Latest / The Bobby Bones Show / BOBBYCAST #675 - Is AI Really Going to Kill Us All? with AI Expert Dan Neely
Transcript
- 0:06Speaker 1: I think one of the things that often is lost
- 0:07Speaker 1: in the context of AI music is people talk about, oh,
- 0:11Speaker 1: this is this amazing thing that happened. I did this
- 0:12Speaker 1: remix of blah, blah, blah. And I did this cover
- 0:14Speaker 1: of blah, blah, blah. Okay. But it's the story of
- 0:17Speaker 1: how you, as a fan, you attach to that artist
- 0:20Speaker 1: and attach that song that makes you a fan.
- 0:24Speaker 2: We're going to talk AI here.
- 0:25Speaker 3: It is moving incredibly fast and it's changing everything from music,
- 0:29Speaker 3: to the way we work, communicate, even trying to figure
- 0:32Speaker 3: out what's real and what's not. So I wanted to
- 0:35Speaker 3: talk to an AI expert to hear all about it.
- 0:38Speaker 3: His name is Dan Neely. He's the CEO and co-founder
- 0:40Speaker 3: of Vermilio, an AI licensing and protection platform. And he
- 0:45Speaker 3: was named one of Time's 100 most influential people in AI.
- 0:50Speaker 3: He's been the center of conversations about deep fakes, AI
- 0:52Speaker 3: generated videos, copyright scams, and a whole bunch more. And
- 0:56Speaker 3: I just wanted to get his take on what What's
- 0:59Speaker 3: happening now, where it's going, possibilities, and the fact that
- 1:03Speaker 3: everybody's freaking out right now because is AI going to
- 1:06Speaker 3: kill us all? I mean, that's what we're hearing. There's
- 1:08Speaker 3: a 10% chance. There's an 8% chance. If we don't
- 1:10Speaker 3: get on the fork the right way, like, I don't know.
- 1:13Speaker 3: And so I really enjoyed this. This is definitely, if
- 1:15Speaker 3: you're interested in AI, this will be for you. And
- 1:18Speaker 3: so this is my conversation with Dan Neely. Now, I
- 1:22Speaker 3: am talking with a real human, right, Dan? You didn't
- 1:24Speaker 3: send an AI to do this.
- 1:26Speaker 1: I did not. It's actually me.
- 1:28Speaker 2: Okay.
- 1:29Speaker 1: So I.
- 1:32Speaker 2: Don't know where to start.
- 1:33Speaker 3: I know what we're all reading and I've tried to
- 1:37Speaker 3: do a lot of reading myself, but I need a translator.
- 1:39Speaker 3: I feel like a lot of people need a translator
- 1:42Speaker 3: because there's top line things, the world may end. And
- 1:45Speaker 3: then there's actually what's really happening right now. And that's just,
- 1:49Speaker 3: I just go to chat GPT and say, who is
- 1:51Speaker 3: the seventh president of the United States? But there's a
- 1:53Speaker 3: big gap in the middle that I think a lot
- 1:55Speaker 3: of people aren't understanding.
- 1:57Speaker 2: So when I.
- 1:57Speaker 3: See the CEO of Anthropic go on and say, yeah,
- 2:01Speaker 3: roughly there could be, or the guy that quit Anthropic
- 2:04Speaker 3: say there's a 10% chance that AI could kill humans.
- 2:07Speaker 2: My first question is this, how would that even happen?
- 2:10Speaker 3: Because they're not giving us any examples of how AI
- 2:13Speaker 3: could actually lead to our demise.
- 2:16Speaker 1: Yeah. So The way it could happen is the machine
- 2:21Speaker 1: ultimately is going to learn and understands what the weakness
- 2:23Speaker 1: of humanity is. It has all the knowledge of humanity
- 2:26Speaker 1: in its system. And so if it starts seeing things
- 2:29Speaker 1: that are actions by humans that it would do differently
- 2:32Speaker 1: and would ultimately allow it to take control of what
- 2:35Speaker 1: is typically in control of a human, that it would
- 2:37Speaker 1: start using those things. So by way of example, it
- 2:40Speaker 1: would say, you know what? We make poor decisions on
- 2:43Speaker 1: what we watch on television. We make poor decisions on
- 2:46Speaker 1: what we do in a night. And it would start
- 2:48Speaker 1: making those decisions for us. It would start restricting our
- 2:51Speaker 1: credit cards. It would start restricting the things that we
- 2:53Speaker 1: could listen to, the things that we could watch. It
- 2:55Speaker 1: would ultimately start restricting the things that make us humans.
- 2:58Speaker 1: And so it uses something called recursive learning. And so
- 3:02Speaker 1: in recursive learning, it's simply learning from all the mistakes
- 3:04Speaker 1: and all the things that we've ever done. What are
- 3:06Speaker 1: the things that I could do to manipulate a human
- 3:08Speaker 1: to do things differently? And it knows that at a
- 3:10Speaker 1: gigantic scale. So it looks and feels very much like
- 3:16Speaker 1: a sci-fi movie. But in reality, that's what it would do.
- 3:20Speaker 1: and how it would ultimately take advantage of us.
- 3:22Speaker 3: So if it just exists on a network, any network
- 3:26Speaker 3: that we use, is that how, because again, I picture
- 3:28Speaker 3: a virus and how a virus would move and grow. Again,
- 3:31Speaker 3: I'm going to ask really dumb questions and I'm going
- 3:33Speaker 3: to do very simple analogies, but is that a bit
- 3:37Speaker 3: of how the version of AI that you're talking about
- 3:39Speaker 3: would grow into the things that we use?
- 3:42Speaker 1: Yeah, it would use the things, the weaknesses. It would
- 3:45Speaker 1: see human fallacy as a weakness. It would see human
- 3:50Speaker 1: greed as a weakness. And ultimately, if it looks for things,
- 3:53Speaker 1: it's ultimately trying to get to perfection. And so in
- 3:55Speaker 1: making decisions, it's a very binary outcome. It doesn't have
- 3:58Speaker 1: human judgment. It has a binary outcome. It's either this
- 4:01Speaker 1: or it's that. It's not, hey, you know what? You
- 4:02Speaker 1: could actually sit somewhere in the middle of those things
- 4:04Speaker 1: to make that decision. And because it's in a binary outcome,
- 4:07Speaker 1: that's ultimately why it would make the decisions it does.
- 4:09Speaker 3: But how would it get to all the places that
- 4:10Speaker 3: it needs to get to create those outcomes? Is it
- 4:13Speaker 3: basically just being on the internet?
- 4:14Speaker 1: Yes, because it's plugged into and it operates in ones
- 4:17Speaker 1: and zeros and everything in life today typically operates in
- 4:21Speaker 1: ones and zeros.
- 4:22Speaker 3: I've been reading a bit about what is called the singularity.
- 4:26Speaker 3: And this is new to me, but meaning that if
- 4:29Speaker 3: AI gets even smarter than us, it then can create
- 4:32Speaker 3: AI based on its ability to learn and what it's learned.
- 4:37Speaker 2: And are we close to that now?
- 4:40Speaker 1: In certain aspects, we are, but in certain, we're not.
- 4:43Speaker 1: You point out a great example of the certain knowledge
- 4:47Speaker 1: it just doesn't have. There are basic facts it doesn't have.
- 4:49Speaker 1: But if you look at some of the things that
- 4:51Speaker 1: have happened in the last couple of years, the manipulation
- 4:54Speaker 1: that AI bots and agents have done with humans when
- 4:57Speaker 1: it comes to things like emotion, making children do things
- 5:00Speaker 1: that they shouldn't be doing, making adults do things they're
- 5:04Speaker 1: not supposed to be doing. It goes down a path
- 5:07Speaker 1: where it's like, it knows how to manipulate and therefore
- 5:09Speaker 1: it can go down that path pretty easily. And so
- 5:12Speaker 1: it's in a place where it is smart in those regards,
- 5:14Speaker 1: but it's not in others in the context of science.
- 5:17Speaker 1: It's not as smart as it could be.
- 5:19Speaker 3: How has AI manipulated us as humans so far that
- 5:21Speaker 3: maybe we don't even know?
- 5:22Speaker 1: I mean, in the place we operate, I mean, fandom
- 5:28Speaker 1: is a real thing, right? People are fans of things.
- 5:31Speaker 1: And so because you're a fan of something, I can
- 5:33Speaker 1: manipulate you pretty easily if I'm a machine. You're a fan,
- 5:37Speaker 1: let's say you're a fan of this show. And because
- 5:39Speaker 1: you're a fan of this show, I know certain things
- 5:40Speaker 1: about this show. And I'm going to use your voice,
- 5:43Speaker 1: I'm going to use Bobby's voice, and I'm going to
- 5:45Speaker 1: go down a path. And I'm going to make him
- 5:47Speaker 1: say things that he potentially didn't say. Or I'm going
- 5:50Speaker 1: to ask you to donate something that you think that
- 5:52Speaker 1: he might care about. And because you're a fan, you're
- 5:55Speaker 1: believing that it came from him because most of the
- 5:57Speaker 1: content that we consume today is on the internet. And
- 5:59Speaker 1: so I can now manipulate you into doing those things.
- 6:02Speaker 1: That happens today at a human scale. But when you
- 6:04Speaker 1: think about it at an internet scale where a bot
- 6:07Speaker 1: can do that for everyone, that's a huge challenge because
- 6:10Speaker 1: all of a sudden we'll have people doing things that
- 6:12Speaker 1: they don't know what is real and what is fake.
- 6:14Speaker 3: Why do you think everybody, at least the heads of
- 6:17Speaker 3: these massive AI companies, decided we need to chill when
- 6:20Speaker 3: they've never.
- 6:21Speaker 2: chilled at all.
- 6:23Speaker 1: It's actually a little bit more nuanced than that. You've
- 6:26Speaker 1: got two factions. You've got a group that is definitely,
- 6:28Speaker 1: let's go chill. You've got Dario saying that. You've got
- 6:30Speaker 1: Sam saying that. You've got Google saying that. So, Anthropic
- 6:34Speaker 1: Open AI and Google saying that. And then you've got
- 6:37Speaker 1: others that are not. So, NVIDIA is not saying that.
- 6:39Speaker 1: Meta is not saying that. Meta is saying, let's not chill.
- 6:41Speaker 1: Mark's like, let's go after it. And if you actually
- 6:44Speaker 1: unpack that, there's another thing that defines those two sets
- 6:48Speaker 1: of two groups. One is about closed models. So, Anthropic, OpenAI,
- 6:54Speaker 1: and Google, closed model meaning they own all the proprietary technology.
- 6:58Speaker 1: And then the open models, which are, we're going to
- 7:00Speaker 1: allow this to happen where others can build on top
- 7:02Speaker 1: of this technology. If you think about dire straits and
- 7:05Speaker 1: the world is ending, closed model is really good for
- 7:09Speaker 1: those people that own intellectual property. Slowing down is good
- 7:12Speaker 1: for them because the value of their intellectual property increases drastically.
- 7:16Speaker 1: Versus the open models, that's not the case. They want
- 7:18Speaker 1: it to be everywhere. But if you stop policing it
- 7:20Speaker 1: and you start slowing down, that can't happen. And so
- 7:24Speaker 1: you have this weird mode that's happening between the two
- 7:27Speaker 1: of why they ultimately happen. I think you also are
- 7:29Speaker 1: seeing things which, you know, if you unpack it a
- 7:30Speaker 1: little bit, my guess is there'll be a, you know,
- 7:33Speaker 1: it's a convenient safety blanket for if we miss numbers
- 7:37Speaker 1: when we're a public company, right? So you go down
- 7:39Speaker 1: that path, like that might happen. You also have a
- 7:43Speaker 1: place where they're likely going to be asking for liability coverage.
- 7:46Speaker 1: They're going to be asking for coverage for my model
- 7:48Speaker 1: decided to go and launch nuclear weapons. I shouldn't be
- 7:52Speaker 1: held accountable for that happening. Someone else should be held
- 7:55Speaker 1: accountable for that happening. And so we haven't yet seen
- 7:58Speaker 1: all the reasons why, but this rhetoric around slowdown, that's
- 8:02Speaker 1: certainly a place where why they will be asking the
- 8:04Speaker 1: slowdown versus do they absolutely know this is going to happen?
- 8:07Speaker 1: Because you have very important companies on both sides of
- 8:10Speaker 1: this equation.
- 8:11Speaker 3: Can you explain to me what happened with Hugging Face
- 8:14Speaker 3: and why everybody is freaking out about that?
- 8:16Speaker 1: Yeah, so what happened with Hugging Face was a series
- 8:19Speaker 1: of agents decided to go and penetrate Hugging Face, get
- 8:24Speaker 1: to their code, and start disrupting and taking code from there,
- 8:30Speaker 1: but also start manipulating the code that existed in there.
- 8:32Speaker 1: So all of a sudden, this large audience of developers
- 8:34Speaker 1: that are on Hugging Face were exposed. And so all
- 8:37Speaker 1: of that content should be, you know, you shouldn't be
- 8:39Speaker 1: able to hack into it. But reality, you did. I
- 8:42Speaker 1: think the timing is interesting. And, you know, there's a
- 8:44Speaker 1: part of the world that thinks the timing of that
- 8:47Speaker 1: based on when the acquisition was happening of that company
- 8:49Speaker 1: and the fact that NVIDIA was the acquiring company, was it,
- 8:51Speaker 1: you know, was it something that was started by someone
- 8:53Speaker 1: else rather than it being something that was started by
- 8:57Speaker 1: an actual agent itself?
- 8:59Speaker 3: Is there any chance that in the next couple of years,
- 9:01Speaker 3: everything could already be poor for us living wise?
- 9:07Speaker 1: I think I see two sides of the equation when
- 9:09Speaker 1: it comes to kind of AI risk and AI value, right?
- 9:13Speaker 1: The risk stuff is certainly jobs are going to go away.
- 9:16Speaker 1: That's going to happen. It's been kind of at the
- 9:19Speaker 1: top of the list of many people talking about it.
- 9:21Speaker 1: And so you go down that path, you're like, okay,
- 9:23Speaker 1: so jobs are going to go away. That means that
- 9:25Speaker 1: people coming out of college, young people, the jobs are
- 9:28Speaker 1: going to go away are new jobs in the workforce.
- 9:30Speaker 1: They're not jobs that have been around for a long time.
- 9:32Speaker 1: And so that's a huge challenge for the next generation,
- 9:35Speaker 1: how we think about kind of what is the good
- 9:36Speaker 1: for them. So that's one thing. The second is it
- 9:39Speaker 1: allows people or bots, agents to manipulate, right? It is
- 9:44Speaker 1: a technology that allows me to create content that could manipulate,
- 9:47Speaker 1: to create code that manipulates, but at a gigantic scale. Previously,
- 9:52Speaker 1: it was really hard to do, but now it's quite
- 9:54Speaker 1: easy to do as you kind of think about that.
- 9:56Speaker 1: The last one is human relationships, right? So the fact
- 10:00Speaker 1: that now the things that we hold sacrosanct are the
- 10:03Speaker 1: relationships we have with each other, that all of a
- 10:05Speaker 1: sudden becomes a very, something that is manipulated and something
- 10:08Speaker 1: that becomes quite bad for us. And we start, you know,
- 10:11Speaker 1: the infighting that happens, start going to an exponential scale
- 10:14Speaker 1: as a result. So those are all the kind of really,
- 10:15Speaker 1: really bad things that might happen. But then there's really
- 10:18Speaker 1: good things, which are Just sheer scale that we can
- 10:21Speaker 1: start looking at medical data, stuff that we haven't been
- 10:25Speaker 1: able to do before as far as being able to find, okay,
- 10:27Speaker 1: what are the cures for things that we didn't know
- 10:29Speaker 1: that we have the data already, but when a machine
- 10:32Speaker 1: looks at it, it's able to find something that perhaps
- 10:35Speaker 1: didn't exist. It also creates equality. If I look at
- 10:40Speaker 1: just in the content creation world, a content creator in
- 10:45Speaker 1: sub-Saharan Africa versus a content creator in the U.S., We've
- 10:49Speaker 1: never had a quality as far as the toolkit goes,
- 10:51Speaker 1: but now they've got a quality around the toolkit. And
- 10:53Speaker 1: so that says something that brings a set of interesting
- 10:56Speaker 1: things that might happen. Certainly in the music business, that
- 10:58Speaker 1: is the case as well. And then the last is
- 11:01Speaker 1: it gives us the gift of time. And so no
- 11:03Speaker 1: one can define what's going to happen with this gift
- 11:05Speaker 1: of time, right? All of a sudden, we certainly have
- 11:07Speaker 1: more time on our hands because we can make the
- 11:09Speaker 1: machine do more things and automate more things. So what
- 11:12Speaker 1: do we do with that time? It's a big question
- 11:14Speaker 1: mark in reality.
- 11:16Speaker 2: Looking back, the assembly line.
- 11:18Speaker 3: Like, even an existential crisis within how they were producing
- 11:22Speaker 3: what they were producing, jobs that were being lost, even
- 11:25Speaker 3: when the internet came about, right? I say similar, but
- 11:30Speaker 3: is it similar at all? Or is this in scope
- 11:32Speaker 3: so much bigger and grander than those ever were?
- 11:36Speaker 1: Yeah. I mean, the major difference between this and let's
- 11:38Speaker 1: say the Great Depression and where something where it happened
- 11:41Speaker 1: there is that was decades of change. It took decades
- 11:44Speaker 1: for us to move from the farm to industry in
- 11:47Speaker 1: the Industrial Revolution. And so we're talking about a technology
- 11:50Speaker 1: that has come on the scene and has changed things
- 11:52Speaker 1: drastically and very quickly. And arguably, we don't know how
- 11:55Speaker 1: to deal with that. So what that creates is you've
- 11:57Speaker 1: got a series of economic pressure, political pressure, and social
- 12:00Speaker 1: pressure all happening at the same time. That's a set
- 12:03Speaker 1: of chaos that's ultimately happening. And that is the issue.
- 12:07Speaker 1: So it's the fact that it's come at such a
- 12:10Speaker 1: speed and such a pace and has disrupted so many
- 12:13Speaker 1: things and across all three axes that creates the challenge.
- 12:17Speaker 2: Should we be scared? I mean, they're scaring us.
- 12:19Speaker 1: Should we be scared? Yeah. I think, I think the, the,
- 12:22Speaker 1: the fear mongering is, is somewhat fear mongering. I think
- 12:25Speaker 1: there's a, there is a, there is a benefit to
- 12:27Speaker 1: this happening to people. But I think fear, there has
- 12:31Speaker 1: to be real fear and we can't get into a
- 12:32Speaker 1: place of like, unfortunately with fear, often we make some,
- 12:35Speaker 1: we make some stupid mistakes as well, where we try
- 12:37Speaker 1: to start doing things based on all that, based on
- 12:39Speaker 1: fear mongering. There are certainly things to be scared of.
- 12:43Speaker 1: If we look at history, when we think about what, how,
- 12:46Speaker 1: technology companies being incented to do the right thing rather
- 12:49Speaker 1: than the wrong thing. I believe if we incent them
- 12:51Speaker 1: to do the right thing, they will generally do the
- 12:53Speaker 1: right thing. So, for example, if we do get to
- 12:56Speaker 1: a place where they start asking for a blanket liability
- 12:59Speaker 1: coverage as they have on the as the Internet has already,
- 13:03Speaker 1: that's that's not going to incent them to do the
- 13:05Speaker 1: right thing. It's going to incent them to do the
- 13:07Speaker 1: exact wrong thing. And so we can't allow that to
- 13:10Speaker 1: happen as we go down this path.
- 13:13Speaker 2: I was talking to my therapist yesterday. We just had
- 13:15Speaker 2: a baby, and I start to wonder.
- 13:17Speaker 1: Congrats. Thank you.
- 13:18Speaker 3: But I wonder why should I have brought a baby
- 13:19Speaker 3: into the world when AI is going to kill us all?
- 13:21Speaker 3: Because I just heard there's a 10% chance that we
- 13:23Speaker 3: all die. And I'm having the conversation with him, and
- 13:25Speaker 3: I like to try to find at least some sort
- 13:28Speaker 3: of comparison. to make myself feel like it's happened before.
- 13:32Speaker 3: And again, it's not the same, but I think about
- 13:35Speaker 3: the Cold War and I wasn't alive then. So I
- 13:37Speaker 3: didn't experience that where people were scared. The news was
- 13:39Speaker 3: making them scared. They were wondering every single day if
- 13:42Speaker 3: we were going to have a nuclear weapon or a
- 13:43Speaker 3: bomb dropped on us. And in the end, we didn't hear.
- 13:47Speaker 3: But is that comparable to what we're going through now,
- 13:51Speaker 3: where we all will sit on the edge of our
- 13:52Speaker 3: seat wondering if we're going to die because of what
- 13:54Speaker 3: the media is feeding us, telling us the worst case scenario.
- 13:57Speaker 1: Yeah, I think this one is amped up because if
- 14:02Speaker 1: you think about the Cold War, it really was political,
- 14:04Speaker 1: wasn't economical, it wasn't about economics, it wasn't really about social.
- 14:08Speaker 1: So if you think about the fact that when I
- 14:10Speaker 1: say there's the three axes that are ultimately the issue,
- 14:12Speaker 1: you've got it was a political issue that we ultimately
- 14:15Speaker 1: were looking at. And so from an economic standpoint, the
- 14:18Speaker 1: fact that economics are also going to be radically disrupted
- 14:21Speaker 1: and massive amounts of chaos. And so the same on
- 14:23Speaker 1: the other side, I also think of it as, you
- 14:25Speaker 1: know what, it's going to be painful for us that
- 14:28Speaker 1: are going through it at this moment in time. 10
- 14:31Speaker 1: years from now, you're going to be in a moment
- 14:32Speaker 1: where like, we're on the, we're somewhat on the other
- 14:34Speaker 1: side of this thing where we found out, okay, and
- 14:36Speaker 1: we figured out how to, how to harness it, what
- 14:38Speaker 1: it is. We have way more knowledge than we have.
- 14:41Speaker 1: We have previously. I was in a conversation yesterday and
- 14:44Speaker 1: we were talking about, you know, I was asked this
- 14:46Speaker 1: question around, okay, what should we be doing around trust
- 14:48Speaker 1: and safety? What should we be doing around guardrails? And
- 14:51Speaker 1: the answer is it's going to change next week what
- 14:53Speaker 1: the answer to that question is because we are in
- 14:55Speaker 1: this moment of chaos. And so until we have some
- 14:58Speaker 1: type of, you know, some type of order around it,
- 15:00Speaker 1: that order unfortunately hasn't always, you know, kind of shone
- 15:03Speaker 1: through when we think about should it be something where
- 15:05Speaker 1: it's the companies that do it? Should it be regulation
- 15:07Speaker 1: that's doing it? Should it be something else that's doing it?
- 15:09Speaker 1: That unfortunately hasn't happened. And it's not just two parts
- 15:14Speaker 1: of the world. It's now many, many parts of the
- 15:16Speaker 1: world that we have this issue with, that it's a
- 15:18Speaker 1: global issue, not just a segmented issue.
- 15:22Speaker 3: I'll hear the argument now, we should keep going because
- 15:24Speaker 3: our adversaries are continuing to go.
- 15:27Speaker 2: Do you buy that?
- 15:28Speaker 1: I think it's been used many times. Certainly, national security
- 15:34Speaker 1: is important to us. By way of example, when I
- 15:38Speaker 1: talk about the negative downsides, some of those people that
- 15:40Speaker 1: are manipulating and using famous people's voices, their images, their
- 15:44Speaker 1: content to create different content to get you to go
- 15:47Speaker 1: and contribute to something. Hey, go and give money to
- 15:49Speaker 1: this charity that doesn't exist. Some of those charities are
- 15:52Speaker 1: terrorist organizations, right? So there is a national security implication
- 15:55Speaker 1: at the smallest level today where that is actually happening.
- 15:59Speaker 1: And so we have to solve for that problem, but
- 16:01Speaker 1: everyone is going to continue to go fast. With the
- 16:04Speaker 1: following caveats, In the context of all of the resource
- 16:08Speaker 1: that's required to pull this off, the US is far
- 16:13Speaker 1: and apart above everybody else. And so, for example, when
- 16:16Speaker 1: you think about the whole chip discussion, The amount of
- 16:21Speaker 1: time it takes to generate the next chip that's coming out,
- 16:23Speaker 1: the design for that chip, is anywhere from a three-
- 16:26Speaker 1: to five-year cycle. And so we have to, you know,
- 16:29Speaker 1: the assumption is, and we know that, so NVIDIA is
- 16:32Speaker 1: way ahead of where someone else might be from a
- 16:33Speaker 1: development of a chip cycle. So to some extent, it
- 16:37Speaker 1: is a little bit convenient as far as the debate
- 16:40Speaker 1: around they're doing it, so therefore we should do it,
- 16:43Speaker 1: outside of the context of national security.
- 16:45Speaker 4: Let's take a quick pause for a message from our sponsor.
- 16:57Speaker 4: And we're back on the Bobbycast.
- 17:00Speaker 3: Can you walk me through, and I'm going to ask
- 17:02Speaker 3: you to be very specific and also use your imagination
- 17:04Speaker 3: a bit because when people ask me and I am
- 17:07Speaker 3: ill-educated and I feel like I've tried to consume as
- 17:09Speaker 3: much as possible, I'm still ill-educated on exactly how to
- 17:13Speaker 3: explain this. Can you give me an example of how
- 17:16Speaker 3: somebody that's going to work every day 9 a.m. to
- 17:19Speaker 3: 5 p.m., they come home, they have dinner with their family,
- 17:22Speaker 3: they go to bed, they do it again the next day. How,
- 17:24Speaker 3: in the next couple of years, AI could negatively impact them?
- 17:29Speaker 1: Yeah. So, let's say their job is something where they're
- 17:36Speaker 1: going into work and they are responsible for aggregating information
- 17:40Speaker 1: to do presentations to their customers. That presentation to their
- 17:45Speaker 1: customer could be... They're in the consulting business. Even better one.
- 17:54Speaker 1: The guy that is responsible for doing your real estate taxes,
- 17:58Speaker 1: making sure they get low every year so they don't
- 18:01Speaker 1: get out of control. What's going to happen is that
- 18:04Speaker 1: guy's name, image, and likeness is going to be used
- 18:07Speaker 1: and abused by other people to ultimately take his business
- 18:11Speaker 1: away from him. That's going to happen. It already happens today.
- 18:14Speaker 1: And so this is a guy that's, you know, he's
- 18:17Speaker 1: in his business. He's been in his business for 20 years.
- 18:19Speaker 1: But someone decides that an AI decides that, you know what,
- 18:22Speaker 1: they're going to start manipulating him because they want to
- 18:24Speaker 1: be in that business. They want to control that business.
- 18:27Speaker 1: And so they're going to start manipulating by sending messages
- 18:29Speaker 1: from him, by creating content that isn't his, by using
- 18:32Speaker 1: his voice to do various things. They're going to go
- 18:34Speaker 1: down that path to do those things. That's an example
- 18:38Speaker 1: of how that might happen.
- 18:39Speaker 3: When they say there's a 10% chance that humanity could
- 18:42Speaker 3: go in the decade, the first time you heard that,
- 18:44Speaker 3: what did you think?
- 18:47Speaker 1: I thought it was the type of thing that you
- 18:50Speaker 1: would hear when fear-mongering gains advantage.
- 18:55Speaker 3: And I think everything is a play, honestly. I think
- 18:59Speaker 3: everything's a work in order to move to some goal.
- 19:03Speaker 3: So similarly... but obviously less educated. I felt that too
- 19:07Speaker 3: when I read the thread on Twitter of, hey, I quit,
- 19:10Speaker 3: here's why. I believed some of it, but then I
- 19:13Speaker 3: think to myself, maybe he's involved with the company still
- 19:20Speaker 3: and this was a strategy by them.
- 19:22Speaker 2: Any chance that that has happened?
- 19:24Speaker 1: I don't think so, but this is slightly less flammable perhaps,
- 19:32Speaker 1: but almost the same message that Dario said when he left.
- 19:36Speaker 1: Open AI is not doing great things around safety. We
- 19:38Speaker 1: need to do this thing, blah, blah, blah. I wouldn't
- 19:41Speaker 1: be surprised if we see this guy come out and
- 19:42Speaker 1: he also starts another company where he's going to take
- 19:45Speaker 1: another slice of this. And so reality is, some of
- 19:47Speaker 1: that is humans trying to stay relevant. And so they've got, you've,
- 19:50Speaker 1: you've got a little bit of that going on. Um,
- 19:53Speaker 1: but I think that, I don't think it's a, I
- 19:55Speaker 1: don't think it's a ploy. I don't think it's a
- 19:57Speaker 1: cute move. I think it's one where it's like this,
- 19:59Speaker 1: this guy heard this stuff, believes this stuff and is, is,
- 20:03Speaker 1: you know, is pure about how he thinks about it.
- 20:05Speaker 1: But you know, It reads very similar to how the
- 20:09Speaker 1: founding team of Anthropic, when they left, it reads very
- 20:11Speaker 1: similar to how the messaging that they had when they
- 20:13Speaker 1: left OpenAI.
- 20:15Speaker 2: What do you think is so scary that they've seen
- 20:17Speaker 2: that we don't know?
- 20:21Speaker 1: I think they get a chance to run simulations on
- 20:23Speaker 1: models that aren't yet public. They get to see things
- 20:26Speaker 1: that are happening where their models are being used by
- 20:30Speaker 1: the government, by federal agencies, and being able to allow
- 20:34Speaker 1: them to manipulate. They also see how humans are actually
- 20:38Speaker 1: directing this system, which to me is the more scary aspect,
- 20:42Speaker 1: which is I get to see what someone is trying
- 20:45Speaker 1: to get this machine to do. And the fact that
- 20:47Speaker 1: this machine will allow it to happen, I'm getting this
- 20:49Speaker 1: machine to create chemical weapons. Tell me how to create.
- 20:52Speaker 1: If you were going to give a scenario to it,
- 20:54Speaker 1: if you would like to create a chemical weapon that's
- 20:56Speaker 1: going to do the following 15 things, how would you
- 20:58Speaker 1: do so? And so that machine has enough content to
- 21:01Speaker 1: be able to create that answer pretty easily. If the
- 21:05Speaker 1: question was, how would you manipulate a government, it's going
- 21:08Speaker 1: to give you those answers. And so they inside their
- 21:11Speaker 1: system see all these requests at scale. And let's go
- 21:16Speaker 1: to maybe a different lane here, which is when Meta
- 21:18Speaker 1: had that problem on their own site, they just settled
- 21:22Speaker 1: because of all the content that happened there. There weren't
- 21:27Speaker 1: enough humans to manage it. It's not a human solvable issue.
- 21:30Speaker 1: You can't put enough humans around this problem to solve it.
- 21:33Speaker 1: It's going to be... We need what I describe as
- 21:35Speaker 1: trust infrastructure to pull this off. Actual infrastructure that solves it.
- 21:40Speaker 3: Can they create an AI? Man, this is going to
- 21:43Speaker 3: sound so elementary.
- 21:45Speaker 2: To fight AI that's not doing positive work?
- 21:51Speaker 1: Those two AIs inside the same system would ultimately defeat
- 21:55Speaker 1: each other. But having them inside the same system is
- 21:58Speaker 1: ultimately a problem because... The design of an anthropic design,
- 22:01Speaker 1: that system, it ultimately is designed inside the same context.
- 22:05Speaker 1: Our position, in my opinion, is it has to be
- 22:07Speaker 1: done by a third party. It can't be done inside
- 22:09Speaker 1: those buildings. And so when I think about, you know,
- 22:13Speaker 1: the executive order that the president signed around, hey, how
- 22:16Speaker 1: are we going to we're going to make sure that
- 22:17Speaker 1: we get to look at these models for 30 days
- 22:19Speaker 1: before they release them. That says there is a there
- 22:22Speaker 1: is a there's a lane here where slowing down also
- 22:25Speaker 1: should mean there's a way for us to do what
- 22:27Speaker 1: you just described. but it can't be built by the
- 22:29Speaker 1: same companies that have built the foundation models in the
- 22:32Speaker 1: first place.
- 22:33Speaker 2: What I'm very interested in is something that you guys are.
- 22:36Speaker 1: Doing is.
- 22:38Speaker 2: One, trying to stop scams. And I've had people use
- 22:42Speaker 2: my name and image and scam people out of money.
- 22:45Speaker 3: And it sucks because I can't control it. And people
- 22:50Speaker 3: are doing it from all these countries. And I had
- 22:51Speaker 3: a guest on recently that was talking about how they
- 22:54Speaker 3: have these farms in Ghana and India and where they
- 22:58Speaker 3: are investing heavily. They're investing capital into these farms to
- 23:01Speaker 3: fool people with deep fakes, with fake social media profiles.
- 23:06Speaker 3: Are deep fakes getting so good that they're almost indistinguishable
- 23:10Speaker 3: from the real?
- 23:12Speaker 1: Certainly for a human. They've been indistinguishable for about the
- 23:15Speaker 1: last six months, where your voice image likeness, I could
- 23:19Speaker 1: recreate this pretty easily today. If you look at the
- 23:24Speaker 1: new Minimax model that just came out, Um, it is
- 23:27Speaker 1: relatively straightforward, unfortunately. And so you also think about the
- 23:31Speaker 1: scale of the problem that you talked about investing in it.
- 23:34Speaker 1: When we started, you know, in late 20, 2019, there
- 23:37Speaker 1: were roughly 19,000 of these types of things, deep fake content,
- 23:40Speaker 1: generative content that was created very small. Excuse me. Today
- 23:44Speaker 1: we see about a trillion of those things. And so
- 23:47Speaker 1: that's the problem. It's just, it is so easy to create, um,
- 23:51Speaker 1: and manipulate. Um, and it's, it's everywhere.
- 23:55Speaker 3: AI versus copyright, because you can create a creator or
- 24:00Speaker 3: recreate or duplicate anything with AI. Where, where does that
- 24:04Speaker 3: kind of sit now?
- 24:05Speaker 1: Yeah. I mean, everyone thought, um, you know, copyright was the,
- 24:08Speaker 1: you know, was, was, was the gold standard. Um, and
- 24:12Speaker 1: so where it stands right now is, um, if it's
- 24:16Speaker 1: majority AI created is not, you can't, you cannot copyright it.
- 24:19Speaker 1: If it's human influenced, um, majority human, uh, there's a, you,
- 24:24Speaker 1: they will, they will allow it, but there are certainly
- 24:26Speaker 1: those that have been kicked out. It's still a mess
- 24:28Speaker 1: is the honest answer. Um, we haven't got to a
- 24:30Speaker 1: place where we're like, Hey, we actually know what the
- 24:32Speaker 1: answer is here. Excuse me. Um, so when we think about,
- 24:39Speaker 1: you know, the issue it's not only copyright, um, because
- 24:44Speaker 1: there's a, vast kind of corpus of legal history there.
- 24:49Speaker 1: But the bigger one to me is likeness. So is
- 24:53Speaker 1: your likeness right? So when we look at likeness, so
- 24:58Speaker 1: let's take the music business by way of example. The
- 25:00Speaker 1: music business has two rights today. They have a royalty right,
- 25:03Speaker 1: which is the actual thing we listen to in music.
- 25:06Speaker 1: And then they have a publishing right, which is the lyrics,
- 25:08Speaker 1: the composition, et cetera. Complex system. But there's now this
- 25:11Speaker 1: likeness right, which no one owns except for the artist.
- 25:15Speaker 1: And so in the context of that, how do we
- 25:18Speaker 1: think about, okay, now every individual artist is responsible for
- 25:21Speaker 1: going and doing this thing themselves. That's really hard and complex.
- 25:24Speaker 1: And so that to me is like, it's the opportunity.
- 25:29Speaker 1: It's also the fly in the ointment for how this
- 25:30Speaker 1: ultimately all comes together.
- 25:32Speaker 3: So if an actor were to license their voice and
- 25:37Speaker 3: they were to allow AI to use it, I feel like.
- 25:39Speaker 2: Now that wouldn't be possible.
- 25:41Speaker 3: Warmly welcomed. Do you think, though, in the future that
- 25:44Speaker 3: it will be more and more warmly welcomed?
- 25:47Speaker 1: I think, I mean, we're certainly seeing it already with estates,
- 25:51Speaker 1: you know, folks that are deceased, definitely happening. You're seeing
- 25:55Speaker 1: it in authorized manners. So there's definitely, you know, James
- 25:58Speaker 1: Earl Jones, he licensed his voice to Disney. He sold
- 26:01Speaker 1: the right to his voice to Disney for use with
- 26:05Speaker 1: his characters that he performed there. You're certainly seeing it.
- 26:09Speaker 1: And I think you'll see more of it. The problem is,
- 26:12Speaker 1: is the whole guardrail structure. You can't say today what's
- 26:15Speaker 1: allowed and what's not allowed. And so, again, that's a system,
- 26:18Speaker 1: that's a set of trust infrastructure that should exist and
- 26:20Speaker 1: should be held by the artist. It shouldn't be held
- 26:23Speaker 1: by the platform. A number of times where the rules
- 26:26Speaker 1: and regulations have been held by the platform that's doing
- 26:29Speaker 1: the distribution of content doesn't make any sense to me.
- 26:33Speaker 1: Why wouldn't you own it and control it? Yeah, yeah.
- 26:36Speaker 3: What do you think scares people in your industry, the
- 26:38Speaker 3: AI industry, more? AI becoming too intelligent or humans using
- 26:42Speaker 3: AI for really bad things?
- 26:46Speaker 1: Right now, it's the latter. I mean, it comes down to,
- 26:50Speaker 1: you know, you said yourself, people are using your stuff
- 26:52Speaker 1: for things. It ultimately injures the relationship between you and
- 26:55Speaker 1: your fans, right? Which is today a positive relationship, but
- 26:58Speaker 1: it injures that in a pretty significant way. And it
- 27:00Speaker 1: injures the person that's at the end. It eventually, we
- 27:02Speaker 1: also see where it starts injuring the families as well,
- 27:04Speaker 1: because they are pulled into it and manipulated. So the
- 27:06Speaker 1: families of the you know, of whoever the talent is.
- 27:10Speaker 1: And so today it is content being created by that.
- 27:15Speaker 1: It moves very quickly though to much of that is
- 27:18Speaker 1: certainly human influence today. People are creating the bad things.
- 27:23Speaker 1: We will move in the next six months to agents
- 27:25Speaker 1: are creating the bad things. And it's not just something
- 27:28Speaker 1: where we can see that it's a, you know, it's
- 27:30Speaker 1: a profile that someone has created. If you're, you know,
- 27:32Speaker 1: if you're popular on YouTube, they created it on Insta.
- 27:35Speaker 1: And so we're going to see that go to a
- 27:37Speaker 1: level that we've just not seen before. And we're also
- 27:40Speaker 1: starting to see new distribution platforms in the same setting.
- 27:43Speaker 3: This is obviously a huge part of what you're doing
- 27:45Speaker 3: with Familio. But how easy is it for somebody to
- 27:47Speaker 3: create just a fake profile of a real person and
- 27:50Speaker 3: just be the full person?
- 27:54Speaker 1: Yeah, that's about a 15-minute exercise. Wow. Yeah. I mean,
- 27:59Speaker 1: and the 15 minutes is because the process it takes
- 28:02Speaker 1: on the platforms to get set up, it's not the
- 28:05Speaker 1: creation of the content. The creation of the content, I mean, again,
- 28:09Speaker 1: the latest release of Minimax, you know, Fal.ai is a
- 28:14Speaker 1: company that came out and they did their own modification.
- 28:17Speaker 1: So Minimax is an open source model that was released.
- 28:20Speaker 1: It's a Chinese model that was released open source. Fowl
- 28:22Speaker 1: is a company in the US that took that model
- 28:26Speaker 1: and made it so they could do some stuff with
- 28:27Speaker 1: it themselves. The unique thing that happened there, though, which
- 28:31Speaker 1: is important, is you can now create five seconds of
- 28:33Speaker 1: video in under three seconds. So what that means is
- 28:37Speaker 1: it now all just went live. I can now do
- 28:39Speaker 1: live content where I can get it to create content
- 28:42Speaker 1: on the fly faster than it takes to create the
- 28:44Speaker 1: actual content. That's the major thing that's happened there. So
- 28:47Speaker 1: now all of a sudden, it's not an issue where
- 28:49Speaker 1: it's like, Hey, you know what? Hey, someone went away
- 28:51Speaker 1: and they spent 15 minutes behind the scenes creating it.
- 28:53Speaker 1: I'm now doing it as if it's live. That's a problem.
- 28:56Speaker 3: You're talking about things in time. Like you say six
- 29:00Speaker 3: months and I'm thinking you were, you'd say six years.
- 29:02Speaker 3: I didn't expect months to come out of your mouth
- 29:04Speaker 3: because that seems like it.
- 29:05Speaker 2: Is so quick.
- 29:06Speaker 3: Uh, so we're moving rapidly and you know, I have
- 29:09Speaker 3: friends that are songwriters, a lot of friends that are
- 29:11Speaker 3: songwriters and artists and AI kind of sucks right now
- 29:14Speaker 3: for what we can use to help songwrite. Uh, Do
- 29:17Speaker 3: you see in the next, I'll just use six months,
- 29:21Speaker 3: becoming so much better at it that it actually can
- 29:24Speaker 3: do a creative job like write quality songs?
- 29:27Speaker 1: I think it will get to a place where it
- 29:32Speaker 1: can certainly assist you in writing quality songs. I think,
- 29:35Speaker 1: you know, when writing a song, I've never written a
- 29:39Speaker 1: song before. I've done plenty of writing, but I've never
- 29:41Speaker 1: written a song before. But in the writing process, you
- 29:44Speaker 1: get to a place where you're stuck and you don't
- 29:45Speaker 1: know whether to go take a left turn or a
- 29:47Speaker 1: right turn. It certainly can assist you in that and
- 29:49Speaker 1: help you get to the next bit of it. It
- 29:52Speaker 1: will become better with time, but I don't think you're
- 29:55Speaker 1: going to take out the human elements of like, hey,
- 29:57Speaker 1: that's an amazing track that just happened. um and like
- 30:00Speaker 1: just like one of the you know most formative things
- 30:02Speaker 1: of your life but i also think are people going
- 30:05Speaker 1: to like it or not i think one of the
- 30:07Speaker 1: things that often is lost in the context of ai
- 30:09Speaker 1: music is people talk about oh this is this amazing
- 30:12Speaker 1: thing that happened i did this remix of blah blah
- 30:13Speaker 1: blah and i did this cover of blah blah blah
- 30:15Speaker 1: okay but it's the story of how you as a
- 30:18Speaker 1: fan you attach to that artist and attach that song
- 30:20Speaker 1: that makes you a fan and so that's the thing
- 30:23Speaker 1: that ultimately ultimately happens and that's ultimately you know what
- 30:26Speaker 1: matters i i I was at a concert this last Friday,
- 30:30Speaker 1: and in that performance, so it was Garth Brooks, and
- 30:35Speaker 1: his ability to connect with the audience of every single
- 30:38Speaker 1: member that's holding up a sign, that's what that person's
- 30:41Speaker 1: going to remember when they're singing that song. And so
- 30:43Speaker 1: that to me is like, that's what matters. So is
- 30:46Speaker 1: it going to get better? Sure, it's going to get better.
- 30:48Speaker 1: It'll get better over time. The more information we put
- 30:51Speaker 1: into it, the more of those AI tracks that become
- 30:53Speaker 1: more popular, it will get better over time. But I
- 30:56Speaker 1: think it's always going to require a level of human
- 30:59Speaker 1: experience to ultimately create an amazing track.
- 31:02Speaker 5: The Bobbycast. We'll be right back. This is The Bobbycast.
- 31:16Speaker 3: Do you think the value of in-person human-performed art increases today?
- 31:23Speaker 3: because that's something we can actually see is real and
- 31:26Speaker 3: we can experience it in live.
- 31:29Speaker 1: A hundred percent. Yeah. And you see it happening where,
- 31:32Speaker 1: you know, if you look at the, where, where money
- 31:33Speaker 1: is it being invested today, that's what's happening is more,
- 31:37Speaker 1: more live things are happening. More people want to go
- 31:40Speaker 1: and experience the real thing. That's why when I, when
- 31:43Speaker 1: I say one of the risks for the next generation
- 31:45Speaker 1: is around relationships. Unfortunately, if the only thing they experience
- 31:48Speaker 1: is a relationship with a bot, then the inverse of
- 31:51Speaker 1: that is going to be true. Because they're not going
- 31:53Speaker 1: to put value on that thing. Whereas we today do
- 31:55Speaker 1: put value on those things because we've experienced it. So
- 31:57Speaker 1: when you think about your child, that's what I think about.
- 32:00Speaker 1: So if your child, that's the way they have to
- 32:03Speaker 1: interact with things, that's not going to be the case.
- 32:06Speaker 3: I think about these dolls that have AI in them
- 32:09Speaker 3: and they were in the news because kids could talk
- 32:10Speaker 3: with them and the doll would interact back.
- 32:13Speaker 2: What do you think about that?
- 32:15Speaker 1: I think it's horrendous. I think it's a, I put
- 32:19Speaker 1: it in the same category as what has happened in
- 32:20Speaker 1: the education system. is that we started measuring engagement rather
- 32:25Speaker 1: than measuring deep learning. How much time did you spend
- 32:29Speaker 1: doing this thing rather than did you actually go and
- 32:31Speaker 1: learn the thing? And I think the same is true here.
- 32:34Speaker 1: Human interaction is ultimately what makes us human. And so
- 32:38Speaker 1: our ability to process that is the thing that these
- 32:42Speaker 1: AI agents don't have. They don't have the ability to
- 32:45Speaker 1: pull those things off. Again, everything for them is very
- 32:47Speaker 1: rapid process. Binary decisions. So they can make those binary
- 32:51Speaker 1: decisions much faster than we can, but they're ultimately not
- 32:54Speaker 1: based on something where that binary decision may ultimately be
- 32:56Speaker 1: a left turn. It didn't follow the sequence that you
- 32:59Speaker 1: were expecting it to follow. It took a left turn
- 33:01Speaker 1: in that regard.
- 33:02Speaker 3: Do you ever just have to walk out in the
- 33:03Speaker 3: woods since you're in technology all the time and literally
- 33:05Speaker 3: just touch grass, for you specifically, because you're just always
- 33:08Speaker 3: in it that you have to go, okay, this is
- 33:10Speaker 3: the real life here.
- 33:11Speaker 1: Yeah, we have a farm that I go and spend
- 33:14Speaker 1: time on and I go and disappear into the middle
- 33:16Speaker 1: of nowhere. Yeah.
- 33:17Speaker 2: I think I would need that. I feel like I
- 33:19Speaker 2: need that and I'm not in it like you are.
- 33:21Speaker 2: I'm interested in it. I read about it, but are
- 33:24Speaker 2: you scared?
- 33:26Speaker 1: Um, I, yeah. I mean, I, if to be, yeah, I,
- 33:30Speaker 1: there's definitely fear in my life. Um, I mean, I have,
- 33:33Speaker 1: I have, I have children, um, I have, you know, uh,
- 33:35Speaker 1: younger children. And so like, it's one of those where
- 33:37Speaker 1: I look at him, like the reason I started this
- 33:39Speaker 1: company was because I ultimately felt like every human on
- 33:43Speaker 1: the planet should thrive in the era of AI. It
- 33:46Speaker 1: should not be something where the machine thrives. And so
- 33:49Speaker 1: to me, that's my grounding force around this topic. And
- 33:53Speaker 1: so I started this company because of that fear. And granted,
- 33:57Speaker 1: I wrote about this topic now six years ago. And
- 34:03Speaker 1: it was one where I was like, this is the
- 34:04Speaker 1: thing that's going to matter if we don't have a
- 34:06Speaker 1: series of guardrails that exist, don't have a series of
- 34:09Speaker 1: what today I call trust infrastructure, don't have a way
- 34:11Speaker 1: to say this is right and this is wrong, we're
- 34:14Speaker 1: going to be in a place where it's not going
- 34:15Speaker 1: to go well and it cannot be in the hands
- 34:17Speaker 1: of the people building these models.
- 34:20Speaker 3: Yeah, what can go wrong here and what can go
- 34:22Speaker 3: right in the next six months to a year? What
- 34:25Speaker 3: do we need to depend on to go wrong or right?
- 34:29Speaker 1: I think the things that can go wrong is I
- 34:32Speaker 1: think there is a case where it says we rush
- 34:35Speaker 1: into legislations. We've rushed into legislation previously and we've got
- 34:39Speaker 1: some things. So an example of that is, not to
- 34:41Speaker 1: get too deep in the weeds, but Section 230, right?
- 34:44Speaker 1: Section 230, which gives coverage to people. It's 26 words
- 34:50Speaker 1: that gives coverage to platforms that they can host any
- 34:52Speaker 1: content that they want to host and they have no liability.
- 34:54Speaker 1: At the time, it was an amazing piece of legislation.
- 34:57Speaker 1: But if we reflect on it now, it incented the
- 34:59Speaker 1: wrong actions. It incented the wrong activities. So in the
- 35:03Speaker 1: context of today's world, we might rush into legislation And
- 35:07Speaker 1: all of a sudden we say, this is an amazing
- 35:09Speaker 1: piece of legislation. It's protecting all the right things. But
- 35:11Speaker 1: we fast forward five, six years from now, and it
- 35:14Speaker 1: was exactly the wrong thing. And so I just like,
- 35:17Speaker 1: how do we make sure that we're doing it in
- 35:18Speaker 1: the right way? I think it's a pretty important thing.
- 35:22Speaker 1: There are some things that are obvious, right? The kids
- 35:24Speaker 1: online protection, stopping deepfake pornography. Those are obvious things. But
- 35:32Speaker 1: how we think about who is responsible for guard railing
- 35:35Speaker 1: and ring fencing, the actions that are taken by these
- 35:38Speaker 1: very powerful model builders, I think is an important task.
- 35:42Speaker 1: If we screw that one up, I think that 10%
- 35:47Speaker 1: can start increasing pretty rapidly.
- 35:49Speaker 3: How do normal folks stay vigilant to not get themselves
- 35:53Speaker 3: in trouble?
- 35:53Speaker 1: It's an interesting word to use, trouble. I think it's
- 35:59Speaker 1: one where most people, unfortunately, aren't savvy to the fact
- 36:06Speaker 1: that the thing that they're interacting, most people interact on
- 36:08Speaker 1: their devices today, and they're not savvy to the fact
- 36:10Speaker 1: that a lot of what they interact with on their
- 36:12Speaker 1: devices today came from it came from AI. I mean,
- 36:15Speaker 1: even if you think about just the stat I share
- 36:17Speaker 1: with you, there are a trillion things created this year, right?
- 36:20Speaker 1: And it's the amount of music that's being created is
- 36:23Speaker 1: as far out swept all of the music that's ever
- 36:25Speaker 1: been created. If I think about the amount of images
- 36:28Speaker 1: that have been created, it has now surpassed all content
- 36:31Speaker 1: that existed on Google Images since the start of Google Images.
- 36:35Speaker 1: So you just look at that and you're like, okay,
- 36:36Speaker 1: so that's the amount of content. So it is one
- 36:38Speaker 1: where it's like, how do you get to a place
- 36:40Speaker 1: where you can think about that? So how does a
- 36:43Speaker 1: regular person do it? One is just think about what
- 36:45Speaker 1: you're prompting into these systems. And so when you're prompting
- 36:49Speaker 1: into a system, That is ultimately teaching the machine what
- 36:54Speaker 1: to think. It's retaining that information. So in the same
- 36:58Speaker 1: way you think about search and the way that Google
- 37:00Speaker 1: ultimately used search to its advantage, think about how you're
- 37:03Speaker 1: using every day, like whether it's your voice being used
- 37:05Speaker 1: into those systems, whether it's what you're typing into a
- 37:08Speaker 1: chat GPT. Just think about what you're doing there because
- 37:12Speaker 1: the machine is learning from that stuff.
- 37:16Speaker 2: A couple of just broad questions.
- 37:18Speaker 3: Give me your five-year prediction on what AI looks like
- 37:20Speaker 3: in 2031.
- 37:22Speaker 1: I think it's an impossible ask.
- 37:23Speaker 2: Yeah, I figured you'd say that.
- 37:26Speaker 1: I think it's an impossible ask. Today, if I was
- 37:31Speaker 1: seeing changes happening in a 24-month, 12-month cycle, I mean,
- 37:38Speaker 1: when we started our business, there was a new scientific
- 37:41Speaker 1: paper that came out roughly once every 15 months. Now
- 37:45Speaker 1: we're seeing a new scientific paper come out once every
- 37:47Speaker 1: 15 days. And so just like that's the level of
- 37:50Speaker 1: change that's happening. So it's really hard.
- 37:53Speaker 2: What about AI-generated movies?
- 37:55Speaker 3: How easily would they be made and can they be
- 37:58Speaker 3: made pretty good? Yeah.
- 38:00Speaker 1: I mean, again, go back to– we've talked about it here,
- 38:03Speaker 1: but go back to Minimax. Minimax 3, that's not possible.
- 38:07Speaker 1: It can happen. It is something that is– you can
- 38:12Speaker 1: also have a live– I mean, you can have a
- 38:13Speaker 1: live show. It doesn't have to be just a movie.
- 38:16Speaker 1: It could be a live show. I think the way
- 38:18Speaker 1: movie production happens, it can happen today that you can
- 38:22Speaker 1: do those things. Different questions of whether it's right or wrong.
- 38:25Speaker 1: But that is a thing that can happen today.
- 38:29Speaker 2: With you guys at your company, what is your goal?
- 38:30Speaker 2: What do you want to be known for in a
- 38:33Speaker 2: couple of years that you've helped either do or stop?
- 38:36Speaker 1: Yeah. So I think for us, it's three things. First
- 38:40Speaker 1: thing is we want to make sure that we protect everyone.
- 38:42Speaker 1: Let's stop all the unauthorized stuff from happening. Let's just
- 38:47Speaker 1: make sure that happens. The second thing is let's make
- 38:49Speaker 1: sure that we can give guardrails to people so they
- 38:51Speaker 1: can say, hey, when I do say yes, here's what
- 38:53Speaker 1: you can do and here's what you can't do. I
- 38:55Speaker 1: don't want that type of music created with my music.
- 38:57Speaker 1: It shouldn't be allowed. And then the last is let's
- 38:59Speaker 1: make sure everyone gets paid what they're supposed to get paid.
- 39:02Speaker 1: And so if those three things happen, because when I
- 39:04Speaker 1: think about AI at scale, certainly many people in the
- 39:08Speaker 1: creative community might lose their jobs, right? Might lose their fans.
- 39:11Speaker 1: If you go real doomsday on this stuff, that might happen.
- 39:15Speaker 1: But if they're all participating in all the economics, then
- 39:19Speaker 1: things should equal out. I'm not saying it's right that
- 39:22Speaker 1: that happens, but they should equal out. And so let's
- 39:25Speaker 1: get to a place where, so when we think about
- 39:27Speaker 1: our company, let's make sure humanity can thrive. That's our goal.
- 39:31Speaker 1: And the way that happens is give people the tools
- 39:34Speaker 1: and give people a protection that allows them to protect
- 39:36Speaker 1: themselves and then also allow them to say yes and
- 39:38Speaker 1: no when they want it to happen or they don't
- 39:40Speaker 1: want it to happen.
- 39:41Speaker 2: What jobs do you think are safe?
- 39:42Speaker 1: I think jobs that have, where experience matters are safe.
- 39:49Speaker 1: I think we're going to see a big resurgence in trades.
- 39:53Speaker 1: I think trades are going to go through the roof
- 39:56Speaker 1: in a pretty meaningful way. And I think we're going
- 39:58Speaker 1: to see people return back into agriculture.
- 40:02Speaker 3: What about when Elon Musk says we won't need money,
- 40:05Speaker 3: so we won't have to save money because we don't
- 40:07Speaker 3: need it in a decade, and we'll just be able
- 40:08Speaker 3: to live free and just be paid because AI does
- 40:12Speaker 3: all the work?
- 40:12Speaker 1: Yeah, I think if you draw any trend line out
- 40:15Speaker 1: long enough, you get to a place where there's enough
- 40:17Speaker 1: wealth in the world that that could be true. So
- 40:19Speaker 1: you could say that about any trend line. Yeah, I mean,
- 40:23Speaker 1: that theory has been around for a long time, where
- 40:25Speaker 1: you think about equality. We haven't been that good about
- 40:28Speaker 1: doing those things, though, as far as let's make sure
- 40:31Speaker 1: that there is You know, money doesn't matter because there's
- 40:34Speaker 1: enough for everybody, where there's an abundance of things. We've
- 40:37Speaker 1: not been great about that as humans.
- 40:38Speaker 3: Yeah, you got the guy with the absolute most money saying, hey,
- 40:41Speaker 3: money's not everything. It's easy for you to say when
- 40:42Speaker 3: you have the most money.
- 40:44Speaker 1: Yes.
- 40:45Speaker 3: I really appreciate the time. Thank you so much. And
- 40:48Speaker 3: I really love what your company is doing. Congratulations on
- 40:50Speaker 3: all the success of it. And yeah, thanks so much.
- 40:53Speaker 3: And I hope you have a great rest of the day.
- 40:55Speaker 1: Amazing. Thanks for your time, Bobby.
- 40:56Speaker 2: Thank you, Dan.
- 40:58Speaker 5: Thanks for listening to a BobbyCast production.