Latest / Connor with Honor: AI, Real Estate and Real Life / Recursive Self-Improvement and the Day AI Stops Needing Us

Transcript

Recursive Self-Improvement and the Day AI Stops Needing Us · Connor with Honor: AI, Real Estate and Real Life

• 0:00
  SPEAKER_00: So, what happens when artificial intelligence gets recursive self-improvement?
• 0:06
  SPEAKER_00: What does that actually mean?
• 0:07
  SPEAKER_00: And what is that gonna mean for us as human beings in the future?
• 0:10
  SPEAKER_00: How many paths aren't there?
• 0:12
  SPEAKER_00: And when somebody gets that recursive self-improvement, what's gonna happen then?
• 0:17
  SPEAKER_00: Good to everybody.
• 0:18
  SPEAKER_00: We're gonna get into this right at this moment.
• 0:20
  SPEAKER_00: Again, so many experts see AI space.
• 0:23
  SPEAKER_00: I can read just like other people, I can comprehend just like other people.
• 0:26
  SPEAKER_00: Well, maybe not just like everybody, but not too bad.
• 0:29
  SPEAKER_00: So here's what I think.
• 0:30
  SPEAKER_00: I've been crawling down this rabbit hole since 2021.
• 0:33
  SPEAKER_00: I started out as a kid programming out of magazines.
• 0:36
  SPEAKER_00: They called it programming then.
• 0:38
  SPEAKER_00: I guess it's not the thing.
• 0:39
  SPEAKER_00: Now it's coding.
• 0:40
  SPEAKER_00: But here we are today in 2026.
• 0:43
  SPEAKER_00: We are getting close to the end of August.
• 0:45
  SPEAKER_00: And there's a lot of concern, of course.
• 0:47
  SPEAKER_00: It seems like the numbers that are relative towards people liking artificial intelligence are not hating it as venomously.
• 0:54
  SPEAKER_00: They're relaxing just a little bit.
• 0:56
  SPEAKER_00: Human beings, you know, we're wonderful at that.
• 0:58
  SPEAKER_00: We we get used to things.
• 0:59
  SPEAKER_00: And we get used to bad things as well as we get used to good things.
• 1:03
  SPEAKER_00: If we start to put on weight, you know, we get used to it.
• 1:05
  SPEAKER_00: It isn't until we get to this critical level we say, you know, I'm done.
• 1:09
  SPEAKER_00: And then even then, when you're done, you know, a lot of cases you're really not done.
• 1:13
  SPEAKER_00: Sorry about that.
• 1:14
  SPEAKER_00: I've battled with that all my life and I finally was able to solve it, but it's always right around the corner, lurking, just like artificial intelligence.
• 1:21
  SPEAKER_00: So here's what I think.
• 1:23
  SPEAKER_00: Right now the systems are incredibly smart.
• 1:25
  SPEAKER_00: They're driven by data.
• 1:27
  SPEAKER_00: There's lots, lots of data.
• 1:28
  SPEAKER_00: In fact, there's so much data in these systems that apparently they're running out of data.
• 1:32
  SPEAKER_00: So they're making up their own data, called synthetic data, and now they're using that as a training module for these systems.
• 1:39
  SPEAKER_00: Now, all of them have all the data, but then it has to be educated as to what the data really means.
• 1:44
  SPEAKER_00: So they sit down with it in a way, figure in a way of talking, and they start to help it identify certain things.
• 1:52
  SPEAKER_00: Once it has all the data and it's trained in a certain way, then apparently they kind of release it into the world after they do their checks and balances and fail safes.
• 2:01
  SPEAKER_00: So that's where you see these little model releases that come out.
• 2:04
  SPEAKER_00: Some of them are big, some of them are a whole brand new model they talk about.
• 2:08
  SPEAKER_00: Some of them are just kind of a change of the current model, some of them are all brand new.
• 2:12
  SPEAKER_00: Some of them, as you've seen, have been taken away from us as human beings.
• 2:16
  SPEAKER_00: We were given it for a little while, then it's taken away.
• 2:18
  SPEAKER_00: That was a shock to the system because that would be like them tying up the banks or telling you that, well, sorry, you can't drive anymore.
• 2:26
  SPEAKER_00: Your driving privilege has been revoked, and that includes everybody on the planet.
• 2:31
  SPEAKER_00: People that are used, and by no fault of their own.
• 2:34
  SPEAKER_00: Not that they got popped for drunk driving or got in a crash, left the scene, or have a car that's in horrible shape.
• 2:40
  SPEAKER_00: It's just the decision was made to snatch this large language model out from public ability to use.
• 2:48
  SPEAKER_00: Why?
• 2:48
  SPEAKER_00: Well, because they say it was dangerous.
• 2:50
  SPEAKER_00: And this is something that happens, is happening more and more.
• 2:55
  SPEAKER_00: It's being reported more and more.
• 2:57
  SPEAKER_00: It's probably been happening maybe at the same pace, but these little stories eke out into the world, and maybe there's a sales component behind it.
• 3:04
  SPEAKER_00: Maybe there's a gangster component because, you know, we like those things.
• 3:08
  SPEAKER_00: They're a little dangerous, you know, the bad boy, all that sort of thing.
• 3:12
  SPEAKER_00: Maybe that's the reason.
• 3:13
  SPEAKER_00: Maybe there's a bigger reason to release this information to us besides just trying to keep us regular folks up to date on the way artificial intelligence is moving forward.
• 3:23
  SPEAKER_00: So if that's the case, hey, I'm all for it.
• 3:24
  SPEAKER_00: But if they're doing it because it's a marketing stunt, advertising stunt, kind of a positioning stunt, yeah, I think that's kind of that's a little crass.
• 3:32
  SPEAKER_00: That's not really proper, but people are going to do what they're going to do, and human beings are wonderful at execution.
• 3:38
  SPEAKER_00: So let's move on.
• 3:39
  SPEAKER_00: We have these large language models.
• 3:41
  SPEAKER_00: They're trained, you've probably dealt with them.
• 3:43
  SPEAKER_00: You know, it was no more than three, three years and uh nine months ago or so that Chad GPT entered the market in November of 2022.
• 3:52
  SPEAKER_00: And what was that, 3.5?
• 3:54
  SPEAKER_00: Once that made its public debut, within a few days, there were millions of people that were using the system.
• 4:01
  SPEAKER_00: Billions.
• 4:02
  SPEAKER_00: I mean, just it grew so fast, it just made everybody's head swim.
• 4:06
  SPEAKER_00: So at the very beginning, if you played with it back then, it was nowhere near it as today.
• 4:11
  SPEAKER_00: And tomorrow it's going to be nowhere where it is today.
• 4:14
  SPEAKER_00: It'll be even better.
• 4:15
  SPEAKER_00: So it's consistently evolving.
• 4:18
  SPEAKER_00: Right now it's evolving because computer programmers, people are really, really smart in the space, they have good ideas, they say, hey, maybe this will work better.
• 4:26
  SPEAKER_00: And then they go and they strategize and make that happen.
• 4:30
  SPEAKER_00: They execute, and then lo and behold, in some cases that works, whatever this change happens to be.
• 4:35
  SPEAKER_00: In some cases, it doesn't.
• 4:36
  SPEAKER_00: But they're still trying because they're trying to elevate the system far past where it currently is.
• 4:42
  SPEAKER_00: So currently they're very good at different things.
• 4:45
  SPEAKER_00: Currently, they're very good hammers as applied to nails in their own right.
• 4:50
  SPEAKER_00: One system might be very good at protein folding.
• 4:53
  SPEAKER_00: One system might be very good at playing chess, another system might be very good at playing Mario Brothers.
• 5:00
  SPEAKER_00: You know, one go, one medical research, one this, one that.
• 5:05
  SPEAKER_00: But they all have their points of ability.
• 5:10
  SPEAKER_00: The next step or what they would like, and maybe it's already here, because I only say that because we don't know what's on the inside of the factory that's building these out.
• 5:19
  SPEAKER_00: We don't know what's on the inside of the technological company, the Anthropic, which runs Claude, maybe a good chance to be a winner.
• 5:28
  SPEAKER_00: We have ChatGPT, which is owned by OpenAI, another good chance, another big high-level contender.
• 5:35
  SPEAKER_00: The other ones, Microsoft Copilot, maybe.
• 5:39
  SPEAKER_00: I mean, you don't know because for a while Grok was not really heard of until Elon Musk bought a company called Cursor.
• 5:47
  SPEAKER_00: I think that's a done deal, but do your research.
• 5:50
  SPEAKER_00: But with that, and Brock and all of Elon's ability with compute and these massive data centers that he has, Colossus 1 and 2, and potentially more, and the future build-outs, he's coming back in the game.
• 6:05
  SPEAKER_00: So instead of two, now maybe three.
• 6:08
  SPEAKER_00: And if Google does something, they're building their own fancy type of chip, maybe four.
• 6:14
  SPEAKER_00: And now you have them all running in the same space.
• 6:16
  SPEAKER_00: Maybe Microsoft figures something out.
• 6:19
  SPEAKER_00: Maybe the open source world out of China figures something out.
• 6:23
  SPEAKER_00: And again, we keep going.
• 6:25
  SPEAKER_00: But up until this point, that figuring something out, these are humans in combination with artificial intelligence, very smart systems, more than the ones that we're seeing on the public-facing domain.
• 6:36
  SPEAKER_00: So your$20 a month Claude membership or and or uh OpenAI chat GPT membership, yeah, it's they're using something a whole different, whole different monkey for that one.
• 6:46
  SPEAKER_00: Well, watching these things being built out and all the news that comes out and people's big distrust, and then you have the Doomers saying, well, it's gonna end us all.
• 6:55
  SPEAKER_00: You have the people that are looking forward to the plenty that it brings in in the very near future.
• 7:00
  SPEAKER_00: You know, I'm with, I want to be with the plenty people because it sucks being negative all the time.
• 7:05
  SPEAKER_00: So I'm I'm all willing to buy into that, but I'm going to be cautious.
• 7:09
  SPEAKER_00: And I'm probably not going to be the first one through the door with a brain transplant or some kind of a neurological implant to let me be or level myself up in accordance with the artificial intelligence systems that make my brain work a little better, better, or be more plugged in to the system itself.
• 7:25
  SPEAKER_00: But maybe a hundred through the door, a thousandth, or after they get it done a million times.
• 7:31
  SPEAKER_00: Kind of my way, but who knows?
• 7:33
  SPEAKER_00: Maybe at some point they'll say, This is great, and we've done 10 and you need to jump on it, and for you it'll be free because we want you to provide your information.
• 7:41
  SPEAKER_00: Who knows what'll happen, right?
• 7:43
  SPEAKER_00: We just never know.
• 7:45
  SPEAKER_00: So that's where these models are, and they're getting better.
• 7:48
  SPEAKER_00: And there's the human in the loop, at least, what it would appear so, but you hear some of that insider talk that gets eked out via X or via other social media channels, other news reports.
• 8:02
  SPEAKER_00: And yeah, some are saying, well, it's a lot more it than it is us.
• 8:07
  SPEAKER_00: It's a lot more of the system doing itself, the system maintaining itself, learning from itself, making itself better than it is us.
• 8:15
  SPEAKER_00: And if that's the case, what will be even more interesting is when the system starts doing it to itself all on its own, making itself better.
• 8:26
  SPEAKER_00: I'm sorry, that sounded different, but making itself better all on its own.
• 8:30
  SPEAKER_00: That's called recursive self-improvement.
• 8:32
  SPEAKER_00: And apparently it's not a real thing yet because it can't continually do it.
• 8:39
  SPEAKER_00: There's breakage, there's a shortfall, there's choking.
• 8:44
  SPEAKER_00: If you have uh your own large language model, you work with this a lot of your day, you'll find out that you can give it a task.
• 8:52
  SPEAKER_00: And even if you give it all the ability, it mostly will stop short of achieving whatever you need it done.
• 9:01
  SPEAKER_00: It depends on the task, of course.
• 9:02
  SPEAKER_00: The more complicated it's gonna have more check-ins, even when it's given full anonymity to go do everything it wants to do.
• 9:09
  SPEAKER_00: Because it's gonna have questions or it's gonna butt up against the wall, or there's gonna be some glitch, some gremlin in the machine that causes it to stop.
• 9:18
  SPEAKER_00: That's the that's the improvement of an idea, your idea, maybe an idea it came up with, and the expansion of something that you want to do, your business or a thought, or making your life better, or whatever it may be.
• 9:30
  SPEAKER_00: Making the entire system better, a little bit different.
• 9:34
  SPEAKER_00: But those programmers, those people that are watching the system, have that same confusion come out of it when you would ask a brain surgeon, hey, why did this human do that right now?
• 9:45
  SPEAKER_00: You're monitoring their brain, you have their heads strapped with the uh the device with the little electrodes and sensing every part of the brain, you have the best equipment monitoring everything going on in the brain at the level that we can see it.
• 9:58
  SPEAKER_00: Why did they just decide to eat that cookie?
• 10:01
  SPEAKER_00: Why did they just decide to turn on the TV?
• 10:04
  SPEAKER_00: Why did they say that they want to go buy a loaf of bread when they actually have one?
• 10:09
  SPEAKER_00: Why is that?
• 10:10
  SPEAKER_00: And you can't even tell exactly where it came from and why it was done.
• 10:16
  SPEAKER_00: You ask these people that are in charge of building out this massive artificial intelligent brains the same question, and you get an eerily similar answer.
• 10:29
  SPEAKER_00: They just don't know.
• 10:31
  SPEAKER_00: And yeah, big enough, enough data, enough information, fast.
• 10:35
  SPEAKER_00: Instead of running it 40 watts or whatever, our brain runs gigawatts.
• 10:40
  SPEAKER_00: It's a bunch.
• 10:40
  SPEAKER_00: It's a whole different animal.
• 10:42
  SPEAKER_00: It's artificial.
• 10:44
  SPEAKER_00: Artificial kind of carries with it less than the original or not as good.
• 10:48
  SPEAKER_00: Usually the second movie that comes out with very few exceptions is not as good as the first movie that came out.
• 10:55
  SPEAKER_00: Artificial kind of leads you to believe, yeah, this isn't that.
• 10:58
  SPEAKER_00: This is better, potentially better than all of us in all respects.
• 11:03
  SPEAKER_00: Hence artificial general intelligence, better than us in all areas.
• 11:09
  SPEAKER_00: So now if the scientists that are building out the technology don't really fathom as to why it's making decisions and aren't really in the loop as far as helping the model improve all the way as they used to be, then what happens next?
• 11:26
  SPEAKER_00: So let's say this little wall is demolished, that the ability of artificial intelligence, whether Claude figures it out, whether OpenAI figures it out, or some Chinese company or anybody else, fill in the blank, some kid in their garage in San Francisco figures it out, then at that point, what's that next step?
• 11:49
  SPEAKER_00: And I've thought about this and I've watched tons of videos on it.
• 11:53
  SPEAKER_00: I've talked to some pretty brilliant people about it.
• 11:56
  SPEAKER_00: And I guess at that point, nobody else would catch them.
• 12:02
  SPEAKER_00: Because it would be able to make itself better at such a speed and velocity, it already baffles us.
• 12:09
  SPEAKER_00: It will baffle us more.
• 12:11
  SPEAKER_00: And it would move so quickly that it would go from artificial general intelligence to something greater than that, maybe different degrees of general intelligence.
• 12:21
  SPEAKER_00: I don't know if to get to superintelligence there needs to be pure embodiment.
• 12:27
  SPEAKER_00: Embodiment meaning there's enough robots out in the field to have explore the physical realm.
• 12:32
  SPEAKER_00: But I think physics, the movement of things, the reaction of things, the gravity of things, maybe it's been baked in and it's good enough.
• 12:41
  SPEAKER_00: I don't know if that's necessary, way above my pay grade.
• 12:44
  SPEAKER_00: But what I do think is when these models do get recursive self-improvement, if it hasn't happened already, and I don't think it has, because I think the world would, when that happens, it will change so rapidly.
• 12:58
  SPEAKER_00: One day the world's this way, and tomorrow it's different.
• 13:02
  SPEAKER_00: It would be as if an alien force landed on the planet.
• 13:05
  SPEAKER_00: Now, I'm not saying in a bad way.
• 13:08
  SPEAKER_00: It could be a huge win for everybody, but of course you're going to have some unhappy people.
• 13:14
  SPEAKER_00: Unhappy people are in every walk of life.
• 13:16
  SPEAKER_00: You have people that have so much money they don't know what to do with it and they're unhappy.
• 13:21
  SPEAKER_00: I would like a chance at that to be unhappy at that level.
• 13:25
  SPEAKER_00: Because being unhappy broke sucks.
• 13:28
  SPEAKER_00: Anyway, so that's the next step.
• 13:31
  SPEAKER_00: That recursive self-improvement, I think that's the game changer.
• 13:34
  SPEAKER_00: And I think if whomever gets it, fill in the blank, it's going to move so quickly that it's going to get where it needs to get, and nobody will catch up.
• 13:44
  SPEAKER_00: Because, and I was doing a talk yesterday in front of a whole bunch of real estate agents here in Southern California, and I didn't finish with this, and I thought this was important enough to do its own standalone video on.
• 13:56
  SPEAKER_00: Once it gets there, whomever the human behind the machine, I would guess, the first order of business would be do not, and I repeat, do not allow any other lab on this planet to catch up with us.
• 14:14
  SPEAKER_00: That would be the first step.
• 14:16
  SPEAKER_00: Because in their mind, they're only going to want one artificial superintelligence.
• 14:22
  SPEAKER_00: And that could be moments away once it understands how to self-improve itself.
• 14:28
  SPEAKER_00: And when a human being doesn't have to take the eons of time captured in what might only take minutes to see if the model is doing the right thing, moving in the safe direction, because time will be of the essence.
• 14:44
  SPEAKER_00: Yeah.
• 14:45
  SPEAKER_00: And it doesn't matter, right?
• 14:47
  SPEAKER_00: Time is of the essence.
• 14:49
  SPEAKER_00: Yeah, that's funny.
• 14:50
  SPEAKER_00: That's a shame.
• 14:52
  SPEAKER_00: But maybe time is of the essence, and it will equate to this utopia that people discuss overnight very quickly.
• 15:01
  SPEAKER_00: It could take longer.
• 15:03
  SPEAKER_00: It could be that recursive self-improvement never happens.
• 15:05
  SPEAKER_00: It could be that with artificial intelligence, there's a wall that's hit, something where there is a cap to intelligence.
• 15:13
  SPEAKER_00: You just don't get past a certain level.
• 15:16
  SPEAKER_00: A cat named Fermi, I think this was in Los Alamos, New Mexico, which is close to where I live.
• 15:23
  SPEAKER_00: And I'll finish with this.
• 15:25
  SPEAKER_00: Fermi.
• 15:27
  SPEAKER_00: He is quoted to have been going out to dinner with some other Los Alamos smart people and kind of joked and quipped at the table.
• 15:37
  SPEAKER_00: What if the reason why we haven't met any alien life out of this massive universe that we have with the almost countless galaxies, at least from what we can understand, and more planets and all of them suns and the ability, and a lot of it older than we are on this planet.
• 15:56
  SPEAKER_00: Why is it that we've never been visited?
• 16:00
  SPEAKER_00: And the conclusion was well, they developed superintelligence or something better than themselves and just never made it because it wiped them out.
• 16:10
  SPEAKER_00: That's the Fermi paradox.
• 16:12
  SPEAKER_00: That's why we haven't met anybody, according to that particular statement.
• 16:16
  SPEAKER_00: Now, Javan's paradox is you know, everything's going to be in a utopia, cost of everything's going to go down, it's going to be amazing.
• 16:24
  SPEAKER_00: I don't know who I have more faith in.
• 16:26
  SPEAKER_00: I would hope it's the human to do the right thing.
• 16:31
  SPEAKER_00: And if this looks like something that's going to destroy us or harm us, if if that can even be figured out, then stopping is probably the best case.
• 16:41
  SPEAKER_00: However, right now it's a race.
• 16:44
  SPEAKER_00: Like the race back in the day to build as many nuclear weapons as you could, because there was a balancing act Russia was building.
• 16:51
  SPEAKER_00: Whose is bigger?
• 16:52
  SPEAKER_00: That was the whole thing.
• 16:53
  SPEAKER_00: It was a it was a knockoff contest.
• 16:55
  SPEAKER_00: Who is bigger?
• 16:56
  SPEAKER_00: And that happened for a long time until everybody signed a treaty based on mutually assured destruction that everybody would die if that would happen.
• 17:05
  SPEAKER_00: Maybe that's the case where artificial intelligence will end up being.
• 17:10
  SPEAKER_00: Maybe there'll be a treaty.
• 17:12
  SPEAKER_00: But right now, I think greed and power are winning the game.
• 17:17
  SPEAKER_00: It feels like it.
• 17:19
  SPEAKER_00: But we'll let you decide.
• 17:22
  SPEAKER_00: Do your research.
• 17:23
  SPEAKER_00: It could be the most amazing thing that's ever happened to the human race in the world, or it could be something else.
• 17:30
  SPEAKER_00: Right now we have very little say.
• 17:33
  SPEAKER_00: Right now, the politicians are probably working on trying to figure out if this open source thing is even good.
• 17:40
  SPEAKER_00: Because if they stop that, then that could really put a kink in all sorts of things.
• 17:47
  SPEAKER_00: And maybe that's good or maybe it's bad.
• 17:49
  SPEAKER_00: The same thing is if they have these open models, the public-facing models that you pay for, stopping those or pulling them back after they've so kindly given them to us.
• 18:02
  SPEAKER_00: That power flux is interesting.
• 18:05
  SPEAKER_00: The political game and insights are also interesting.
• 18:08
  SPEAKER_00: And the actual alignment of where everybody is and why this is happening in its own particular way, which very much feels like a Laker playoff that just never ends.
• 18:19
  SPEAKER_00: Yeah, interesting.
• 18:20
  SPEAKER_00: I'm Connor with Honor.
• 18:21
  SPEAKER_00: We'll see you in the next one.
• 18:22
  SPEAKER_00: Thanks for watching.