WEBVTT - AI Regulation Talk and Tesla AGM 0:00:01.360 --> 0:00:05.680 From Marhart. We're Innovation, Money and Power Collie in Silicon 0:00:05.800 --> 0:00:10.200 Vallet NBN. This is Bloomberg Technology with Caroline Hyde and 0:00:10.440 --> 0:00:11.040 Ed Ludlow. 0:00:25.680 --> 0:00:28.280 Ed Ludlow here in San Francisco. Caroline Hyde off today. 0:00:28.360 --> 0:00:32.080 This is Bloomberg Technology. Massive show coming out. Full coverage 0:00:32.200 --> 0:00:35.800 on artificial intelligence. We speak to someone who testified before 0:00:36.080 --> 0:00:38.920 lawmakers just twenty four hours ago on the risks of 0:00:38.960 --> 0:00:41.880 the technology, as well as the CEO of character AI 0:00:42.200 --> 0:00:45.320 and Dina Trace. Last, we bring you the biggest takeaways 0:00:45.320 --> 0:00:48.120 from the Tesla AGM and break it all down with 0:00:48.240 --> 0:00:51.680 Tasha Kini of ARC invest and we'll discuss the state 0:00:51.680 --> 0:00:54.680 of venture capital investing and go life to sol Y 0:00:54.760 --> 0:00:58.760 Connections Forum focus on global finance, tech and public policy. 0:00:58.800 --> 0:01:00.360 But first it's going to check on these markets. 0:01:00.520 --> 0:01:02.800 There's a lot of attention from Wall Street on what's 0:01:02.800 --> 0:01:05.480 happening in DC and the debt ceiling, a smaller group 0:01:05.760 --> 0:01:10.240 trying to accelerate negotiations. There's fighting talk from the President 0:01:10.280 --> 0:01:11.440 and Chuck Schumer as well. 0:01:11.680 --> 0:01:12.839 How that looks in markets? 0:01:12.840 --> 0:01:15.800 Now's that one hundred up half a percentage point outperformance 0:01:15.800 --> 0:01:18.160 in semiconductors. You look at names, I can video AMD 0:01:18.280 --> 0:01:20.240 a number of names in that basket up in the 0:01:20.319 --> 0:01:24.400 high single digits, some of it AI related, I am sure. 0:01:24.400 --> 0:01:26.920 In terms of the bomb market, US Tenure, Yeald three 0:01:26.959 --> 0:01:30.640 point five six percent a little higher, and Bitcoin interesting. 0:01:30.640 --> 0:01:34.000 We're back now below twenty seven thousand US dollars per token. 0:01:34.000 --> 0:01:36.480 We'll bring you some details of newsflow in the crypto 0:01:36.560 --> 0:01:39.679 space later in the program in terms of specific names 0:01:39.680 --> 0:01:42.639 and movers. One piece of news out this Wednesday morning 0:01:42.680 --> 0:01:46.319 is Amazon with new Echo devices, a real emphasis on 0:01:46.360 --> 0:01:50.680 Alexa and part of the narrative around generative AI tools. 0:01:50.720 --> 0:01:52.840 I guess in the voice context of what they'll do 0:01:53.080 --> 0:01:54.760 to push back in some of the advances of their 0:01:54.800 --> 0:01:57.760 peers like Alphabet Dina, Trace now down one point four percent, 0:01:57.800 --> 0:02:01.320 have been much higher in pre market lease on strong 0:02:01.360 --> 0:02:04.280 earnings beating the quarter gone in a full year outlook, 0:02:04.320 --> 0:02:06.840 there was above expectations. We will speak to the CEO 0:02:07.120 --> 0:02:10.120 letter later in the program. Then Tesla up four percent, 0:02:10.320 --> 0:02:13.600 now really accelerating in terms of its gains. The big 0:02:13.639 --> 0:02:16.079 takeaways out of the annual general meeting. They will now 0:02:16.120 --> 0:02:19.519 look at advertising in a limited way. Jbi Strawbell added 0:02:19.520 --> 0:02:22.320 to the board and Elon Musk is there to stay. 0:02:22.360 --> 0:02:24.080 He will remain as CEO. 0:02:24.400 --> 0:02:26.840 He's going to really focus now a lot more on Tesla, 0:02:26.880 --> 0:02:29.680 having of course been distracted by Twitter in recent weeks 0:02:29.720 --> 0:02:32.320 and months. Let's stick with a Tesla story. Yesterday, CEO 0:02:32.360 --> 0:02:35.200 Elon Musk held that company and your shareholder meeting, and 0:02:35.200 --> 0:02:39.000 while taking questions from investors, he excited the crowd basically 0:02:39.160 --> 0:02:39.799 by saying this. 0:02:40.840 --> 0:02:42.920 And although there's there's obviously a lot of people that 0:02:43.360 --> 0:02:47.480 follow like to say the Tesla count and the you 0:02:47.480 --> 0:02:52.280 know my account whatever on Twitter, to some degree, it 0:02:52.360 --> 0:02:55.560 is preaching to the choir, and the choir is already convinced. 0:02:57.360 --> 0:03:01.280 So I think what you're saying does does have some merits, 0:03:01.360 --> 0:03:06.639 and you know what, I believe in taking taking suggestions. 0:03:06.680 --> 0:03:10.400 So we'll try a little advertising and see how it goes. 0:03:11.840 --> 0:03:14.160 Let's get into Tesla and his ad shift with Tasha 0:03:14.240 --> 0:03:18.040 Kini Ark invest director of Investment Analysis. That's choir as 0:03:18.120 --> 0:03:23.600 convinced Tasha, does advertising convince others that aren't in the choir? 0:03:26.240 --> 0:03:28.840 Yeah, I mean, I think it's certainly the point that 0:03:28.880 --> 0:03:32.040 everyone is harping on here. You know, I think generally 0:03:32.360 --> 0:03:35.040 it's not a bad idea to at least try advertising, 0:03:35.520 --> 0:03:38.520 right I mean from our research, we know that on 0:03:38.960 --> 0:03:42.000 both a sticker price basis now and a total cost 0:03:42.000 --> 0:03:45.600 of ownership basis, evise are better cars. Right, They're cheaper, 0:03:45.640 --> 0:03:49.040 they're more performant. So you know, if you're not buying 0:03:49.040 --> 0:03:51.760 an electric vehicle, the question is why. Maybe you know 0:03:51.960 --> 0:03:54.920 that information just hasn't reached you yet. Maybe an ad will. 0:03:55.400 --> 0:03:58.360 So I don't think it's a bad idea there, you know, 0:03:58.520 --> 0:04:01.840 I'd say what I was most excited about from last 0:04:01.920 --> 0:04:05.200 night was all the talk around autonomy. You know, Elon 0:04:05.280 --> 0:04:08.040 said he thinks this will be the greatest asset unlock 0:04:08.120 --> 0:04:11.200 in human history, and i'd agree. You know, we've modeled 0:04:11.200 --> 0:04:14.520 that autonomous driving could add an additional twenty six trillion 0:04:14.560 --> 0:04:17.200 dollars to GDP in the next ten years. It is 0:04:17.240 --> 0:04:20.080 going to totally change touslis business model and we're e's 0:04:20.080 --> 0:04:20.600 ated for it. 0:04:21.400 --> 0:04:23.240 I want to get into a five year forecast. But 0:04:23.400 --> 0:04:26.600 let's react to the other announcements, which is JB. Strubble 0:04:26.640 --> 0:04:29.320 being added to the board. And also you know the 0:04:29.360 --> 0:04:32.800 comments from Elon Musk about his focus on Tesla remaining 0:04:32.839 --> 0:04:36.960 as CEO, particularly to get the company through its next 0:04:37.279 --> 0:04:42.080 focus on artificial intelligence. What did you make of that, Tasha, Yeah, 0:04:42.120 --> 0:04:43.039 you know, I think JV. 0:04:43.080 --> 0:04:45.800 Strubble is a great addition to the board, right you know, 0:04:46.040 --> 0:04:48.919 right now he's working with his own company on battery cycling. 0:04:48.960 --> 0:04:52.000 We know that's crucial to scale the battery industry as 0:04:52.000 --> 0:04:54.559 a whole. They're seeing great efficiency, I think over ninety 0:04:54.560 --> 0:04:57.840 percent and recycling. You know, we've always thought that it's 0:04:57.880 --> 0:05:00.400 important for Elon to stay on as CEO as they 0:05:00.400 --> 0:05:03.040 reach full autonomy again because we think that this is 0:05:03.720 --> 0:05:08.080 the next greatest milestone for Tesla. And you know, arguably, 0:05:08.120 --> 0:05:10.080 you know, any automaker out out there, this is what 0:05:10.080 --> 0:05:12.320 they really should be going for. So we think that 0:05:13.200 --> 0:05:17.200 autonomous driving or autonomous ridehill, which is how we expected 0:05:17.200 --> 0:05:20.479 to play out, will constitute roughly two thirds of Tesla's 0:05:20.560 --> 0:05:24.160 enterprise value in the next five years. So this is huge. 0:05:24.240 --> 0:05:27.720 Tesla has an enormous data advantage and you know you 0:05:27.720 --> 0:05:28.400 shouldn't miss it. 0:05:30.200 --> 0:05:34.800 Investors or shareholders rejected a proposal for Tesla to publish 0:05:34.880 --> 0:05:38.000 a key person risk report, and Tessa's argument was, well, 0:05:38.000 --> 0:05:39.920 if we put all of our talent in the shop window, 0:05:39.960 --> 0:05:42.880 our competitors will try and take them. But do you 0:05:43.000 --> 0:05:45.600 still have a key man risk concern around Elon Musk. 0:05:45.680 --> 0:05:48.360 I guess, you know, a new CEO at Twitter goes 0:05:48.400 --> 0:05:51.360 some way to allaying those concerns. 0:05:53.800 --> 0:05:56.159 Yeah, you know, I think there's a lot of focus 0:05:56.160 --> 0:05:58.839 on this question. What I what I'd broadly say is 0:05:59.160 --> 0:06:00.919 I'd be a lot more concerned if he said he 0:06:01.000 --> 0:06:04.200 was stepping away from Tesla. Right again, I think he's 0:06:04.240 --> 0:06:08.200 crucial to cross that autonomy finish line. You know, autonomous 0:06:08.279 --> 0:06:11.760 driving lowers the cost per mile of ride hill significantly 0:06:11.839 --> 0:06:14.120 invites a lot of people that are not currently in 0:06:14.160 --> 0:06:16.480 the red hill market into it. You know, it's going 0:06:16.520 --> 0:06:19.360 to be a multi trillion dollar industry over the next 0:06:19.400 --> 0:06:23.960 five years. So, you know, and on Elon's time. You know, 0:06:24.279 --> 0:06:27.320 we've heard this question come up so many times over 0:06:27.360 --> 0:06:30.520 the past five years. I mean, Tesla's SpaceX. You know, 0:06:30.600 --> 0:06:33.280 he was one of the founders of Open AI. He's 0:06:33.360 --> 0:06:36.000 clearly demonstrated that he can juggle a lot of tasks 0:06:36.040 --> 0:06:39.919 and ultimately we're just focused on his execution, right and 0:06:40.000 --> 0:06:42.640 I think given that Tesla has the best you know, 0:06:42.800 --> 0:06:47.440 cost relative to performance of any electric vehicle maker out there, 0:06:47.920 --> 0:06:50.520 you know, he's already proved sort of that that he 0:06:50.560 --> 0:06:52.840 can juggle multiple things and succeed at it. 0:06:54.400 --> 0:06:58.960 There was a demonstration of progress, apparent progress in Optimus 0:06:59.040 --> 0:07:00.400 that the humanoid. 0:07:01.440 --> 0:07:01.800 Bolts. 0:07:02.600 --> 0:07:05.279 When you saw that, did it give you any feeling 0:07:05.400 --> 0:07:07.679 that they are making progress in the field of AI 0:07:08.240 --> 0:07:10.280 in that use case at least? 0:07:12.040 --> 0:07:12.280 Yes? 0:07:12.400 --> 0:07:15.440 I think Optimist is really interesting. So it's definitely out 0:07:15.480 --> 0:07:18.800 of our five year forecasting window. You know, our our 0:07:18.920 --> 0:07:21.960 price target for Teslas that our expected value is roughly 0:07:21.960 --> 0:07:24.800 two thousand dollars per share in twenty twenty seven. That 0:07:24.840 --> 0:07:27.880 actually doesn't include Optimists because I think that you know, 0:07:27.880 --> 0:07:30.400 by the time it's a meaningful contributor. Again, it could 0:07:30.440 --> 0:07:33.720 be further into the future. But I think the important 0:07:33.720 --> 0:07:36.000 thing to look at here is you know why Tesla, 0:07:36.200 --> 0:07:39.480 Why why humanoid robot? Well, given that they have this 0:07:39.600 --> 0:07:42.760 massive data advantage from autonomous driving. You know, they have 0:07:42.840 --> 0:07:46.040 millions of cars on the road, they're collecting over a million, 0:07:47.000 --> 0:07:50.760 they have over a million miles driven in FSD daylight daily, 0:07:50.920 --> 0:07:53.600 which they can then pull data from. You know, that's 0:07:53.840 --> 0:07:57.480 an order of magnitude more than competitors. So I think 0:07:57.560 --> 0:08:01.440 this sets them up well to base create robots that 0:08:01.520 --> 0:08:04.440 move through the physical world and the autonomous car will 0:08:04.440 --> 0:08:07.000 be the first version of that, but they can you know, 0:08:07.080 --> 0:08:10.280 poort that knowledge over into other robots, and I think 0:08:10.800 --> 0:08:14.320 the Optimist robot is a good example. You know, we 0:08:14.360 --> 0:08:17.240 talk a lot at ARC about backwards integration. I think 0:08:17.280 --> 0:08:19.280 the fact that this is a humanoid robot and it 0:08:19.280 --> 0:08:22.200 can move through spaces that were already built for humans 0:08:22.880 --> 0:08:25.440 is big. And I think this, you know, in the future, yeah, 0:08:25.480 --> 0:08:30.320 could be a major productivity enhancer like all AI. You know, 0:08:30.360 --> 0:08:33.600 we'll see you know, non market labor activity turn into 0:08:33.600 --> 0:08:37.320 market labor activity, and I think the productivity of the 0:08:37.400 --> 0:08:40.199 individual worker will ultimately increase because of this. So we're 0:08:40.240 --> 0:08:40.960 excited about it. 0:08:42.000 --> 0:08:44.680 So your twenty twenty seven coal is two thousand dollars 0:08:44.720 --> 0:08:46.880 per share, which I'll give you an opportunity to explain. 0:08:47.679 --> 0:08:51.120 But the question is, did anything in that presentation or 0:08:51.160 --> 0:08:54.240 in the interview with CNBC afterwards move the needle for 0:08:54.280 --> 0:08:57.320 you guys change the course of where you see this 0:08:57.400 --> 0:08:58.000 company going. 0:09:00.600 --> 0:09:03.079 Yeah, I think again, since we're so focused on the 0:09:03.440 --> 0:09:07.880 long term horizon, that five year view, you know, I 0:09:07.880 --> 0:09:10.440 would stick to that two thousand dollars per share estimate 0:09:10.480 --> 0:09:14.280 that we put out, And you know, I think again, 0:09:15.000 --> 0:09:19.079 what I'm more looking forward to is everyone else realizing 0:09:19.280 --> 0:09:23.160 the opportunity and autonomous driving, because I actually think that 0:09:23.200 --> 0:09:27.440 Tesla does not get enough questions about this opportunity, given 0:09:27.480 --> 0:09:30.520 how monumental it will be. So I was glad that 0:09:30.559 --> 0:09:34.160 we heard Elon talk about that last night again because 0:09:34.160 --> 0:09:37.760 I think that most don't truly understand how big of 0:09:37.800 --> 0:09:42.320 a productivity unlock and a cash flow unlock, because it 0:09:42.360 --> 0:09:45.520 will be because we think autonomous ride hill could have 0:09:45.920 --> 0:09:50.120 software like Margins, and we heard affirmation of that last night. 0:09:51.200 --> 0:09:55.079 So again, it's a recurring revenue stream software like Margins, 0:09:55.600 --> 0:09:58.320 So you know, it's up to let's say, roughly ten 0:09:58.360 --> 0:10:02.040 thousand dollars in cash flower car per year. I mean 0:10:02.040 --> 0:10:06.080 that's unheard of, and no other traditional automaker is, you know, 0:10:06.200 --> 0:10:08.959 close to the data advantage that Tesla has at least, 0:10:09.800 --> 0:10:11.520 so they're in a major position there. 0:10:12.040 --> 0:10:14.480 Yeah, it's interesting because ahead of you coming on, a 0:10:14.559 --> 0:10:16.640 number of people tweeted at me saying, actually they wanted 0:10:16.679 --> 0:10:19.040 to hear a little bit more about the plan for 0:10:19.160 --> 0:10:21.760 autonomous driving or at least the robotaxi vision of the 0:10:21.760 --> 0:10:25.560 future are Thanks to Tasha Kini Promark invest for that reaction, 0:10:25.640 --> 0:10:37.760 to Tesla's AGM before a Senate Judiciary subcommittee on Tuesday. 0:10:37.800 --> 0:10:41.800 Open AI CEO Sam Outman praised AI's potential, but warned 0:10:41.800 --> 0:10:45.400 that the emerging technology is powerful enough to change society 0:10:45.400 --> 0:10:48.760 in unpredictable ways. Joining us now as someone who also 0:10:48.880 --> 0:10:53.040 testified on the Hill alongside him, NYU Professor emeritus and 0:10:53.120 --> 0:10:56.200 Geometric Intelligence founder Gary Marcus. Of course, you're also the 0:10:56.200 --> 0:10:59.920 host of the podcast Humans Versus Machines, which is kind 0:11:00.040 --> 0:11:03.240 kind of the debate that we're having as a nation 0:11:03.440 --> 0:11:06.680 and globally. Right now, let me ask you this, what 0:11:06.760 --> 0:11:08.600 good came out of yesterday's hearing? 0:11:08.679 --> 0:11:09.760 What was the net result? 0:11:11.240 --> 0:11:14.320 I thought the hearing was actually fantastic. It far exceeded 0:11:14.360 --> 0:11:18.240 my own expectations that I think probably most others. It 0:11:18.280 --> 0:11:21.560 was really bipartisan, and I think we all agreed there. 0:11:21.600 --> 0:11:26.079 Almost everybody except the IBM executive all agreed that we 0:11:26.120 --> 0:11:30.480 need to have some kind of national agency governing AI, 0:11:30.960 --> 0:11:34.400 and probably want some global agency doing that. Sam Altman 0:11:34.480 --> 0:11:36.440 was supportive of that. That's an idea I've been pushing 0:11:36.480 --> 0:11:38.480 for the last month or two. So it was wonderful 0:11:38.480 --> 0:11:42.360 to have his endorsement and the government or the Senators 0:11:42.360 --> 0:11:45.520 as a whole were pretty positive towards it. I think 0:11:45.640 --> 0:11:48.719 the notion is that the United States should try to 0:11:48.800 --> 0:11:50.679 lead the way if we're going to do something global, 0:11:50.880 --> 0:11:54.679 and I hope that will happen. And there was also 0:11:54.760 --> 0:11:59.559 strong support for having something like FDA kind of regulations 0:11:59.679 --> 0:12:02.640 where you have a sufficiently large model, you need to 0:12:02.640 --> 0:12:05.760 show that it is sufficiently safe. 0:12:05.800 --> 0:12:08.000 You can't just release something to one hundred million people. 0:12:08.040 --> 0:12:10.280 So there's lots to discuss, but I thought it was 0:12:10.320 --> 0:12:13.959 a very positive atmosphere, very bipartisan, and people recognized how 0:12:13.960 --> 0:12:15.000 serious the problems were. 0:12:15.160 --> 0:12:16.880 And also I think there was a lot of. 0:12:18.840 --> 0:12:20.679 Sense from the senators that they feel like they didn't 0:12:20.679 --> 0:12:22.839 do the right thing yes with Section two thirty in 0:12:22.880 --> 0:12:25.920 the Internet, and that they wanted to do better this 0:12:26.000 --> 0:12:27.880 time in there here a lot faster. You know, it 0:12:27.880 --> 0:12:30.600 took them like fourteen years after social media before they 0:12:30.640 --> 0:12:33.880 really did anything, and you know, it's only six months 0:12:33.920 --> 0:12:36.280 after shat GPT made it big that they're trying to 0:12:36.400 --> 0:12:38.920 yes handle it. So I saw a lot of seriousness 0:12:38.960 --> 0:12:39.960 on the part of the senators. 0:12:40.200 --> 0:12:40.680 Gary. 0:12:41.400 --> 0:12:44.200 I invited our audience to put forward questions. A lot 0:12:44.200 --> 0:12:47.040 of people interested on you coming on the program, We'll 0:12:47.080 --> 0:12:50.640 give you a second to take on some water. There 0:12:50.679 --> 0:12:53.400 are many that tweeted at me saying that what you 0:12:53.440 --> 0:12:59.000 were doing was scare mongering. Is that a fair accusation 0:12:59.120 --> 0:13:00.480 that they've led at you. 0:13:02.200 --> 0:13:02.920 I don't think so. 0:13:02.960 --> 0:13:05.600 I mean, it's true that I'm trying to raise alarms 0:13:05.640 --> 0:13:08.000 about things that I think are genuinely risky. But I 0:13:08.040 --> 0:13:11.160 think scare mongering is if you don't actually think that 0:13:11.240 --> 0:13:12.000 there's a risk. 0:13:11.840 --> 0:13:13.520 And you're just trying to manipulate people. And I think 0:13:13.520 --> 0:13:14.200 the real risk. 0:13:14.600 --> 0:13:17.839 I've put all my own effort, I'm not getting paid 0:13:17.840 --> 0:13:20.800 for this into trying to help us address those risks 0:13:20.800 --> 0:13:23.120 because I think they're real. Sam thought they were real too, 0:13:23.200 --> 0:13:25.800 you know. Sam agreed with me. Yes, there were risks 0:13:25.920 --> 0:13:28.840 to our elections and possibly much graver risks in the 0:13:28.840 --> 0:13:30.920 long term if we don't figure out how to control 0:13:30.920 --> 0:13:31.880 our AI systems. 0:13:33.080 --> 0:13:37.560 One of the proposals that Sam put forward was to establish, 0:13:37.600 --> 0:13:40.520 at least here in the United States, an agency to 0:13:40.760 --> 0:13:44.280 license or have some sort of license seeing system for 0:13:44.320 --> 0:13:47.360 the development of AI. There are many that believe that 0:13:47.400 --> 0:13:54.800 would basically centralize activity control power among the biggest tech companies. 0:13:55.840 --> 0:13:58.680 What's your response to that, I mean, just to clarify. 0:13:58.840 --> 0:14:02.600 Even before this meeting, I wrote our ed and Economist, 0:14:02.640 --> 0:14:05.560 and I gave a TED talk on April eighteenth about 0:14:05.600 --> 0:14:08.800 having an international agency to regulate AI, and. 0:14:08.880 --> 0:14:10.720 Sam was supportive of that. 0:14:11.600 --> 0:14:15.120 He emphasized licensing, and he emphasized excuse me, sorry about 0:14:15.120 --> 0:14:18.280 the cough. I've been doing so many interviews. He emphasized 0:14:18.400 --> 0:14:25.000 licensing for large scale models, not necessarily for small models. 0:14:25.000 --> 0:14:26.760 I think we all agree that we don't want to 0:14:26.800 --> 0:14:28.760 cut off research. We don't want to cut off small 0:14:28.800 --> 0:14:32.080 companies from having a role. But the larger the model, 0:14:32.120 --> 0:14:34.560 the larger the impact, the more we might need licensing. 0:14:37.400 --> 0:14:39.920 One of the questions from our audience is for you 0:14:39.960 --> 0:14:44.720 to explain what is the difference between fooling a human 0:14:45.200 --> 0:14:50.320 with artificial intelligence and fooling a system with artificial intelligence. 0:14:52.120 --> 0:14:53.840 I'm not sure what you mean by fooling a system 0:14:53.840 --> 0:14:55.560 with artificial intelligence, but I think. 0:14:55.440 --> 0:14:57.800 What I mean is that if you're a consumer and 0:14:57.800 --> 0:15:01.120 you're confronted by information from a generative a tool and 0:15:01.200 --> 0:15:03.640 it's false, you have fooled. But you can also use 0:15:03.880 --> 0:15:07.560 llms and foundation models to automate all kinds of processes 0:15:07.600 --> 0:15:11.640 from SaaS through to internal workflow, whatever it will be. 0:15:11.680 --> 0:15:14.360 And I think the root of that question is what 0:15:14.480 --> 0:15:17.880 is the greater risk about human interaction with AI or 0:15:18.000 --> 0:15:20.400 broadly automation that comes from those tools. 0:15:22.840 --> 0:15:25.040 That's a good question, I guess. You know, there are 0:15:25.040 --> 0:15:27.680 different risks of different timescales. The thing that I'm most 0:15:27.760 --> 0:15:31.120 currently worried about is the risk to democracy, and the 0:15:31.200 --> 0:15:34.560 risk to democracy actually comes both from kind of accidental 0:15:36.280 --> 0:15:38.600 mistakes from these systems. We know they can fabulate, or 0:15:38.600 --> 0:15:41.840 some people call it hallucinate. They do that automatically without 0:15:41.880 --> 0:15:45.160 human intervention. They're just not a very reliable technology, and 0:15:45.200 --> 0:15:48.160 then bad actors can deliberately use them to make enormous 0:15:48.200 --> 0:15:51.840 amounts of misinformation that's incredibly plausible. They can do that 0:15:51.840 --> 0:15:53.880 with deep fix for images, they can do it with 0:15:53.960 --> 0:15:57.720 TEX and so they're both the fact that the systems 0:15:57.720 --> 0:16:00.240 are unreliable, they don't know what they're talking about, can't 0:16:00.280 --> 0:16:02.360 verify what they're saying, and they can be abused. So 0:16:02.440 --> 0:16:05.520 it's kind of like both similarly, in terms of long 0:16:05.600 --> 0:16:08.320 term risk, you can think of deliberate scenarios where people 0:16:08.360 --> 0:16:11.760 try to manipulate the markets and things go wrong and 0:16:12.600 --> 0:16:17.240 there's violence because people misattribute things that are happening, or 0:16:18.280 --> 0:16:20.920 maybe machines do things that are entirely different from what 0:16:20.920 --> 0:16:24.080 we program them do. So I don't know if you 0:16:24.080 --> 0:16:27.240 can neatly dichotomize it them that way. The reality is 0:16:27.240 --> 0:16:31.640 the machines are not reliable, and the machines are not 0:16:31.840 --> 0:16:33.680 very well controlled, and so that leads to all kinds 0:16:33.680 --> 0:16:34.160 of risks. 0:16:35.680 --> 0:16:39.760 Gary, why is the large language model approach not the 0:16:39.840 --> 0:16:43.840 right approach to achieve AGI in your opinion? 0:16:45.240 --> 0:16:47.360 Well, I just saw other words on your screen safer 0:16:47.400 --> 0:16:51.920 and more relied, more aligned. Excuse me, These systems aren't 0:16:51.960 --> 0:16:54.680 that safe. They're not that sophisticated. They don't have a 0:16:54.720 --> 0:16:56.880 model of the world. They don't understand what's going on, 0:16:57.280 --> 0:16:59.600 and so, for example, they make stuff up all the time. 0:16:59.640 --> 0:17:03.120 They up a sexual harassment charge. They said Elon Musk 0:17:03.280 --> 0:17:06.159 was dead when he's alive. I mean, all kinds of craziness. 0:17:06.440 --> 0:17:07.000 They're just not. 0:17:07.320 --> 0:17:10.960 Systematic, trustworthy bits of AI. You know, when we look 0:17:11.000 --> 0:17:13.080 back twenty years ago, I mean twenty years from now, 0:17:13.119 --> 0:17:15.200 it'll be like looking back at cell phones from twenty 0:17:15.320 --> 0:17:17.560 years ago. It's like they had a phone that big. 0:17:17.560 --> 0:17:20.600 We're going to say they used AI that was that unreliable? 0:17:20.640 --> 0:17:21.439 Like, what were they. 0:17:21.320 --> 0:17:27.159 Thinking, Gary, what do you want to happen next? You know, 0:17:27.600 --> 0:17:33.200 you actually quite praising or complementary of the hearing itself 0:17:33.280 --> 0:17:37.119 it being bipartisan. But if there is a concrete step 0:17:37.200 --> 0:17:39.919 for law makers or regulators to take, what does it 0:17:39.960 --> 0:17:41.040 look like to your mind? 0:17:42.680 --> 0:17:45.280 I think the next step is to actually figure out 0:17:45.400 --> 0:17:48.800 what regulation would look like. I think we probably need 0:17:48.920 --> 0:17:52.320 a cabinet level agency and I think we should start 0:17:52.440 --> 0:17:55.400 drafting plans for how that would work. And there's lots 0:17:55.400 --> 0:17:57.639 of complications in terms of how it would work with 0:17:57.720 --> 0:18:00.639 existing agencies, how it would work international. I think we 0:18:00.680 --> 0:18:03.680 should start, you know, taking the consensus that we've got 0:18:03.720 --> 0:18:07.600 and try to represent what that might look as actual legislation. 0:18:08.960 --> 0:18:11.560 And finally, Gary, you talked about the need for a 0:18:11.600 --> 0:18:15.720 global regulator or a global agency. Do you recognize a 0:18:15.760 --> 0:18:18.960 sort of multi speed approach what Europe is doing, what 0:18:19.040 --> 0:18:21.920 China is doing when it comes to the regulation and 0:18:22.240 --> 0:18:24.879 development of AI technology. 0:18:25.080 --> 0:18:26.879 So I think we have an opportunity here to do 0:18:26.920 --> 0:18:30.240 something rational rather than just sort of arbitrary. In Balkani 0:18:30.440 --> 0:18:33.280 so it's not actually in the interest of the AI 0:18:33.359 --> 0:18:35.960 companies if we have like one hundred and ninety three 0:18:36.359 --> 0:18:39.560 or one hundred and ninety five rivers. You know, many 0:18:39.600 --> 0:18:43.320 different places where you have to train your own language model. 0:18:43.560 --> 0:18:46.080 So as you probably well know, it's very expensive to 0:18:46.119 --> 0:18:49.840 train these models. It's very costly in terms of climate impact. 0:18:50.040 --> 0:18:53.000 And so if everybody is requiring their own set of rules, 0:18:53.119 --> 0:18:55.600 their own set of giant language model that takes you know, 0:18:55.680 --> 0:18:58.160 millions of dollars to train and you know a certain 0:18:58.240 --> 0:19:00.520 number of jet flights in terms of emissions and so forth, 0:19:00.640 --> 0:19:01.640 that's not a great thing. 0:19:01.680 --> 0:19:02.400 And so I think the. 0:19:02.359 --> 0:19:05.199 Companies themselves would like some kind of alignment here, some 0:19:05.280 --> 0:19:06.520 kind of systematic. 0:19:06.119 --> 0:19:07.240 Way of doing business. 0:19:07.280 --> 0:19:09.560 And then you know, there's all this talk about global 0:19:09.560 --> 0:19:10.960 tension and stuff like that, and some of it is 0:19:11.000 --> 0:19:14.119 of course real, but in terms of AI, like, no 0:19:14.240 --> 0:19:17.920 country wants their citizens to be completely overwhelmed by misinformation, 0:19:18.040 --> 0:19:21.720 nobody wants to be completely overwhelmed by cybercrime, and nobody wants, 0:19:21.760 --> 0:19:23.480 you know, robots to take over the world, which is 0:19:23.480 --> 0:19:25.199 not an immediate concern, but in the long term we 0:19:25.200 --> 0:19:27.080 do have to make sure we get that right. So 0:19:27.160 --> 0:19:29.320 I think even if you know some countries are going 0:19:29.359 --> 0:19:31.960 to do some things differently. I think there's a lot 0:19:32.000 --> 0:19:35.720 of intersection between what different countries want and even the companies, 0:19:35.800 --> 0:19:38.080 you know, want some alignment here. So I think there's 0:19:38.119 --> 0:19:40.840 a real opportunity, even though, of course the politics are difficult. 0:19:41.880 --> 0:19:45.360 M YU Professor Emeritis Gary Marcus, also the host, of course, 0:19:45.359 --> 0:19:48.159 of the podcast Humans Versus Machines, were very grateful for 0:19:48.200 --> 0:19:48.600 your time. 0:19:48.680 --> 0:19:50.520 Thank you. Out of Washington, d C. 0:19:51.520 --> 0:19:55.440 Now sticking with AI, Goldman Sachs says artificial intelligence offers 0:19:55.680 --> 0:19:59.280 the biggest potential long term support for US profit margins. 0:19:59.320 --> 0:20:02.960 AI can boost net margins by nearly four hundred basis 0:20:02.960 --> 0:20:05.800 points over a decade, but the Goldman team notes that 0:20:05.920 --> 0:20:09.400 predicting AI's impact is tricky due to the large number 0:20:09.440 --> 0:20:12.399 of unknown factors surrounding it, such as as we just 0:20:12.440 --> 0:20:16.199 discussed regulation. There have been about sixteen hundred mentions of 0:20:16.240 --> 0:20:20.000 AI by US and European firms in the first quarter 0:20:20.040 --> 0:20:24.800 earning's conference calls alone, of course, a record number now 0:20:24.800 --> 0:20:28.120 coming up. How Kim Kardashian is using her social reach 0:20:28.480 --> 0:20:33.000 to attract investors for a private equity fund. Keeping our 0:20:33.040 --> 0:20:35.600 eye also on shares of Cisco. We get their earnings 0:20:35.640 --> 0:20:38.000 after the bell. Another name to what's we're hired by 0:20:38.000 --> 0:20:41.600 around seven tenths of one percent. Again, would we be 0:20:41.720 --> 0:20:45.080 surprised if AI is a key term when it comes 0:20:45.119 --> 0:20:47.440 to them, but more I guess on the networking side. 0:20:47.560 --> 0:20:48.320 This is Bloomberg. 0:20:57.800 --> 0:21:01.080 Welcome back to Bloomberg Technology, imed Love in San Francisco. 0:21:01.119 --> 0:21:03.440 It's got a quick check in on the markets as 0:21:03.440 --> 0:21:07.520 you want to pay pretty close attention to bitcoin. We're 0:21:07.520 --> 0:21:11.360 down now below twenty seven thousand US dollars per token, 0:21:11.480 --> 0:21:13.600 off six tens of a percent in a session. 0:21:13.640 --> 0:21:15.040 Remember we're trading twenty four seven. 0:21:15.040 --> 0:21:18.040 When it comes to bitcoin, there is some technology news 0:21:18.080 --> 0:21:22.200 out there in terms of engagement between platforms. 0:21:21.800 --> 0:21:23.040 No real news driver. 0:21:23.240 --> 0:21:25.800 There isn't really a close correlation between the trading we 0:21:25.880 --> 0:21:28.720