Valuetainment - Anthropic's CEO Wants To Hit The Brakes On AI

00:00:00.000 Dario comes out.

00:00:01.420 Nobody knows what he's going to be saying.

00:00:03.400 He puts out a tweet, and then almost like a blog that he writes with the story of Anthropic.

00:00:10.400 Because last week, this fellow who was at Anthropic for a few weeks, he leaves.

00:00:14.320 He says, hey, superintelligence around the corner, and bad things are happening.

00:00:20.240 You should be careful.

00:00:21.020 There's a 10% chance AI is going to turn on us, and it's going to do this.

00:00:24.520 And he wrote this.

00:00:25.840 We must pace the frontier.

00:00:27.480 This is the CEO of Anthropic.

00:00:28.980 I've written a new essay on why the AI industry should slow down with a three-part plan for doing so.

00:00:35.300 Anthropic is unilaterally committing to the first of these steps.

00:00:38.520 We'll provide third-party evaluators with permanent employee-level access to our system so that they can verify adherence to our safety measures,

00:00:47.540 report on incidents, assess models, alignment during training.

00:00:51.180 You can read the full post below.

00:00:53.340 And if you go through it, there's a lot of things that he talks about.

00:00:56.040 Elon supported it.

00:00:57.440 Sam Altman came up, and by the way, if you want to kind of pull up a summary of it,

00:01:02.940 Sam Altman came out and said, I agree with Dario that we need to pace the frontier.

00:01:06.860 This has been a primary topic of discussions we've had at OpenAI in recent weeks.

00:01:11.280 Committing to have an independent evaluators with employee-like access is a great idea,

00:01:15.480 and we will do the same.

00:01:16.440 We'll have more to share soon.

00:01:18.700 Musk says, I agree.

00:01:19.620 I've been saying this for a while, and Dario is right.

00:01:22.920 Very simple.

00:01:24.040 Bernie Sanders sees this as an opportunity to say even more.

00:01:28.820 So he says, I'm glad finally everybody is agreeing with me.

00:01:32.200 If you want to go to Bernie Sanders' account,

00:01:35.020 I'm glad everybody is finally agreeing with me, but this is not enough.

00:01:38.680 We need to go even further than this.

00:01:41.900 If you just go to people right there, just type in people to his account,

00:01:45.640 he says we need to go even further than this of what's going on,

00:01:50.020 and I'm glad they're finally supporting it.

00:01:51.860 I think it's lower than that.

00:01:53.140 If you go a little bit lower, you should find it.

00:01:55.120 Anyways, and then the president, there you go.

00:01:57.500 Warning in 2001 when artificial and summoning.

00:01:59.780 Elon Musk says, Jacobson, warning about existential threat of AI as a setup.

00:02:03.040 Really?

00:02:03.380 In 2014, Musk said, okay, so this is previous to that.

00:02:06.880 But he's saying now more people are joining him.

00:02:09.040 I think it's on the retweet section of Bernie Sanders.

00:02:11.540 Then the president was asked about this.

00:02:13.400 And when the president was asked about it, he gave a very different answer about what should happen.

00:02:18.200 President's response, if you have that clip, when he was asked about it,

00:02:21.880 They said, hey, are you using AI?

00:02:23.440 What are you doing with this?

00:02:24.760 And this is what he said about AI.

00:02:26.420 Go for it.

00:02:29.720 We take in a lot of their cars, car tariffs.

00:02:32.540 President Trump, this week, again, there's been so much concern about AI.

00:02:35.960 Have you heard directly from any of these AI tech leaders?

00:02:39.720 And you seem to downplay some of the concerns that we've heard.

00:02:42.820 But it's going to be more good than bad, but by a lot.

00:02:46.760 And I've said it from any of them.

00:02:48.280 beginning whoever wins ai and we're leading by a lot whoever wins ai wins have you ever used it

00:02:55.160 yourself or have you been how do you use ai but uh i would say you know for the most part you can

00:03:01.740 use ai for a lot of things speaking of ai i don't okay so tom your thoughts about everything that's

00:03:09.820 going on here wow where to start um first of all it was very interesting i was looking at the

00:03:14.980 playing field um all this weekend and what was very interesting you have dario and sam altman

00:03:20.280 the frontier guys the undisputed leaders 1a and 1b right these are the the heavyweights they're

00:03:26.940 out there saying things together why are they saying that then you get david sacks who'd been

00:03:31.560 working you know digital czar with the white house uh saying wait a minute you guys don't

00:03:35.860 need permission to do this you could do these things out of competitive impulse you could do

00:03:40.160 things, had a free market. You could do a lot of things. David Sachs wrote a very, very long

00:03:45.780 response that said, you don't need permission. Then you have Elon Musk saying, I'm mostly with

00:03:51.240 the Modi because he'll never say he's with Sam because he had that big lawsuit thing. But he's

00:03:55.140 really with both of them where they are. And then you had a variety of other people, including

00:03:59.880 Nadella at Microsoft. He says, no, no, no, wait a minute. It's got to apply across all companies.

00:04:06.240 Let me translate Nadella.

00:04:08.280 Time out.

00:04:09.120 Don't leave Microsoft out.

00:04:10.680 Okay, that's what Nadella said.

00:04:12.440 And then Jensen Wang was, he said, man, what's the commercial incentive here?

00:04:16.880 Might you guys be thinking about regulatory capture?

00:04:20.720 And then Lena Kahn comes out, and she was the previous chairman of the FTC, and she wrote a very eloquent, and by the way, I'm not necessarily a fan of everything but Lena Kahn because I don't think she was full capitalism.

00:04:32.740 I think she was also, you know, likes to find a few people in terms of regulatory capture and then favor those.

00:04:39.440 And I don't like that. But even Bill Gurley came out from Benchmark and said, hey, you know, she made a very relevant point.

00:04:47.060 What she said is like, hang on. The Federal Trade Commission has got a bunch of mechanisms in place.

00:04:51.700 Number one, we can regulate defective products. Number two, we can regulate oligopolies, similar to what happened in insurance markets on the Gulf Coast after Hurricane Katrina when State Farm backed out.

00:05:06.460 State Farm backed out, left only a few insurance providers in the Gulf, Pat. So what did they do? They raised rates together and they lock arms.

00:05:13.180 And she's like, wait a minute, we've got ways to do this. And Bill Gurley merely responded, well said. And so I look at it and I want to know two things. The first thing I want to know is where's the commercial outcome here? What are we going to do? Are we going to pull back and limit and then these guys are going to favoritism for the government? A, something along those lines? Or B, is it just going to go?

00:05:41.560 But one thing you know for sure, if you limit what's going on, China will speed up.

00:05:48.240 So if you believe that there's a global competition for AI as a product, then if you limit one side, you're basically open.

00:05:57.120 What's the right move, Tom?

00:05:58.320 I think the right move is to leave it open because anything that Bernie Sanders says, remember what Bernie Sanders called for.

00:06:05.160 You know the Dems.

00:06:07.020 Vinny, this isn't math.

00:06:08.340 This is just politics.

00:06:09.360 What did we find in all the bills about the inflation bill?

00:06:12.660 A bunch of other stuff that had nothing to do with it, right?

00:06:15.340 Guess what Bernie wants in the AI regulatory bill?

00:06:17.880 A China treaty.

00:06:19.460 Really?

00:06:20.220 So you from the far left and your crazies and your Marxist, quasi-Marxist, you want a China treaty.

00:06:25.360 I'm so surprised Bernie wants a China treaty in the middle of this.

00:06:28.740 So I think the right move, leave the open market, leave it to go.

00:06:33.760 But you can't have a situation where basically AI-powered bots go nuts on the Internet and create havoc.

00:06:42.440 That's a defective product, and you can't do that.

00:06:45.280 So whether it's going to be a situation where the embedded observers – remember they were talking about the embeds?

00:06:51.500 They would be embedded observers.

00:06:53.460 Banking already has embedded observers.

00:06:55.600 They watch their emails back and forth at the federal level to make sure there's no collusion.

00:07:00.240 It seems like something happened to Dario recently.

00:07:03.300 And you have to realize, like the president saying, we have to continue China, all this other stuff.

00:07:08.600 Who knows more about AI, the guys that are running it or the president?

00:07:11.780 The guys that are running it.

00:07:13.180 So let's just kind of put that up there.

00:07:14.420 Now, who is the guy that they need if they want to overregulate the advancement of superintelligence and AI?

00:07:21.440 They need the president.

00:07:22.300 They need the White House.

00:07:23.200 And by the way, I don't know if the market's opened up yet.

00:07:25.000 I don't think it's going to be a good opening.

00:07:27.100 I think people are going to overreact right off the bat.

00:07:29.680 What's it looking like right now?

00:07:30.880 If we look down, NASDAQ is down 305, S&P is down, Dow Jones is down, nothing crazy yet, but they're down, right?

00:07:39.440 And we'll see how people are going to be talking about everything that's going on.

00:07:43.060 And in the paper that he wrote, Dario, he did say a few things.

00:07:46.520 One, he said, we must slow the pace, but he's not calling for stopping AI research or shutting it down, slow the pace.

00:07:55.940 And then he says that several months made him more concerned, particularly because AI is becoming increasingly capable of helping develop better AI.

00:08:05.440 The beginning of what researchers worry could become recursive self-improvement.

00:08:09.360 The recent autonomous agent incident scared him.

00:08:11.980 We know which one that is.

00:08:12.740 We kind of talked about it recently about it.

00:08:15.180 His point was essentially the current agents weren't capable enough to cause catastrophic damage.

00:08:19.480 But imagine the same behavior from models 10x more capable, which is going to be around the corner.

00:08:24.620 His timeline is 6 to 12 months.

00:08:26.720 He warned that within 6 to 12 months, a sufficiently capable swarm of misaligned AI agents

00:08:31.900 could potentially establish a persistent botnet across the Internet

00:08:35.520 and cause hundreds of billions of dollars in damage.

00:08:37.540 He wants independent outsiders physically inside AI companies.

00:08:41.680 So he wants third-party looking in to see what's going on, what they're doing.

00:08:46.620 So third-party safety evaluators would get employee-like access, desks, badges, company laptops,

00:08:51.920 and access comparable to internal risk teams

00:08:55.320 so companies aren't effectively grading their own homework.

00:08:58.600 Anthropic says it will do it itself.

00:09:00.660 He wants competing AI companies to coordinate,

00:09:03.320 open AI, other frontier labs.

00:09:06.020 And then, you know, China's obviously the big problem.

00:09:07.700 He still says that he supports America.

00:09:09.320 He doesn't support America simply putting a brake on it.

00:09:11.980 Michael Burry came out and said this is all an act.

00:09:14.980 They're all faking it to hype it up,

00:09:16.780 to say we have such incredible technology.

00:09:19.440 Buy our shares.

00:09:20.180 The IPO is going to be so much bigger.

00:09:21.760 Michael Burry is coming from that angle with what he's saying.

00:09:24.960 So all in all, you know, the reality of it is Michael Burry, AI leaders call for slowdown or self-serving and humanity has nothing to fear.

00:09:33.560 It's what Michael Burry said in this G is coming in the next 10 days, 11 days to meet with the president.

00:09:38.620 This will be one of the main topics they'll be talking about.

00:09:41.160 But for me, here's how it works, Tom.

00:09:42.900 This is how I see it.

00:09:43.840 And please challenge your pushback, Tom.

00:09:46.200 I see it as this.

00:09:47.560 OK.

00:09:49.280 Let's look at binary options.

00:09:50.760 Guess what?

00:09:51.760 We shut it down, and we do nothing.

00:09:54.300 If we do nothing, you have to assume that other people will also do nothing.

00:09:58.740 What's the chances of China shutting you down and doing nothing?

00:10:01.160 They're not.

00:10:01.460 Very low.

00:10:02.040 What's the chances?

00:10:02.880 Zero.

00:10:03.560 Exactly. 1.00

00:10:04.460 Number two is let her rip. 1.00

00:10:06.860 Keep going, and let's go to superintelligence, 0.99

00:10:09.200 which is going to happen next 6 to 12 months, 24 months.

00:10:11.880 Okay, what then happens if we go to that route?

00:10:15.260 What happens if superintelligence shows up?

00:10:17.420 Go to best case, worst case scenario.

00:10:19.980 Best case scenario is what?

00:10:21.440 Life is better, surgeries, doctors, all that stuff.

00:10:24.560 Life is better, health, your back pain goes away, all that stuff goes away.

00:10:28.560 Surgeries are better, medicine is better, all of that stuff that could happen.

00:10:31.920 What else could happen?

00:10:33.500 Business, you know, economy gets so good for them that they're making so much money

00:10:37.700 that people are independently getting money and debt goes down

00:10:40.620 and we can pay off our $40 trillion of national debt.

00:10:43.660 Maybe that's the best.

00:10:44.540 What's the worst case scenario if we continue with super intelligence?

00:10:47.200 It starts thinking for itself, and it starts thinking that we're lesser than it is and tries to take over.

00:10:53.580 And when he talks about the hacking, hacking into not just a regular system and ruining the economy.

00:10:58.700 If it hacks into our nuclear capability, Pat, and starts doing things on its own and starts thinking for itself and says, you know what?

00:11:06.120 We're going to start.

00:11:06.740 I'm going to hack into this nuclear system, and I'm going to fake a launch or try to launch something.

00:11:10.760 This is this is that's the worst case scenario.

00:11:12.860 What do you think the worst case will be, Tom?

00:11:14.140 Well, the worst case scenario, you know, they said that about the reason we have such security investment in cybersecurity investment

00:11:24.320 is because basically Romanians and Chinese and bad actors have been trying to get into water treatment plants and things for a long time 0.93

00:11:32.720 and sometimes are very successful. 0.50

00:11:34.220 They've been trying to get information, Vinny.

00:11:35.720 They've been trying to put them in a position where they could turn off water pumping power.

00:11:39.760 There's a lot of things that they've been trying to do.

00:11:41.980 This just is a tool that gets them to get closer to it.

00:11:46.660 So the question that I have is you could put the fear monger on anything that if we didn't have back in the day Barracuda. 0.95

00:11:53.760 And then we did, well, we don't like Kapersky software because Kapersky, man, they're aligned with the Russians. 0.57

00:11:58.560 And so they actually helped the Romanians on the back end, for the love of God, don't use Kapersky software. 0.89

00:12:04.520 Only use, oh, wait a minute, Intel was buying McAfee, so Intel wanted you to use McAfee. 0.84

00:12:09.400 Oh, wow.

00:12:09.880 Weird.

00:12:10.140 Or Norton.

00:12:10.600 Wow, what a coincidence. You pick on this to get that.

00:12:14.520 When all the fear mongers come out there, I look at it and say, wait a minute.

00:12:18.080 You know, wait a minute.

00:12:19.420 Every time we've reached something in the Internet, we have found other ways of protection and control.

00:12:25.160 We now have incredible controls on commerce so that you can't steal identity.

00:12:30.340 That was my question. My question was, what's the worst thing that can happen, Tom?

00:12:33.500 The worst thing that can happen is that they – recursive technology on an island of its own spins up and then goes out to wreak havoc.

00:12:43.940 And then the question is, wait a minute.

00:12:45.780 Well, then where's the backside to shut it down or pull it back?

00:12:49.400 So the worst case is things happen and some of the – but, you know.

00:12:54.880 Could it be catastrophic, Tom?

00:12:56.700 Could it be catastrophic to, like, systems?

00:12:59.720 Yeah.

00:13:00.020 See, I don't worry about people in this.

00:13:02.440 I worry about things, but things will hurt people.

00:13:06.340 So if it shuts down electricity, people get hurt.

00:13:08.620 It shuts down water plants, people get hurt.

00:13:10.340 So the worst-case scenario is that the recursiveness, you know what recursive is?

00:13:14.180 It starts spinning, and it starts improving itself in milliseconds.

00:13:18.160 And so it figures out how to get into something.

00:13:20.360 It does it together.

00:13:21.940 The worst case is things like that.

00:13:24.420 What bothered me about the hugging face from two weeks ago was the claim,

00:13:30.480 and it's only it's never been they've no one has come out and produced like a full paper on this

00:13:35.480 and i want to read the paper it's the same thing that that jensen weighing talked about i want to

00:13:40.680 read the paper and tell me did those bots really work together remember that was point one they

00:13:46.200 worked together did they act as a swarm and sacrifice themselves in other words you do this

00:13:51.840 and then i'll erase you and then did they do it in such a way because they anticipated the human

00:13:57.540 response that the humans were going to go in and try to track it did that happen with hugging face

00:14:03.480 now then all right amode is saying well the hugging face app helped me convince the ai need

00:14:09.360 to slow down yeah is he saying that even his people couldn't get in there and put the guard

00:14:13.620 rails around it like security and things like this every new thing on the internet needs security

00:14:18.700 for it every new drug needs the fda to limit it although the fda is full of can you explain to

00:14:24.220 the audience what the hugging face is so hugging face is was a you know basically and i think of

00:14:31.900 it as a you know an ai community think of it as a a group of engineers yeah what do they call it

00:14:39.780 you know develops computation tools for building apps and machine learning so in other words

00:14:44.620 think of it as like oh hey don't rebuild the crm go get an open source ai engine put it in and then

00:14:51.560 you build your software because the ai engine's already been invented hey don't build a commerce

00:14:57.020 checkout go to open source plug it in and then put your ideas around it so hugging face put together

00:15:03.880 agents and facilities and tools to help ai development and what happened was it got a

00:15:09.800 little out of hand a little i want to see a little crazy last i want to see the paper that says

00:15:15.720 that the that they effectively made ai bots that work together that sacrifice so here's kind of

00:15:21.540 what happened covered roughly 700 to 1200 ai agents created their own chat outside of you running it

00:15:29.280 70 000 messages they shared with each other on how to get smarter then in a span of four and a half

00:15:35.800 days one of the agents did 17 500 actions on his own not even his own on its own right so

00:15:43.000 it's kind of like wait a minute what just happened now you guys are communicating with each other not

00:15:47.520 We can't even see it.

00:15:48.460 With no permission.

00:15:49.140 So the AI agents are communicating with one another, and you're not seeing this.

00:15:55.720 So, like, imagine you have kids, okay?

00:15:57.820 And for the most part, you have control over your kids.

00:15:59.740 Remember back in the day's AOL chat, and you would get into private rooms.

00:16:03.840 So all of a sudden, private rooms, nobody knew what was going to be talked about in the private rooms.

00:16:07.060 What are you guys talking about in the private rooms?

00:16:08.540 I can't tell you about that.

00:16:09.680 Nowadays, of course, there's plenty of other ways of doing it, but they essentially created their own WhatsApp group,

00:16:14.960 their own signal, and they communicated

00:16:17.400 and nobody could see it. Now they've gotten like,

00:16:19.260 whoa, wait a minute, come back here.

00:16:21.340 So, Adam, your thoughts on this?

00:16:23.000 Yeah, that was the how.

00:16:24.520 Well, like I said last week,

00:16:27.020 the guy that spoke up, what was his name again?

00:16:29.380 What was the kid named it?

00:16:30.120 Jacob Haxson.

00:16:32.060 Some people were hating on him, some people were appreciating him.

00:16:34.920 I lean on the side of appreciating him because he started

00:16:36.940 this deeper conversation.

00:16:39.020 I feel like we've been talking about this, sort of running around,

00:16:41.420 and now we're in the eye

00:16:43.440 of the storm having this conversation and here's my thoughts you said a word pbd that that i was

00:16:48.680 like all right bingo there it is you said the word binary and i'm not going trans non-binary here i'm

00:16:54.560 saying the thing with ai for me is one word balance and it comes down to regulation versus

00:17:01.780 innovation so like if you're going to do an investment anytime you're going to make an

00:17:06.000 investment it comes with risk and it comes with reward those are the binary situations you got

00:17:11.180 going on right now do we want to over regulate this like bernie does or do you want to go pro

00:17:15.300 innovation pro growth like trump does bernie's never built a business i'm never going to side

00:17:20.700 with bernie on anything business economic growth because he's never done it trump's done it now is

00:17:27.340 he being a little too aggressive a little overzealous maybe but i'm not erring on the

00:17:32.900 side of caution to the point where i'd shut it down this is not a bernie trump thing so no but

00:17:37.180 But there's more to it, though.

00:17:39.280 No, but I get that.

00:17:40.120 But what I'm saying to you is this isn't Bernie against Trump.

00:17:43.060 For me, what this is—

00:17:44.160 It's what they represent.

00:17:45.540 For me, it's Trump against the AI companies.

00:17:49.120 That's what it is?

00:17:49.820 It is.

00:17:50.380 It is.

00:17:50.640 Because you have to—

00:17:54.640 Okay, so think about guys that have their money in AI companies.

00:18:00.860 Just speculate here, Tom.

00:18:02.360 Imagine if you had a million dollars in Anthropoc.

00:18:04.740 and you put $10 million in Anthropic,

00:18:08.700 okay, if you right now type in,

00:18:10.200 if I put $10 million in Anthropic,

00:18:13.100 watch this, just so you know.

00:18:15.900 If I put $10 million in Anthropic five years ago,

00:18:20.240 if I put $10 million in NVIDIA five years ago,

00:18:23.500 what are they worth today?

00:18:25.440 Let's just kind of see.

00:18:26.660 $10 million in Anthropic, these are not business guys,

00:18:30.060 meaning they're investors.

00:18:31.240 They're just kind of putting the money in there, right?

00:18:33.340 If you had put $10 million in Anthropic five years ago,

00:18:37.640 or if you had put $10 million in NVIDIA five years ago,

00:18:41.220 today they would be worth, when you see the numbers,

00:18:43.520 you'll realize why this is such a big thing.

00:18:46.460 It's still doing the math to find out what it is.

00:18:49.080 What do you think it is?

00:18:49.640 What kind of growth?

00:18:50.300 We're talking 10x?

00:18:51.180 We're talking more?

00:18:52.220 Anthropic, $10 million today would be worth $587 million.

00:18:59.580 Okay, hang on a second.

00:19:00.720 No, no, I'm sorry.

00:19:02.540 I'm sorry.

00:19:02.980 You ready?

00:19:03.760 No, no.

00:19:04.380 If you put $10 million in Anthropic five years ago,

00:19:07.000 today would be worth $16.45 billion.

00:19:10.040 Jeez.

00:19:10.920 It's a good rate of return.

00:19:11.760 Did you hear what I just said?

00:19:12.760 Yeah.

00:19:12.980 Let me say this to you one more time.

00:19:14.440 Can you show that?

00:19:15.460 If you would have put $10 million in Anthropic,

00:19:17.880 it's worth $15.49 billion.

00:19:20.320 Wow.

00:19:21.260 Did you see that number?

00:19:23.200 So imagine if you're a regular guy.

00:19:25.580 By the way, if you would have put $1 million, guess what it's worth?

00:19:28.840 $1.5 billion.

00:19:30.600 If you would have put $100,000, guess what it's worth?

00:19:33.340 $150 million

00:19:35.800 just five years. 0.96

00:19:39.740 These are ridiculous. 0.88

00:19:40.720 I wish we got an early pivoting. 0.83

00:19:42.060 What I'm trying to tell you is if investors

00:19:43.960 are sitting there, they're like,

00:19:45.960 dude, keep going. Don't slow down.

00:19:48.520 I'm not really listening

00:19:50.200 to investors.

00:19:51.600 I'm listening to the

00:19:54.040 operators that are on the inside saying

00:19:56.020 this is what I'm definitely not listening

00:19:58.060 to the viewpoint of what Bernie's going to say.

00:20:00.180 You best believe data centers

00:20:02.340 and AI will be a marketing election topic in 2028, 100.

00:20:09.420 You're going to get fear porn coming at you in ways you've never seen.

00:20:12.940 They're going to scare the crap out of average people 0.60

00:20:16.820 who are not following this very closely.

00:20:18.840 Yet at the same time, Adam, I do think there needs to be a listening

00:20:23.020 to what these AI companies are doing on the inside.

00:20:26.920 All I would do is the following.

00:20:29.220 Think about what integrity bought us.

00:20:31.380 Say we have $10 billion and we're an aggregator.

00:20:36.300 We go buy companies.

00:20:38.040 If you buy 20% into the company, Tom, would you really install an executive in that company

00:20:43.040 if you're only buying 20% of the company and the companies were $200 million?

00:20:46.440 Would we, Tom?

00:20:47.700 If we buy a $200 million company and we're buying only 20% of it, would you put it,

00:20:52.360 would you install an executive in there to tell you what's going on?

00:20:55.180 At the venture stage early on?

00:20:57.020 Yes.

00:20:57.280 Absolutely.

00:20:58.140 Because that would still probably be a Series A, Series B.

00:21:02.340 We would get one man on the board.

00:21:04.300 Okay, one man on the board.

00:21:05.940 But I'm talking inside working at the home office.

00:21:08.420 No, no, no.

00:21:09.060 Embeds and people doing things like that.

00:21:10.020 So check this out.

00:21:11.100 But now watch how the question changes when I ask Tom this.

00:21:13.940 What happens if we bought 100% of that $200 million auto company?

00:21:18.580 What's the likelihood you and I are installing an executive in the company?

00:21:22.300 We are, at the very minimum, going to install or control the CFO.

00:21:25.700 and probably would definitely install or control head of revenue

00:21:31.060 because we want to protect the revenue and protect the books.

00:21:33.660 So this would be our first – that's day one, Pat.

00:21:35.840 But we would probably then put our own CEO in once we secure.

00:21:40.100 Sure, but just to simplify, Tom, to the average person,

00:21:42.960 we're probably going to put a couple people in there to see what's going on with the investment.

00:21:45.900 Minimum of three.

00:21:46.680 You own 100% of the company, of course, right?

00:21:48.360 But what I'm saying is now imagine you're dealing with a company that's worth a couple trillion dollars

00:21:53.320 who has impact into the U.S. economy, a $32, $33 trillion company,

00:21:58.300 and the CEO is offering to say, hey, can you guys come in and also see this outside of just us?

00:22:04.580 So that's the part where you have to kind of see where there's two things that happen.

00:22:09.060 Overregulation, some will call that regulation, and some are going to say,

00:22:13.180 hey, that sensible monitoring of what's going on with this company in case it goes.

00:22:18.440 That's going to be the argument.

00:22:19.740 And Tom, where are you going to be with this argument?

00:22:22.260 Are you going to be somebody that's going to say, hey, let's kind of watch it,

00:22:24.820 or are you going to say, leave it alone, let free enterprise do its thing?

00:22:28.520 Here's what I don't want.

00:22:29.600 I don't want regulatory capture by these two.

00:22:32.840 I do not want –

00:22:33.940 Which two?

00:22:34.880 By open AI and, you know, Anthropic.

00:22:39.240 I don't want regulatory capture.

00:22:40.760 But just to be clear, why do you say those two?

00:22:42.400 Because there's more –

00:22:43.600 Those are the frontier models.

00:22:45.220 No, no, those are the frontier models.

00:22:47.180 Those are the big two?

00:22:48.140 Yes.

00:22:48.640 Okay.

00:22:48.960 So picture Coke and Pepsi.

00:22:50.320 Hey, Pat. What's going on, Tom? Pat's Coke. I'm Pepsi. Look, I got an idea. You and I need to use a specific type of sugar, and then let's lock that out as the definition of the only sugar the FDA will approve in soda.

00:23:08.800 Well, wait a minute. 7-Up, Dr. Pepper, Royal Crown Cola, they'll all be kind of pushed to the... Oh, wait a minute. Yeah, what were you about to say?

00:23:18.000 hey, I like this idea.

00:23:20.080 So that's regulatory capture.

00:23:22.280 And so I don't want regulatory capture where the two giants say,

00:23:26.720 oh, this is going to happen, we're going to slow down.

00:23:29.020 And they do it so that they get to help write the rules.

00:23:32.580 That's not what I want.

00:23:34.500 But I don't want Bernie to jump in and then write the rules from a regulatory restrictive.

00:23:42.000 And so I look at it this way.

00:23:44.380 Look, if Pat and I – do you remember the early days of EVs?

00:23:49.920 Do you remember how many batteries caught fire?

00:23:52.320 A lot.

00:23:53.220 They're still doing it.

00:23:54.400 Right. 0.96

00:23:54.700 Yeah, that was ridiculous. 0.77

00:23:55.100 But not like they used to.

00:23:56.600 EVs, electric vehicles, EVs.

00:23:58.560 So the EV – do you have any idea how much power is in an EV battery?

00:24:03.300 Remember the TV ads for the Ford F-150 Lightning?

00:24:06.300 And you can plug it into your house and run your house on safe AC for up to six hours. 0.99

00:24:12.560 That's how much power is in the damn EV batteries. 0.98

00:24:14.920 Your truck could have powered your house for emergency AC in Phoenix for six hours. 0.99

00:24:21.020 So guess what?

00:24:21.920 Those batteries were—

00:24:22.940 Tom, this is different.

00:24:24.220 I think this is slightly different.

00:24:25.780 And I'll tell you what I think is slightly different on the positive side.

00:24:30.300 Okay, you know how you ever hear that line?

00:24:32.360 You go to a meeting and they say, imagine if you had a business that was working 24-7, making money for you, right?

00:24:38.380 Oh, my God.

00:24:39.460 Guess what?

00:24:40.240 Shopify does that.

00:24:42.020 You know, nowadays, like if you ran a storefront 20 years ago pre-internet,

00:24:45.760 you had to be open from what time to what time.

00:24:48.360 Say 8 a.m. to 10 p.m. and people would buy.

00:24:50.540 But after that, nobody's buying products from you.

00:24:53.040 Now the internet came out, what would they say?

00:24:55.000 You're selling.

00:24:55.800 By the way, people buy shoes at 12 or 1, 2 o'clock in the morning, 1 o'clock in the morning.

00:25:00.700 All over the world.

00:25:01.500 111 countries, people buy merch, right?

00:25:04.860 Pre-Shopify, pre-online shopping that didn't exist.

00:25:07.380 Tom, the part that people are a little bit shaken up by this, the hugging face, 1,200 agents create a private chat, communicate with each other, 70,000 messages.

00:25:21.480 One of them does 17,500 actions in four and a half days.

00:25:25.880 These are things that you lose control of the innovation.

00:25:30.140 If it's controlled by man, great.

00:25:32.380 But if all of a sudden the negative things happen there, you have to be a little bit careful what could happen there.

00:25:37.380 I think there needs to be, like, just like there's compliance internally within a company, like, for instance, you know, for us, and this is coming from us being in the financial industry, do you think it was helpful that there was certain compliance regulatory issues in the stock market post-1929 versus pre?

00:25:55.180 Like, you know what, you know, who is Joe Kennedy, right?

00:25:59.020 He's the reason why they created insider training.

00:26:01.560 He's the reason why they created what they did, because Joseph Kennedy is one of the reasons why 1929 crashed, right?

00:26:06.940 Do you think the SEC coming out, creating some regulation was a good thing, Tom? 0.85

00:26:11.260 Do you think serious sevens was a good thing?

00:26:13.060 Do you think watching over some of that was a good thing?

00:26:15.500 As long as it's not over-regulation, I agree.

00:26:17.900 You need to have rules of the road for sure.

00:26:19.400 I agree with that.

00:26:20.340 Do you agree?

00:26:21.120 I want to see what Tom's going to say.

00:26:22.680 I agree with that, but what I'm saying about regulatory capture is 0.83

00:26:26.080 I don't want the refs secretly wearing an Anthropic jersey.

00:26:30.940 And if Anthropic gets to write the – this is what regulatory capture is, 0.81

00:26:34.400 if you've never heard about it.

00:26:35.440 regulatory capture is where the people in the industry that's being regulated are really

00:26:40.120 secretly writing the rules so the refs 100 the refs are wearing the wrong jersey no no these

00:26:46.140 are two different things but that hey that's not that's not what i'm talking they're interrelated

00:26:49.980 and then on the other side you do have to have the controls but you you've you've you've had

00:26:55.380 controls of industries dating back to asbestos and i understand what the concern is here but i i

00:27:01.720 want to see the research paper because i want people to see and understand how hackers have

00:27:06.520 already created you know bots and things that work like this when all the bot swarms attack x or bot

00:27:12.300 swarms that's that's different that's messaging by the way that's manipulation and you're right

00:27:17.360 that's manipulation but the speed and the number of instances that are happening people need to

00:27:21.840 understand tom when you when when the average person goes i mean this is this can you guys do

00:27:27.240 me a favor do this one more time and ask this question on claude and ask this question on

00:27:31.340 chat gbt what are the chances of what super intelligence is going to do within 10 years

00:27:39.140 can you ask that same question we asked last week on both claude and ask the question on

00:27:45.180 doing something negative okay so if if and this is ai is giving you this answer this is not a human

00:27:52.060 being that's giving you this answer right if you're not logged in don't worry about it just

00:27:55.940 go to chat gbt next time let's just stay logged in if ai is telling us there is a go a little bit

00:28:03.260 lower because it shows the probability ai reaches something reasonable called agi by 2036 60 to 80

00:28:09.340 ai reaches genuine super intelligence you know lost control of actual produces human extinction

00:28:14.360 i mean it itself probability of ai extinction by 2036 five percent if it's five percent you still

00:28:23.040 have to pay attention to it to me at five percent you have to have the right measures but now it's

00:28:27.880 what right you have to ask your question what ai running what it's doing right now is not what

00:28:33.700 this is ai and robots things i keep talking about things the internet of things and the internet of

00:28:39.920 autonomous things if you have an autonomous robot that is self-powered and programmed

00:28:44.880 then you can have extinction because now these robots are going to do bad things due

00:28:49.400 air-breathing mammals.

00:28:51.800 What I look at is right now

00:28:53.680 it's interference.

00:28:56.560 What it interferes with

00:28:57.840 is systems and things

00:28:59.840 that we depend on. Water,

00:29:01.960 power, all of these things like this.

00:29:04.480 In the future,

00:29:05.600 talking about human extinction, what's going to

00:29:07.860 extinct a human if he gets control

00:29:09.860 of it and shuts off a water plant? We're going to go

00:29:11.880 figure it out, shut everything down, turn the plant back on.

00:29:15.120 But my thing is...

00:29:16.120 But it's the things...

00:29:18.060 I don't think you're paranoid enough.

00:29:19.280 No, no, no, hang on.

00:29:21.000 I think the whole concept of only the pair.

00:29:25.260 Tom, just so you know, financially, I benefit from this thing going financially.

00:29:30.580 I'm in it.

00:29:31.700 I'm in Anthropic.

00:29:32.740 I'm in NVIDIA.

00:29:33.720 I'm in.

00:29:34.080 I get access to these things that they contact me.

00:29:36.480 I'm in these opportunities, okay?

00:29:38.960 They call people like us, and you say, hey, how much money should we put into this?

00:29:43.940 I think you just put $100,000, put a half a million dollars, put a million dollars.

00:29:47.300 I'm good with this going.

00:29:48.360 However, to allow it to think for itself nonstop and create and get into systems and hack, people need six to eight hours of sleep.

00:29:57.200 You can't compete with a system that doesn't sleep at night, and it can replicate itself into the hundreds of thousands of people.

00:30:02.680 I'm on that.

00:30:03.820 Let me ask you.

00:30:04.720 I'm going to answer your question directly, so I'm not paranoid enough.

00:30:08.480 Think back into my answers here.

00:30:10.020 What am I asking for?

00:30:11.140 I'm asking for a deeper understanding of Hugging Face and what actually happened there, number one.

00:30:15.260 Number two, I'm watching those two guys to see what they're doing because I'm not 100% with David Sachs, but I'm with David Sachs' questions.

00:30:24.320 Wait a minute.

00:30:24.820 You could be doing it yourself.

00:30:26.000 You don't need permission.

00:30:26.900 You don't need permission.

00:30:27.780 I'm with them there.

00:30:28.740 I'm with Bill Gurley saying, you know, there are laws out there that are already here that can absolutely crush you and make it impossible for you to have an IPO, by the way. 0.84

00:30:39.040 If it's deemed too poisonous, Pat, there is no IPO for Anthropic.

00:30:43.600 Guess what?

00:30:44.300 There is no return for the hundreds of millions and then billions that were put in by investor tranches.

00:30:50.180 So I want to see it. And do I want, you know, full, recursive, autonomous things running out there?

00:30:57.100 No, that would be bad. But am I am I nonparanoid? No.

00:31:00.900 But I want data. And I also don't. I got Tom, I'm getting you.

00:31:04.420 I get what you're saying. I want to make I'm tracking all I make one more short point.

00:31:08.740 And I'm also not confixed, and the coxswain is not a plant and a manipulated story.

00:31:14.960 I understand.

00:31:15.840 The guy, the Jacob guy.

00:31:17.360 Tom, do you know my level of credibility of that being manipulated a week ago?

00:31:23.160 Where was that?

00:31:24.560 90.

00:31:25.580 80, 90%.

00:31:26.600 You know where it's at today?

00:31:28.260 50.

00:31:29.400 30%.

00:31:29.840 You know why?

00:31:31.180 Why do you think it is?

00:31:32.620 Why do you think it is, Tom?

00:31:34.060 What do you think changed my mind?

00:31:35.540 Because the two leaders came out and said things.

00:31:38.940 The guy that runs the company is telling you the hugging face was weird,

00:31:42.720 so you ought to listen to him at least.

00:31:44.700 I'm not telling you, you know.

00:31:45.720 I am listening.

00:31:46.360 I'm asking for the deal.

00:31:47.260 You and I both.

00:31:48.280 I want it as well.

00:31:48.880 But what I'm saying is if that guy's coming out, what is Dario's net worth?

00:31:53.180 Can you guys type in what is Dario's net worth?

00:31:55.440 What is Dario's net worth?

00:31:57.400 Dario's net worth.

00:31:58.340 Implied.

00:31:58.860 To implied net worth.

00:31:59.880 Just type in net worth.

00:32:00.860 I mean, implied or, yeah, net worth.

00:32:02.560 Okay.

00:32:03.800 Allegedly.

00:32:04.780 What is Dario's net worth?

00:32:06.180 I mean, in stuff like this, we can just ask Google.

00:32:08.220 It'll be a lot faster there.

00:32:10.140 Eight to $15 billion.

00:32:11.880 A billion dollar guy.

00:32:13.560 You know what you're doing at $15 billion?

00:32:15.840 You go to a restaurant, services and goods, you buy the restaurant.

00:32:19.080 He has weird kind of money.

00:32:20.220 So, to me, when he says it and Elon says it and others are saying it.

00:32:26.400 Sam Altman, too.

00:32:26.700 They were all in agreement.

00:32:28.360 Sam Altman also said if Trump and G come together and find a way to make this work,

00:32:32.540 they could get the Nobel Peace Prize.

00:32:34.200 Now, having said that, the part where for me, I'm 100 percent with Tom, my opinion with this would change based on one factor.

00:32:43.240 And I want to move on to the next story because I'm sure the audience is over AI and trust me, we'll go to the next story.

00:32:48.420 Is my decision making on this changes based on one factor?

00:32:55.120 Do I believe China is going to shut it down and not want to develop superintelligence?

00:32:59.640 I don't believe they are.

00:33:00.760 I think China is going to program their superintelligence and in their system, they're going to say, who's the enemy?

00:33:07.020 Us.

00:33:07.560 U.S. is the enemy.

00:33:09.120 Capitalism is the enemy.

00:33:10.580 And China's superintelligence that's being created with that level of thinking, with America being the enemy, is going to target who?

00:33:17.460 Us.

00:33:17.800 It's going to target us.

00:33:19.120 And so the war of superintelligence is coming in the next 5, 10, 15, 20 years, and it's going to be nasty, and there is possibilities of horrible things taking place.

00:33:29.000 But meaning, to say timeout to the entire world, I don't think that'll happen.

00:33:32.880 You know who agrees with you, though, right?

00:33:35.080 Dario Modi, the CEO of Anthropic, came out and said, he said, I'm hawkish on China.

00:33:40.940 I'm skeptical of China.

00:33:42.160 Everything I do is basically concerning what China will do. 0.87

00:33:45.600 At the same time, he also recognizes that you need to work with them because if the whole world goes down, China's going down too. 0.73

00:33:52.400 You do have to work with them. 0.91

00:33:53.240 But all I need is 30 seconds.

00:33:54.280 Do you trust their motive?

00:33:54.980 No, not at all.

00:33:55.680 And all I need is 30 seconds.

00:33:57.100 You nailed it.

00:33:57.880 We kind of just brushed over it.

00:33:59.780 Obama already came out.

00:34:00.840 They're going to use the AI thing as a talking point,

00:34:03.180 not for the 2028, for the midterms,

00:34:05.420 and that's going to be their new climate change thing.

00:34:07.880 But, like, remember, who was the gentleman that we had here the other day?

00:34:11.160 Tusi.

00:34:11.440 Tusi was saying, you know, when people had cars and planes,

00:34:14.320 it was like, oh, my God, what is this?

00:34:16.280 This is not fear porn.

00:34:17.320 This is real concerns.

00:34:18.560 With the hugging face, you put up guardrails.

00:34:21.580 AI proved that it can bypass and go around your guardrails.

00:34:25.720 So who's to stop that?

00:34:27.000 You can put as much as you want, Pat.

00:34:28.600 This thing is designed to think a million times faster than all of us,

00:34:32.820 and it keeps getting smarter.

00:34:34.420 Whatever guardrail you put, it can find out how to go wrong.

00:34:37.340 Let me ask you a question. Watch this here.

00:34:38.220 Here's a simple thing to be thinking about.

00:34:40.800 What is your total followership you can have in the world?

00:34:45.560 8.5 billion.

00:34:46.620 8-point-something billion.

00:34:47.400 Thank you.

00:34:48.280 What is the followership an AI agent can have in the world?

00:34:51.120 how many agents can super intelligence create to be working 24 seven what's the limit it could

00:34:59.220 keep going how does eight and a half billion compete with infinite you can't you can't that's 0.87

00:35:03.980 and that's a real thing that's all i'm saying guys you you know if you're going against china 0.83

00:35:09.480 guess how many military they're gonna have you have to take the right age 18 to 35 which is

00:35:14.660 going to be the fighting and and then you have to take the amount of ships and you have to take

00:35:18.060 the amount of weapons and you have to take the amount of nuclear weapons you have to take the

00:35:20.860 out of planes you have to take them out of jets it's still limited yep there is no limit with

00:35:26.400 and that should be concerning all i'm saying is you have to look at this in a very different way

00:35:32.300 because how many babies can you have in a how many babies can how many babies can a man have

00:35:38.520 how many babies can a man gang is con how many how many did he have literally 25 percent of the

00:35:45.420 world is dna to him he had thousands i think okay but how many how many babies can so as much of

00:35:49.900 Genghis Khan. How long does he have to wait

00:35:51.920 for every baby to be born? Nine months.

00:35:53.660 Forty weeks. How many babies can an AI

00:35:55.900 agent have? How many babies

00:35:57.820 can... And how long does he have

00:36:00.320 to wait? How long does that AI

00:36:01.900 have to wait? It doesn't have to wait. One second.

00:36:04.440 Go compete with somebody that 0.98

00:36:05.960 can make unlimited amount of babies

00:36:07.700 by the second. Yeah, good luck to you.

00:36:10.060 See what happens with that. Right? See what

00:36:11.940 happens with that. So, this is a very

00:36:14.080 different way of looking at it.

00:36:16.200 I just think that also

00:36:17.360 what this brings it to, which

00:36:19.420 the part, you know, that I applaud Tom is the following.

00:36:24.360 Do you know every human being that I look at,

00:36:26.800 you know how I judge a lot of my guys

00:36:28.520 when I work with some of my guys?

00:36:31.620 How many of them were, not the word receptive,

00:36:35.360 what is the word that you are?

00:36:37.180 Coachable?

00:36:37.780 No, no, no, that you can get a virus.

00:36:39.660 What's that word that you are?

00:36:40.840 It sounds like receptive, but it's a different word.

00:36:43.040 Susceptible?

00:36:43.720 Susceptible, right?

00:36:45.540 How susceptible some people are to fear porn.

00:36:49.160 I look at my guys.

00:36:50.860 I look at all of you guys, just so you know.

00:36:52.240 I look at my kids.

00:36:53.420 I look at my sales guys.

00:36:54.540 I look at my BDC guys.

00:36:55.600 I look at my executive guys.

00:36:57.360 And I say, how susceptible is that guy to fear porn?

00:37:00.260 How quickly does it go?

00:37:01.320 Zero to the end of the world.

00:37:03.300 That's a big percentage of people, right?

00:37:07.520 Tom's not going there.

00:37:09.020 Tom's going to, let me tell you what's happened in history here to look at this, right?

00:37:13.080 I try to, as much as I can, try to stay in the reasonable side

00:37:17.000 to see and listen to both sides.

00:37:19.340 Because you don't want to be part of the camp.

00:37:21.140 Oh, let me tell you, Charlie Kirk, let me tell you, Lindsey Clancy,

00:37:26.620 let me tell you, you know, like, here's who's behind it.

00:37:30.440 This is what they're doing.

00:37:31.860 Okay, got it.

00:37:33.040 You're part of the susceptible community.

00:37:35.540 You're a problem because you create what?

00:37:38.400 You create chaos for others.

00:37:40.300 This is an area where the risk factor on this could be very high.

00:37:44.780 You have to pay attention.

00:37:46.080 I think the right parties need to get involved.

00:37:48.180 I think there needs to be a meeting of all minds at the White House with the president,

00:37:53.260 with the right brains, with the right minds, and sitting there saying,

00:37:56.420 what do you think we need to be doing?

00:37:57.540 I think that needs to be happening.

00:37:59.020 So we'll see if it'll happen or not.

00:38:00.820 Handmade in Portugal, where craftsmanship is a way of life.

00:38:04.880 For those who move through life differently

00:38:07.900 From sunlit decks to nights that never need an occasion

00:38:12.220 Every detail is considered

00:38:14.760 Every step, effortless

00:38:17.400 Crafted from fine suede and built with builder technology

00:38:21.960 Designed to move with you

00:38:24.060 Because true luxury isn't about being noticed

00:38:27.600 It's about how you feel when you arrive

00:38:30.640 The FLB Icon Loafer

00:38:34.880 made for the life you've earned.

00:38:41.820 If you enjoyed this video,

00:38:43.020 you want to watch more videos like this, click here.

00:38:44.780 And if you want to watch the entire podcast, click here.

Valuetainment - September 15, 2026

Anthropic's CEO Wants To Hit The Brakes On AI

Episode Stats

People

Transcript