to do that.
24:53But
24:54so between
24:55chip smuggling
24:56and distillation
24:57there's a lot
24:57going on
24:58in the Chinese
24:59AI market
25:00that we don't
25:00yet understand
25:01and one of the
25:01major questions
25:02going forward
25:03is like
25:03how much of this
25:04did Jensen Huang
25:05know about?
25:06You know
25:06was he aware
25:07that NVIDIA chips
25:08were being smuggled
25:09into China
25:09and how much
25:11did his
25:12sort of relationship
25:13with Alice Huang
25:14and Megaspeed
25:14factor into that?
25:16So what are
25:16the ramifications
25:17of this?
25:18I mean
25:19we'll see
25:19if Jensen Huang
25:21ends up
25:21you know
25:22he's now saying
25:23he would like
25:23to return
25:24to NVIDIA
25:25I suspect
25:25that won't work
25:26probably
25:27No I'd like
25:27to win the Super Bowl
25:28too but I don't
25:28think that's happening
25:31I mean
25:32I think
25:32in the short term
25:34what it means
25:35is that
25:35the entire
25:37industry
25:38will kind of
25:38have to
25:39approach
25:40things
25:41slightly differently
25:42I mean
25:43for many years
25:43Jensen Huang
25:44has been
25:45the sole guy
25:46with control
25:48of the commodity
25:49that everyone
25:50in AI
25:50needs
25:51that is
25:51the key
25:52ingredient
25:52and his
25:53relationships
25:54really determined
25:55who got
25:56how many
25:56of NVIDIA's
25:57latest chips
25:58it was not
25:58just a pure
25:59like auction
26:00you know
26:01market
26:01it was like
26:03oh
26:04Alice Wong
26:04and I are
26:05friends
26:05so I'll save
26:06some of these
26:06chips for Megaspeed
26:08or oh
26:08Elon Musk
26:09and I go way
26:09back
26:10I'll give him
26:10a big allotment
26:11of some of these chips
26:12who benefited most
26:12from the relationship
26:15I think
26:16Elon Musk
26:18has gotten
26:18a very large
26:20allotment
26:21of chips
26:21that maybe
26:22were out of step
26:23with his company's
26:24importance
26:25in AI
26:26because he and
26:27Jensen Huang
26:28are old friends
26:29I think
26:30Sam Altman
26:32has benefited
26:32from a close
26:33relationship
26:34with Jensen Huang
26:35Dario Amade
26:36and Anthropic
26:37have actually
26:37not had
26:38a close
26:38relationship
26:39with Jensen Huang
26:40in fact
26:40as I reported
26:41in my book
26:41there was this
26:42very tense
26:43meeting
26:44between
26:45Dario Amade
26:46and Jensen Huang
26:47the first time
26:48they ever met
26:49where
26:50you know
26:50Jensen is basically
26:52grandstanding
26:53about how big
26:54the data center
26:54they're going to
26:55build is
26:55and Dario
26:56is sort of
26:56peppering him
26:57with sort of
26:58rude questions
26:59oh is it
26:59how big is it
27:00going to be
27:00what order of
27:01magnitude
27:01where are you
27:02going to get
27:02the power
27:03and eventually
27:04Dario just
27:04sort of
27:05mutters to
27:05someone sitting
27:06next to him
27:07that Jensen
27:07is sort of
27:08acting Trump-like
27:09and the two
27:10of them
27:11didn't like
27:12each other
27:12for a long time
27:13but if Dario
27:14didn't have
27:15that kind of
27:16access for Anthropic
27:18how has he
27:19prevailed anyway
27:19they have adopted
27:21a very unusual
27:24compute strategy
27:25which is that
27:26they have
27:26essentially
27:27gotten a little
27:28bit from everyone
27:29so they have
27:30a bunch of
27:31Amazon chips
27:32called Tranium
27:33they rent a bunch
27:34of chips
27:34from Google
27:35and they rent
27:36a bunch of chips
27:37they have a bunch
27:38of NVIDIA chips
27:39too
27:40and so
27:40they have
27:41rather than
27:42kind of
27:42tethering
27:42themselves
27:43to one
27:44chip company
27:45or one
27:46hyperscaler
27:47they have
27:48kind of
27:49drawn from
27:50all of them
27:51to varying
27:51degrees
27:53but they're
27:54compute constrained
27:55just like
27:56everyone else
27:56in the industry
27:57they wish
27:57they had
27:58more data
27:58centers
27:59they wish
27:59they had
27:59more chips
28:00no one's
28:01got enough
28:01you know
28:03here we are
28:04again
28:04on the brink
28:05of another
28:05revolution
28:06that is arguably
28:07much greater
28:08than anything
28:09represented by
28:10the internet
28:11or
28:13the iPhone
28:14and the like
28:15something vastly
28:16more complicated
28:17despite your optimism
28:19in terms of human
28:21relations
28:22in terms of the way
28:23we live our lives
28:24what are you
28:26expecting in the
28:27next decade
28:28or so
28:30a decade
28:31is so long
28:32from now
28:33David
28:34I mean I
28:34I really
28:35that's fair
28:36I struggle to think
28:37about what the world
28:39will look like
28:39in a year
28:41really
28:42I do
28:43I mean I think
28:44parts of it
28:45will look
28:45largely the same
28:46I will still
28:47probably
28:49you know
28:50drop my kid
28:50off at school
28:52go to the
28:53grocery store
28:54there will be
28:55parts of my life
28:56that are
28:56much the same
28:58when you and I
28:59talk next Thanksgiving
29:00what will we have
29:02experience
29:03what changes
29:04will have been
29:04brought into our lives
29:05over the course
29:06of a year
29:07I think
29:08on the good path
29:09if the next year
29:10plays out
29:11in a way
29:12that is
29:13thoughtful
29:15I think
29:15we will have
29:16put in place
29:17vastly
29:18more
29:19guardrails
29:20where do you
29:22get your
29:22confidence
29:22that thoughtful
29:23is
29:25something
29:26that's
29:27anywhere in
29:28abundance
29:29in the US
29:30government
29:30or anywhere
29:31else
29:33thoughtful
29:33or maybe
29:34just
29:35self-interested
29:36I think
29:38we
29:38have a deep
29:40desire
29:40to
29:42remain
29:43in control
29:44of
29:45our
29:46society
29:47of
29:48our
29:48government
29:49of our
29:49surroundings
29:50of our
29:51lives
29:51I think
29:52that we
29:52still
29:54want to be
29:55in the driver's
29:56seat
29:56and as long
29:58as we
29:58recognize
29:59that AI
29:59is not
30:00just
30:00a tool
30:02or a hype
30:02bubble
30:03but is
30:03actually
30:04in some
30:04sense
30:04a
30:07type of
30:08entity
30:08that
30:09could
30:10displace
30:10us
30:11at the
30:11top
30:11of the
30:12intelligence
30:12ladder
30:15we
30:15could
30:16be
30:16the
30:16second
30:16smartest
30:17thing
30:17on
30:18earth
30:18a
30:19year
30:19from
30:19now
30:21and
30:24I
30:24think
30:25we
30:25have
30:25a lot
30:26to do
30:26between
30:27now
30:27and
30:27then
30:27but
30:28my
30:29hope
30:29would
30:29be
30:29that
30:30we
30:30look
30:31back
30:31and
30:31we
30:31say
30:32wow
30:33that
30:33was
30:33a
30:33crazy
30:34year
30:34and
30:35a
30:35bunch
30:35of
30:35cyber
30:36security
30:37incidents
30:37happened
30:38and
30:38we
30:38fixed
30:38a
30:38bunch
30:39of
30:39stuff
30:39just
30:39in
30:40time
30:40and
30:40it
30:40was
30:40like
30:40Y2K
30:41you
30:42know
30:42it
30:42could
30:42have
30:42been
30:42this
30:43major
30:44crisis
30:44but
30:45we
30:45averted
30:46it
30:46through
30:46heroic
30:47effort
30:47and
30:48late
30:48nights
30:48at
30:49the
30:49computer
30:49terminal
30:51that's
30:51that
30:52I
30:52think
30:52is
30:52the
30:52good
30:52outcome
30:52that's
30:53a lot
30:53of
30:53close
30:54calls
30:54to
30:55endure
30:55and
30:55still
30:56be
30:56an
30:56optimist
30:57we've
30:58done
30:58it
30:58before
30:59when
31:01Y2K
31:01the
31:02Cold
31:02War
31:03we
31:04have
31:05lots
31:05of
31:06examples
31:06I
31:06mean
31:06the
31:07industrial
31:07revolution
31:08played
31:08out
31:09over
31:0975
31:10or
31:1080
31:10years
31:11but
31:12we
31:12muddled
31:13through
31:13it
31:13did
31:13not
31:14lead
31:14to
31:15human
31:16extinction
31:17Kevin
31:17there's
31:17a
31:18I
31:18don't
31:18know
31:18if
31:18it's
31:18a
31:18metaphor
31:19or
31:19an
31:19actual
31:20scientific
31:20thing
31:21but
31:21a lot
31:21of
31:21people
31:22talk
31:22about
31:22a
31:22kill
31:23switch
31:23for
31:24AI
31:24as
31:25if
31:26when
31:26things
31:26get
31:27out
31:27of
31:27hand
31:27suddenly
31:28somebody
31:28goes
31:28to
31:28a
31:29special
31:29room
31:29and
31:30turns
31:31a
31:31light
31:31switch
31:32and
31:32AI
31:32is
31:32turned
31:33off
31:33and
31:33things
31:34can
31:34settle
31:35down
31:35whatever
31:36is
31:36broken
31:37can be
31:37fixed
31:41a
31:41construct
31:42or
31:42is
31:42it
31:42a
31:42real
31:42thing
31:43it's
31:44an
31:44imaginary
31:44construct
31:45I
31:45think
31:45it
31:45comes
31:45from
31:46science
31:46fiction
31:46movies
31:47or
31:47thrillers
31:48where
31:48the red
31:50lights
31:50in the
31:51control
31:51room
31:51start
31:51blinking
31:52and
31:52the
31:52engineer
31:53sprints
31:53over to
31:53the
31:54server
31:54rack
31:54and
31:54yanks
31:55out
31:55the
31:55cord
31:55right
31:56where
31:56clips
31:56the
31:57blue
31:57wire
31:58instead
31:58of
31:58the
31:58red
31:59wire
31:59Bruce
32:00Willis
32:00always
32:01chooses
32:01correctly
32:03unfortunately
32:04that is
32:05not how
32:05modern AI
32:06systems
32:07are
32:07running
32:08they
32:08are
32:08distributed
32:09across
32:10huge
32:11numbers
32:11of
32:11servers
32:12and
32:12data
32:12centers
32:13and
32:14there's
32:14been
32:14some
32:15research
32:15into
32:15how
32:15these
32:16things
32:16can
32:16actually
32:16replicate
32:18themselves
32:18on
32:19smart
32:19devices
32:20like
32:20the
32:20smart
32:21ovens
32:21and
32:22toasters
32:22in your
32:23home
32:23and
32:23so
32:24we
32:25can't
32:25just
32:25pull
32:25the
32:25plug
32:26if
32:26it
32:26means
32:27you
32:27know
32:27it
32:27might
32:27mean
32:28taking
32:28down
32:28some
32:29huge
32:29chunk
32:30of
32:30computers
32:31across
32:31the
32:32world
32:32in a
32:32few
32:33years
32:33are
32:33we
32:33going
32:33to
32:33be
32:33at
32:3325%
32:34unemployment
32:36I
32:37hope
32:37not
32:39I
32:40hope
32:40I
32:41hope
32:41we
32:41will
32:42continue
32:42to
32:42find
32:43useful
32:43and
32:44valuable
32:44things
32:44to
32:45do
32:45but
32:46I
32:46think
32:46a lot
32:46of
32:46that
32:46depends
32:47on
32:48not
32:49just
32:49the
32:49decisions
32:50we
32:50make
32:50as
32:51a
32:51public
32:52sector
32:52as
32:53the
32:53government
32:53as
32:54regulators
32:54but
32:55as
32:56employers
32:57and
32:58business
32:59leaders
32:59I
33:00think
33:00we
33:02as
33:02recipients
33:04as
33:04users
33:05of
33:05and
33:06in
33:07some
33:07cases
33:07victims
33:08of
33:08this
33:09technology
33:09we
33:11are
33:11not
33:12powerless
33:12and
33:13impotent
33:13in the
33:14face
33:14of
33:14this
33:15change
33:15and
33:17we
33:18have
33:19seen
33:19examples
33:20before
33:21of
33:22large
33:24society
33:24altering
33:26technological
33:26transformations
33:27that have
33:29turned out
33:29great
33:30because
33:31people
33:32made it
33:33turn out
33:34great
33:34they
33:34organized
33:34they
33:35fought
33:36for
33:37human
33:38rights
33:38even
33:39as
33:39capital
33:41owners
33:41tried to
33:42dehumanize
33:43them
33:43through
33:43things
33:43like
33:44factory
33:44automation
33:47you
33:47know
33:48I am
33:48my
33:48optimism
33:49also
33:49comes
33:50David
33:50from
33:50this
33:50general
33:51belief
33:51I have
33:52that
33:52technology
33:53usually
33:53makes
33:53our
33:54lives
33:54better
33:54in
33:54the
33:55long
33:55run
33:56Kevin
33:56Roos
33:56thank
33:57you
33:57so
33:57much
33:57thank
33:58you
33:58David
33:58this
33:59was
33:59great

Recommended