Will AI Replace Scientists?

Please join my mailing list here 👉 https://briankeating.com/yt to win a meteorite 💥 In this video, we discuss the rapidly evolving intersection of artificial intelligence and fundamental science and how AI is transforming the way physicists handle massive streams of data from modern observatories—while also explaining where its power ends and human creativity begins. Tune in for insights, surprising stories, and a behind-the-scenes look at how scientists are navigating the AI revolution. 🚀 - Key Takeaways: 00:00 - Cosmologist Exploring Universe and AI 05:30 - AI Success and Physics Limits 09:48 - How Did the Universe Begin? 14:00 - Limits of AI in Scientific Insight 16:04 - Cosmology's Data Revolution 19:29 - AI Insights: Science and Writing 23:04 - Limits of Predictive Models 25:32 - AI's Academic Impact: Pros & Cons 33:24 - LLM Chips and AI Accuracy 34:41 - Accountability, Writing, and Discovery - Join this channel to get access to perks like monthly Office Hours: https://www.youtube.com/channel/UCmXH_moPhfkqCk6S3b9RWuw/join 📚 Get my books: Think Like a Nobel Prize Winner, with productivity tips from 9 Nobel Prize winners: https://a.co/d/03ezQFu Focus Like a Nobel Prize Winner, with life-changing interviews with 9 Nobel Prizewinners: https://a.co/d/hi50U9U My tell-all cosmic memoir Losing the Nobel Prize: http://amzn.to/2sa5UpA The first-ever audiobook from Galileo: Dialogue Concerning the Two Chief World Systems: Ptolemaic and Copernican https://a.co/d/iZPi9Un Follow me to ask questions of my guests: 🏄‍♂️ Twitter: https://twitter.com/DrBrianKeating 🔔 Subscribe https://www.youtube.com/DrBrianKeating?sub_confirmation=1 📝 Join my mailing list; just click here http://briankeating.com/list ✍️ Detailed Blog posts here: https://briankeating.com/blog 🎙️ Listen on audio-only platforms: https://briankeating.com/podcast #universe #podcast #briankeating #intotheimpossible #science #astronomy #cosmology #cosmicmicrowavebackground #artificialintelligence #aiinscience #physics #machinelearning #futureofscience #aiandscience #scienceexplained #aiuniverse Learn more about your ad choices. Visit megaphone.fm/adchoices

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:01Speaker 00

  This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome, that's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration blog, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses set up required. Compatibility and availability varies 18 plus.

• 0:26Speaker 01

  This episode is brought to you by Indeed. You're ready to level up your business, but the pressure to find the right person to make that happen is intense. Indeed sponsored jobs can help carry the load. It boosts your job post to help it reach quality talent fast. Cut through the chaos with Indeed sponsored jobs. Get a $75 sponsored job credit at indeed.com/podcast. Terms and conditions apply. Hiring now? Then this is a job for Indeed sponsored jobs.

• 0:56Speaker 02

  All right, Brian Keating, thanks so much for joining us today.

• 0:58Brian Keating?

  That's great to meet you finally, Jake. You're quite persistent.

• 1:02Speaker 02

  Oh, yeah. How many? It's been two years, I think.

• 1:05Brian Keating?

  It's two years since I gave you an assignment and you followed through. So I can't, I can't, you're better than a lot of my students. So.

• 1:13Speaker 02

  There we go. Well, thanks a lot. So for those who haven't seen you on Joe Rogan or seen your other podcasts or read your books, can you give an introduction to who you are? And what do you want people to know about you?

• 1:26Brian Keating?

  Yeah, so, you know, in terms of my different hemispheres in my mind, you know, I break it into work and non-work. And most of what dominates my work is research into the origin of the universe, the big bang. Lately, I'm getting a lot into AI systems and what they can do for fundamental science and experimental science. So I'm an experimental cosmologist, which doesn't mean I do hair and makeup. It means I do. Experiments, experiments that look and search and hope to find evidence for the origin of the universe, the existence of black holes, the possible existence of other planets and life, maybe on those planets all within the purview of what a cosmologist can do.

• 2:09Brian Keating?

  And outside of that, I, yes, I have my own podcast, YouTube channel, which I use to interview, you know, so far about 22, 23 Nobel Prize winners and top thinkers and, and some people that have left this earth, which is the most treasured interviews I have, the three or four of my guests, not because of my interview, Jake, but because of times, you know, scythe taking them at the ripe old age of 99 in one case. So I'm not too, I'm not too distraught over it, but I really cherish the conversations I have because most of my work is not, you know, stroking my, you know, graying beard and looking up at the cosmos, although I

• 2:50Brian Keating?

  do get several moments to do that a week, but most of it's, you know, dealing with people I have to talk to, things I have to do. And the podcast gives me an outlet for things that I want to do. And some of those things involve writing books. I've written four books, working on my fifth one now. And the podcast is sort of a labor of love where I get to talk to people just like you, you know, intelligent people that are curious about the state of the universe. And my most controversial opinion is that scientists, oh, the public, you know, the explanation for what we do and the failure to do so will result in the termination of the job that we would do for free, which is to explore the universe. So I try to give back in that way. And I give my colleagues crap for not

• 3:34Brian Keating?

  doing it and they don't do it. And I salute them when they do do it. So those are kind of my work hemisphere things. And then outside of that, I'm a father, husband, do my best to do that. And I also am in very involved in my synagogue, my temple here, do a lot of stuff involved with my Judaism. And I also very much enjoy. Being, you know, kind of involved in the local San Diego sport scene and doing all sorts of fun stuff like that. So I do a bunch of a lot of things, probably not, you know, not the best at any one of those, but I think putting them all together, it's a pretty fun combination.

• 4:10Speaker 02

  Amazing. And focusing focusing on your work in science, what do you think about the developments in AI over the last three, six, 12 months? Is that going to materially influence the the trajectory of, let's say, your work or others in your field? And, you know, one example I thought of is a few days ago, researchers applied a machine learning technique to uncover unexpected features of the non reciprocal forces that shape the behavior of the many body system. I don't know if you saw you saw that, or that was that was a headline. So they're saying that AI is already influencing the the speed of development in some cases of physics discoveries. Is that aligned with yours?

• 4:47Brian Keating?

  Yeah, well, it's certainly true that AI is have a tremendous impact and has tremendous potential. The question is, what kind of AI and what will it eventually, you know, be able to do? And from my perspective, the breakthroughs are sort of incremental. There have been, you know, just massive headlines. Everything's, you know, coming for every job under the sun. And I just feel like that is completely off base and it's completely counter historical. If you look back at history, the great breakthroughs that we had were not precipitated by, you know, the computer or the invent. Now, there have been many advances because of the computers. But if you look at the classical discoveries in the pinnacle of science, which I am not a member of, but, you know, theoretical physics, for example,

• 5:32Brian Keating?

  or mathematics, etc. I'm an experimental physicist. I'm not a theorist. So I can speak to their credit of what they do. None of it involves like, oh, the calculator and the slide rule, the transistor and the microprocessor. The those are extremely powerful augmenting force multipliers. And I use them every day. And I'm using them right now, as a matter of fact, I'm doing some AI, you know, kind of work, deep research on a new theory that I'm trying to test using an optical telescope. But but besides that, the question of whether or not the style of of deep learning, which is the predominant form of AI that we're familiar with in the LLM space. That these are, you know, kind of these saviors that are gonna come, you know,

• 6:16Brian Keating?

  basically to the rescue of physics and break it out of the rut that it's supposed to be been in for 50 years since the string theory revolution. I think that's all nonsensical. And the reason is is that they're too damn good. Um, you know, if I could go back to save physics, you know, I would, I would, uh, you know, cause Geoffrey Hinton and, and, uh, and John Hopfield, I'd, I'd put them in, you know, in a, in a, in a, you know, in a timeout for, for many years because I think that they're, you know, the neural network and the L M architecture is so successful, it's just gobbled up everything. It's kind of like, you know, the internal combustion engine is so successful and gobbled up a much better, you know, superior alternative, the

• 6:57Brian Keating?

  electric motor. I mean, imagine if you have the opportunity to start over again and, and have, you know, start with the electric motor, you know, electric vehicles, and then you had to convince somebody to get a gas powered car, which you, you know. So I think that the A I. Systems that we are familiar with, L L Ms plus G P U's are not gonna lead to at least new discoveries in my field in physics, but they may be, you know, extremely powerful for coding and for, you know, generating transcripts and doing all sorts of great stuff. The, the issue is that, you know, physics is not a language. It's not, certainly not a language that is informed by a corpus of knowledge that's reinforcement trained on,

• 7:39Brian Keating?

  you know, the past history of humanity on the internet. Uh, that's not how Einstein, you know, he didn't scour all of civilization, you know, from ancient Macedonia to modern day, you know, Bulgarian poetry, and then all of a sudden come up with the theory of relativity. So I think that that's far fetched, and that's not the way physics works. So it's gonna have tremendous impact on all forms of science in society, but it's not necessarily going to have the breakthroughs that I'm most interested in. And I've was pleased last week to see Demis Hassabis, who won the Nobel Prize in Chemistry in 24, 2024. You know, proposing a similar test that I proposed in 2023 and have been working on for quite some time, which is to see if an LLM system can reproduce

• 8:23Brian Keating?

  the physics that Einstein did, just given the knowledge of science that Einstein had in 1911. So if you cut off all knowledge and you don't allow it to peek under the curtain of what happened in the future, can these tools replicate what Einstein did? If they can't, we don't have AGI. We don't have artificial general intelligence because it can't do what a human can do. We certainly don't have artificial superintelligence. So I am optimistic. It gives me the power of a thousand PhDs and postdocs working around the clock, literally. And I'll give it tasks before I go to bed and I'm really into it. You know, I set up an open claw server, eight different times because it didn't really work that well the first seven. But I finally got that dialed in. I've got subscriptions to every major tool that there is.

• 9:07Brian Keating?

  I use it all day, every day. But it's really to augment what I do. The ultimate test will be, can it build the conducting supercollider? Can it build a large hadron collider? Can it build the James Webb Space Telescope with no human input? Because that's what humans did. That's the minimum bar that they have to be able to show in order to prove that I should be worried about them taking my job or my colleague's job. So they may take over parts of it and things like education. We could talk about that. I think that's right for disruption, as they say. But I'm not worried about it in terms of taking my job just yet.

• 9:46Speaker 02

  Right. And along those lines, like as someone who understands the limits of physical measurement, do you think there's a hard ceiling on what silicon-based intelligence can understand about the universe? Things that require embodied intuition?

• 9:57Brian Keating?

  There's no hard ceiling. I think anything that's a universal computer is capable of doing everything that we can do. I mean, we may have a ceiling. It may be that we have a ceiling and so everything has a ceiling. As far as I understand, any kind of computer, even a TRS-80 from when I was seven years old, could theoretically have no limit as long as it can do enough computations. Now, there are hard limits in terms of the age of the universe, the lifetime of black holes and Hawking radiation. You know, the practical so-called limits are trillions of year limits. But in principle, no, there's nothing that prevents them from discovering everything that's discoverable as far as we know.

• 10:37Speaker 02

  And then what's the most important question in cosmology that you think AI could could help us figure out in the next five, 10 years?

• 10:43Brian Keating?

  I think we're, yeah, so the most important question I'm convinced of because of perhaps my bias and that's what I work on for the last 20 years is, is how did the universe begin? I think that's the ultimate question. It's sort of like asking, you know, a biologist, what's the ultimate question in biology? They may be partial to their own field, but how did life begin? You know, not just on earth, but in the universe, because they may be interrelated and I'm curious about that, too. So I think the ultimate question cosmology is how does the universe begin? How do you go from a non-universe to a universe? How do you go from non-matter to matter? How do you go from matter in simplest possible form, hydrogen, helium, protons, neutrons, electrons?

• 11:28Brian Keating?

  How do you go from that to organic chemistry? How do you go from organic chemistry to biological, simple molecules? How do you go from molecules to, you know, large life forms, cells? How do you go from, you know, prokaryotic cells to eukaryotic cells? How do you go from those to mammals to, you know, all these different contingent things? But they all start with the big bang as the, you know, the big bang theory television show, theme song, highlights. Everything has to start with the big bang. So in cosmology, that's the most important thing that I feel I can study and that's what I've dedicated the experiments and the careers of the scientists that work, you know, in my lab, that's what we're working on.

• 12:06Speaker 02

  And what about the AI hype, you know, frustrates you the most as a scientist? Is it what you just said about people thinking it's going to take over everything or?

• 12:15Brian Keating?

  I think it's, you know, that they don't, they don't really have any humility about it. Like, there's more to being, you know, scientist or even a citizen or anything than just raw intelligence. I mean, some of the most important intelligent people in the world are in the Epstein files. My name's in the Epstein files, but I'm not, I'm not, you know, fortunately is because I attended a conference, not because that he supported, he wasn't even there. And you know, some of the most brilliant Nobel Prize winners are in this are in these Epstein files. They're far more intelligent than me, and yet they have no wisdom, or they have very little wisdom, or they had, you know, horrible, horrible, you know, sociopathic behaviors. Some of each category are probably true. And so, no, raw intelligence should never be confused

• 13:00Brian Keating?

  with with anything approaching wisdom. I think what we could use is artificial wisdom, you know, more than artificial intelligence. I mean, we've had tremendous intelligence for decades, you know, since the internet came around. Wikipedia is far smarter than any human being, but in terms of raw knowledge and even speed and processing, I mean, we've had supercomputers and superintelligence, if you narrowly define it as ability to do math problems really quickly, we have that for, you know, 55 years or 60 years now. But that's not what most people care about. I mean, I care more about, you know, the person that I'm interacting with, their wisdom, and, you know, someone taking care of your your kids, you don't really care how smart they are, if they're evil, malicious, malevolent people, you know, just because Stephen Hawking is so brilliant, you know, he went to Epstein

• 13:44Brian Keating?

  Island multiple times, you know, so what are the actual implications of intelligence? I think they're way overstated.

• 13:53Speaker 00

  This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome, that's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses setup required. Compatibility and availability varies 18 plus.

• 14:19Speaker 02

  Got it. And it's more, and we need to balance that with like wisdom morality.

• 14:22Brian Keating?

  Yeah, I think that we don't have any ethical training. I mean, a scientist, we get none. I mean, it's funny, we make fun of our colleagues in in the business school, you know, all they just care about money, or the law school, and they're, you know, they just want to be around the world or whatever. And then the medical students, you know, that that come to our classes, we always kind of say, oh, they just care about getting a good grade on the MCATs. But all those three disciplines, business, law, and medicine, have ethical training requirements for their students. We don't have any of that. You know, we just go, oh, you'll you'll pick it up by osmosis. And I'm like, oh, yeah, you're going to pick up quantum mechanics by osmosis? Oh, no, that's really serious and important. We're not going to pick that up by osmosis. So why don't we, why don't we teach ethics the way we teach quantum mechanics? Oh, well, we don't have time, or no, you do have time,

• 15:06Brian Keating?

  you just don't want to prioritize it.

• 15:09Speaker 02

  Huh, that's interesting. And so another question back to the thing you brought up earlier about, it sounds like you don't believe that AI is going to come up with any novel scientific theories, right? Like it can help us.

• 15:22Brian Keating?

  I wouldn't say I don't think they're going to come up with any novel. I think they're going to come up with novel things. I just think the biggest, most important things that I care about, you know, understanding new theories of unification of forces and fields of matter and energy, of the ultimate, you know, model that describes how the universe began. I think that they're, you know, as incapable of doing that as, you know, a computer in 1965, when computers were brand new, you know, they weren't able to just necessarily predict the existence of dark energy. I mean, that took human insight to kind of conjecture, and then later it took observation. So I don't think it's possible as a scientist to conjecture things and then prove them without any human contact with some way of acquiring data.

• 16:07Brian Keating?

  I mean, you could say that the humans might, you know, will work for the computers. And there is a rent to human function that you can hire. These Claude bots have been hiring humans to do things like CAPTCHAs and all sorts of things. So that's not really what I mean. I mean, someone has an idea for something or discovers something. I mean, half of the most interesting things are discovered serendipitously. The CMB, the Cosmic Microwave Background that I study, was discovered serendipitously. And I always say it's very hard to plan on serendipitous discoveries because by definition, you don't know that one, they're going to happen. So how does an AI, you know, really enhance that? In one sense, they can't. In another sense, you can't, you know, ignore how powerful they are

• 16:50Brian Keating?

  and how supplemental they are. And they should be used as a compliment, not a supplement for what physicists do.

• 16:56Speaker 02

  So with the CMB, so you call it the Simmons Observatory in Chile. So how could an AI?

• 17:02Speaker 02

  What did I say? So how could, how could a, so you think an AI-driven approach like the one used in the dusty plasma study be applied to CMB data to find physics that we're currently missing?

• 17:12Brian Keating?

  Um, it could be.

• 17:14Speaker 02

  And, and just to elaborate a little more, like, you know, CMB is the ultimate many body data set, right? Billions of interacting photons from the early universe. So is anyone really applying physics-informed neural networks to it?

• 17:27Brian Keating?

  Yeah, for sure. I mean, right now, the, where we're at with physics and, and a lot of, you know, understanding of, of high-energy particle interactions and cosmological interactions is we're drowning in data. We have tons and tons of data, which wasn't the case 50 years ago. There's no data. There was like 10 galaxies that we knew about, you know, and then we had their redshift and we saw they're higher than the redshift seems to be that they're moving away from us. You know, so cosmology was really data-starved for the first 30, 40 years after 1929 and Hubble and Einstein and Le Metra, kind of conjecture that the universe began in a hot, dense state. And then it wasn't until the discovery of the cosmic microwave background, this three Kelvin, all pervasive radio wavelength, microwave

• 18:10Brian Keating?

  wavelength radiation that bathes us in a constant three-degree Kelvin glow, that we had a precision number. Finally, we had a physical cosmological observable, a temperature which we could convert to a density, which we could convert to, you know, a basically an ionization fraction of the universe, that we had a, we had a, you know, quantitative science rather than a qualitative science. And since then, it's just exploded. I mean, we've got new telescopes. The Rubin Telescope came online last year. We're gonna have, you know, the, the upcoming Nancy Grace Roman Telescope is launching soon. We have, you know, the James Webb Telescope, obviously. Hubble Space Telescope, the Simons Observatories came online two years ago, is gonna release data in the next year or so. It's just, you

• 18:54Brian Keating?

  know, it's a golden age for data, but we're drowning in it. And, and each day of data, we get about a terabyte worth of, of acquisition. Now, we don't keep all that data, but imagine like your hard drive fills up every day. And, and a lot of it, and when the weather's good and the, and the seeing is good and the conditions are right, you want to keep 500 gigabytes of that one terabyte, and it's every day for 10 years. So you're talking about petabytes of data. And my colleagues at the LHC will laugh at me. That's like, they get a petabyte in a couple hours, but they throw away 99.999. You know, they have six events over 12 years. That's a big deal for them. So we're in different regimes. So AI is perfect for that. AI, machine learning,

• 19:38Brian Keating?

  just incredible data sets. The math is not that complicated. I mean, LLM math is not complicated. That's basically what's known as linear algebra, matrix multiplications, diagonalization, gradient descent method. It's very powerful that they were optimized, but they came out of serendipitous. They were serendipitously discovered, too. They came out of the computer gaming field where GPUs were needed to blast pixels at a high rate to a monitor. And so there's nothing like that that is, has to do with how a new field theory might come about. But on the other hand, it's very good at image analysis, which is one of the things that we do, is one of the preliminary steps that we do on many different branches of astronomy and cosmology. So no, they're going to be very useful. And they may come up with things. Like,

• 20:23Brian Keating?

  I had to do things, you know, where I felt like, well, this is kind of magical, where it kind of surfaced something. It was actually from my books and, and podcast. Like, I fed it every transcript I've ever done. You know, it's a huge, 1,700 total videos, audio, you know, fed it all in. I don't know if it, you know, what it does to it, to be honest with you, how it compresses all that information. But it was a 40 megabyte text file, plus my books, you know, four books, plus every appearance. I've been on other people's podcasts, like, this will go in it to the next version, right? So I fed it all in. And then it surfaced like really interesting connections. And I, you know, I remembered writing it, but you know, I didn't. The good thing about having written books before ChatGPT 3 came out is that, you

• 21:06Brian Keating?

  know, no one can claim I wrote it with the help of AI. So I'm kind of proud of like that it resurfaced. And it really picked out some things in my childhood and, and the connections between, you know, say, science and religion that I didn't think about so much when I was writing my first book in 2018. To now when I'm writing my fifth book, you know, and it was just kind of magical moment where it surfaced this thing and I could imagine it doing the same where I'm looking for this very peculiar type of polarization signal from the early universe and also looking at how like a crystal behaves on earth and saying, oh, I have data from that in the lab. There's only one universe. There's many crystals. Maybe it's telling me to put these things together, not like magical woo woo crystals, but about study of the behavior

• 21:48Brian Keating?

  of optical polarization in the laboratory as a tool to understand the polarization in the universe. I felt that was really clever. And so there are magical things about it for sure. And it's just not clear to me that the main limitation, as you know, for a lot of these LLMs, the training data and it's just vast and that 40 megabytes of data took me 10 years, you know, more to assemble, right? So it's, it's not like it's, it's easy to get the data. And so what do the companies do now? They do two things. They either generate data themselves after they, they stole a lot of data, including my books. You know, Anthropic has admitted that they, you know, basically scanned a half a million books and then burned them for some reason, which is very strange, but they scanned all

• 22:31Brian Keating?

  this. So they stole basically intellectual property from people. But then other company, you know, then they're using people to do reinforcement learning using humans. And then they're also doing training data. And then the Chinese models are basically sampling the American models or the French models or whatever. And they're doing what's called distillation and they're basically, you know, reverse engineering what the, what the weights are. And so if the weights are so dependent on data, then that data by definition was recovered in the past. So it's only past looking. So it can't by definition predict the future. So it would be like looking at 1911 data for Einstein and just saying, no, there's no, there's no weird behavior over the planet Mercury. And then we'd be

• 23:13Brian Keating?

  stuck in that, in that prison. So I call it like the GPU LLM prison. I'm not sure. Yeah, I could be wrong. We may discover that a new law of physics tomorrow. And people are doing great things with it and especially in math and checking stuff and proofs. But it's not, as far as I know, it's not able to pass this test that I coined the, you know, the Keating test, but Hassebus coined the Einstein test. And I think that's a better name, honestly.

• 23:40Speaker 02

  Well, then I want to get your quick opinion on this article that came out a week ago. I don't know if you had seen this, but AI reveals unexpected new physics in the force of matter. Not sure if you'd come across this, but basically, what this had said is that the physicists at Emory had used a custom neural network trained on 3D particle tracking from dusty plasma experiments to infer the force between particles. And so they figured out that two longstanding theoretical assumptions were wrong, that charge doesn't scale proportionally to particle radius, and also that force drop off between particles does depend on particle size with 99% accuracy. So, I mean, this came out a week ago. You know, this stuff's not coming out every day, but...

• 24:19Brian Keating?

  I guess the analog for me would be, does it predict what's called the Coulomb force, which is the ultimate electrostatic force? Can it predict Maxwell's laws? Like, if you put an LLM and, you know, just, again, train it on data from 1860 and before, and nothing else, and it's impossible to do that now because the data's so corrupted, and even in physics data sets, like, you just have a physics paper corpus of training data, it'll still have like, you know, it'll know somehow about like, the Fast and the Furious 3, or Interstellar, and it's very hard to decontaminate it. But I would say, no, things like that don't surprise. It would surprise me, again, because that would be effectively the Einstein test. Like, if it predicted a new force of nature, which,

• 25:04Brian Keating?

  you know, some people are claiming, and you always have to be a little bit careful. It's like, scientists discover life in another exoplanet, and you have to look really deeply, and what does it mean? Oh, they've discovered a molecule which does something tricky, and that's unusually, you know, usually associated with life. You know, those are all potential biased results. So, that's, it sounds like a really cool result, a tour de force mathematically. I don't know if it qualifies for the Hassebus-Kitting-Einstein test. I'm going to throw my name in all those Nobel laureates. But it's certainly, again... It's not really clear what I could do with a thousand, you know, 190 IQ PhDs working for me like, but, but you know, you need some kind of orchestra. I even find this, I don't know if you've used this open claw.

• 25:49Brian Keating?

  Yeah, but, but I find when

• 25:51Speaker 02

  I have it, I have it running on this Raspberry Pi right here.

• 25:53Brian Keating?

  Oh, you do? Okay. You're gonna have to tell me how to set it. I got a Raspberry Pi, but I have it running on a Mac Mini.

• 25:59Speaker 02

  You just gotta ask Claude.

• 26:01Brian Keating?

  Yeah, yeah. I'll, I'll get your, uh, get your tips for that offline. So I was setting it up and then like I was blasting through all my tokens for Claude 4.6 and I was like, this is crazy when like, you know, a lot of it was just like, send me my daily weather briefing, plus, you know, local papers on the physics archive and, and, um, you know, what's, what's my first meeting? And it was all tech. It did it. And then I was like, you've reached your session limit on, I'm like, what the hell? And so I figured out how to change the orchestrator from 4.6, use the orchestrator only in 4.6. And then, uh, use other models, Gemini Flash and whatnot, right? I don't want to get too nerdy, but, but you're, you're familiar with it.

• 26:39Brian Keating?

  And I realized that's kind of like what we have to do as scientists. We have to take like, we have to be the orchestra. We have to be the Claude 4.6 or, you know, whatever. And we have to, you know, have our students and colleagues and you'd be surprised. 90% of my colleagues don't use AI at all. 10%, 9% of the remainder are hostile to it for good reason. There's, I mean, I, I wouldn't say there's good reason, but it's causing sort of more problems for academicians like me and professors like me than it seems to solve if you're, if you're an experimental physicist or even a theoretical physicist, like I said, and you're just like a pencil and paper pusher guy or gal, you don't really have much use for it. But, um,

• 27:19Speaker 00

  This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome, that's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses setup required. Compatibility and availability varies. 18 plus.

• 27:44Brian Keating?

  But they're, you know, we're finding kids are using it to do every single assignment. We found kids are going to the bathroom during final exams and coming back 10 minutes later and getting, you know, identical answers to different final exam problems. And so we're seeing, you know, kind of like only the negative side. So like I said, 90% don't use it, 10%, 9% use it, but are, you know, kind of vehemently opposed. They're trying to use it to figure out ways to catch people using LLMs. And this is AI versus AI, spy versus spy. And then there's the 1%, you know, maybe like me that are tinkering, you know, get up early in the morning, play with it, have our kids use it.

• 28:26Brian Keating?

  Your kids are young, but, but, you know, when your kids get older, you're going to want them to use it. I have a son who's wanting to learn, you know, differential equations in high school and, and I think he can do it. I couldn't do it. I didn't learn, you know, calculus A B until my senior year, you know, and I had to self teach myself. But I said, look, you can have this book. You can take like Steven Weinberg's book on gravity and say, explain it to a smart high school student. Um, and then break down the calculus into chunks that I've already covered in calculus A B or whatever. And it will do that. And so you can make, and I said, like, tell it not to make any spoilers, like there are spoilers in calculus. Like you don't think about it like that, but calculus is, is it like a love story?

• 29:07Brian Keating?

  It's a beautiful mankind's, one of mankind's highest accomplishments up there with, with Beethoven and Mozart and Picasso. So I don't want him to get spoiled of that benefit. You know, he's reading The Great Gatsby. I'm like, you know, have a reading guide, but don't, but always tell it not to spoil it because it will spoil it. But it will also hallucinate. And that's the other problem with it. A lot of my colleagues don't trust it, again, for good reason, because it's very sycophantic. Oh, that's a great idea. I get, I get literally, I get, you know, probably three to four emails a day. Professor Keating, I just got one. I could read it to you. It's, I'm not even going to respond to the person, very earnest people, but, but, you know, they, they, they write to me. They're desperate. They're, they're wonderful people, I'm sure.

• 29:50Brian Keating?

  And, and I would love to be able to help every one of them, but I can't. I have, you know, kids and a wife and a family and all sorts of other things. But here's one, you know, that I just got, a falsification of the Torah. And and then how the how what is it? The many body problem comes into play, the refractive index that takes place. It causes a function that then decays geologically. And that explains the Torah and the seven days of creation. And and that's that's one example. That's the least maybe technical one I've gotten. And then I get people that, you know, like, oh, I just want five minutes of your time. And I'm like, well, you know, my five minutes of my time could be worth like a million dollar. Like, if it's on my deathbed, you know, that's that's worth a lot, right?

• 30:35Brian Keating?

  So so I started charging people. I said, because the L L M's are telling them, every single this guy's just said, here's a script that I got from Gemini 3.0. And you can prove that it works. Just run it on your computer. You have to download this model first, and then you download the other. Like, because it won't work. I'm like, OK, you're telling me this whole homework assignment. You're telling me it only take five minutes of my time. But I'm going to tell you, my time is worth millions of dollars perhaps per minute. So depending on where I am in my life. So I have a standard thing now. If you want an hour of my time, you can either pay a thousand dollars an hour. And that goes to charity. UCSD has a food bank, if you can believe it. There are starving or, you know, basically food insecure kids on campus, which is a tragedy.

• 31:18Brian Keating?

  I mean, we we charge so much for tuition that there are kids that can't afford to eat, Jake. So it's sick. I I feel like universities are gotten totally out of control. But that'll have to be a topic for our next podcast when you come down here. But the point is that I want to give to these kids. So I donate my own money every year, a decent amount of money. And then anytime that someone wants an hour of my time, I say, you can have an hour, but it's going to cost you a thousand bucks and you have to donate to charity. And I'm also going to start doing that with podcasts pretty soon. Because, again, it's just like, I have so much cool stuff I'm excited about. I love talking to people like you, but, you know, I have to prioritize. So you got off cheap this time, but maybe next time it'll be a little bit more expensive. Just kidding. But, you know, the thing is, it has to go to charity. And then they have an hour with me.

• 32:02Brian Keating?

  If they don't want to do that, I charge $20 a month for a group coaching session. So I've got like seven to 10 people every month. Yeah, I have 20 or 30 that are subscribed to it, but they don't they don't come. Seven or 10 people have an hour a month with me, and we just talk about their ideas. I love it. But The problem with these AIs is that they're giving people this kind of this this sort of hallucinatory, you know, sycophancy that there are the next Einstein and that they just need me to sign off with a .edu email address and then they can get the paper published. I mean, literally, I get people asking me to endorse their papers, to put them on the, uh, submit them for publication. I'm not going to do that. I won't do that for anyone.

• 32:42Speaker 02

  I'm just trying to connect. I'm trying to connect and make sure I followed. So the, what you're saying about people emailing you, the, the story there is that they're emailing you to get your sign off on what they think is the discovery.

• 32:53Brian Keating?

  After getting chat GPT sign off on it. Yeah, it's always there. Chat GPT is allowed the proliferation.

• 33:00Speaker 02

  But they're, but they're saying, hey, I discovered this great amazing new thing.

• 33:02Brian Keating?

  New law physics. Yeah, I had someone call me in.

• 33:04Speaker 02

  Damn, you're, that's the content you're getting from everyone.

• 33:07Brian Keating?

  Someone, another guy called me in a hole because I wouldn't sign off on his theory or I wanted to.

• 33:12Speaker 02

  What does it mean sign off? What, what does it mean sign off?

• 33:14Brian Keating?

  So to get a paper published, you have to submit it to a journal, right? And so the journal gets peer reviewed. So they have, um, typically you have to submit that from a university, a .edu account, or sometimes a government account. It's very rare you can submit with a Gmail account and there's another venue by which people publish their papers called the archive. It's actually a public domain service going back 30 years now. It's called archive with an X. Uh, and it is, uh, published that, you know, Cornell is supported by the Simons Foundation as well. And that, uh, that also typically will, uh, very highly de-weight people that don't have a .edu email address.

• 33:55Brian Keating?

  So you need something called an endorsement before you can submit of any kind. If I submit and it was my first paper, but I'm a .edu, like my graduate student, when she wants to submit a paper, she has to get an endorsement from someone else, usually like me. And typically they have, they have to have published on the archive or they have to have a .edu email address to just get above the referees that are there. And that's not refereed. It's just, it's kind of like vetting process. So that's what I mean. But what's happening is people are saying, I don't want to waste your time. So I had chat GPT do this. Look at this chat that I paste to a thousand. Lines of text in the email, which literally happens. And here's it going through step by step, every single equation and verifying it's like, yes, you know, Jake, you're absolutely correct. This is incredibly,

• 34:39Brian Keating?

  you know, transformative result that and and I can make up like a thousand different theories right now and have chat. I could guarantee you, we could go to chat GPT right now. I could come up with some crystal dust plasma theory that explains the CMB. And there's no big bang and chat GPT will endorse, well, effectively say it's correct. So that's the problem in that you can make up to complete and you could ask it facts. I said to it recently, Brian Keating, what books has he written? Losing the Nobel Prize, correct. Into the Impossible, correct. A Brief History of Time, not correct. I had it say to me that I I had this new one that's a chat function on a built in supercomputer chips called Talus.

• 35:21Brian Keating?

  And they make these chips now that have LM architecture encoded in the chip. And I asked it, who's Brian Keating is like Jimmy chat or something dot AI. And it came back, Brian Keating is the cosmologist at San Diego. And he was the former head of the Large Hadron Collider. And so it's like, you can't trust it on facts, but you're gonna, I mean, a couple of years ago. If you asked it, how many Rs are in strawberry, it got it wrong. So all the more so for checking a theory of physics that we don't even know exists, but has never been peer reviewed, identified or vetted.

• 35:54Speaker 02

  Wow, I want to take a 90 degree turn here and go back to something you said minutes ago, which was that you fed AI all of your podcast content, all that stuff. I wanted to ask you, what were the questions you asked it after you fed it all that?

• 36:08Brian Keating?

  So I just asked, based on the, you know, kind of text that you've seen, you've received. What are my top five priorities? What am I? What am I working on? What holes am I missing? What connections can you see between these ideas and the different podcasts? I also fed it.

• 36:26Speaker 02

  The craziest thing you what did it? Did it tell you anything you didn't know?

• 36:30Brian Keating?

  It gave it was kind of like my personal trainer, you know, like my personal trainer will say, you're like, eat less, you know, carbs and workout more. And like, I know that, but it was good. It's good to have accountability, right? To get weighed, to get measured, to get, you know, checked and surfing and lifting weights and all the stuff I like to do. So. It didn't tell me anything I didn't know, but it told me stuff that I was unwilling to see maybe, or that I didn't want to see, including aspects of my upcoming book, which, you know, I'm not going to talk too much about because sometimes you talk about stuff and it takes away a little bit of the thrill of doing the actual thing. You gotta just do the thing and not think about doing the thing or talk about doing the thing. So for me, yeah, it had this very interesting connection, you know,

• 37:10Brian Keating?

  between this, you know, person in my life that, you know, I was considering writing about, but not sure how to orient it, you know, in the book, you know, their private person. And so how do I do that and do it? And I came up with a very, a very warm way to do it. And I still have to write it, you know, I have to sit down and write it. But the architecture, the skeleton of it, you know, is like, the main challenge I'm finding is that to describe this particular physical effect that we're looking for, it is as a ancillary kind of finding and discovery for the Simons Observatory potentially. It involves really very abstract math and physics, like way beyond what's in A Brief History of Time

• 37:54Brian Keating?

  or, you know, A Fabric of the Cosmos or, you know, Brian Greene's books as great as they are. You know, they kind of gloss over them, like string theory is a billion vibrating dimensions in a super high dimensional space of energy. And it's like, okay, what does that mean? They won't tell you. Or waves are correlated and particles are waves and they have this weird collapse of the wave. You know, like they never really explain what that means or why it's a problem. So I want to do it justice, but the problem is a lot of people that are popular science consumers, they're not going to be willing to go through the trouble of learning the math and physics behind it. So how do I do that elegantly, you know, to quote Brian Greene? How do I do that in an elegant way?

• 38:36Brian Keating?

  But it's also like, I don't want to sugarcoat it and make it like fanciful. Because his books and Stephen Hawking's books and Lisa Randall's books and Janet Levin's book, like they don't have to be, they don't have to ever confront reality. Like they could talk about string theory and say, oh, it's going to be discovered in 2020, you know, 2222, you know, and 200 years from now. I'm an experimentalist, I can't get away with that. You know, I'll lose my reputation, my funding, my livelihood. So from that perspective, an experimentalist has to be a lot more practically oriented. And so those are some of the things that's kind of helping me work through as we speak.

• 39:16Speaker 02

  Amazing. Well, Brian, this was incredible to get a chance to talk to you. It's really on.

• 39:21Brian Keating?

  I hope we can meet in person, either up there or down here, and keep doing what you're doing. Congratulations on all your success. I mean, you've grown a lot in two years. I was, I think you, if I could have your growth, you know, 60x or whatever, since I gave you the homework assignment, I would take.

• 39:36Speaker 02

  Oh, yeah. Well, I, yeah, we'll see. Here, let me stop the recording and, yeah, well.

• 39:44Speaker 04

  Stonefire naan is made with real buttermilk and ghee. It's the perfect flatbread for any occasion.

• 39:49Speaker 05

  I put out stonefire naan with dips at every dinner party.

• 39:53Speaker 00

  I make naan grilled cheese when my daughter has friends over.

• 39:56Speaker 04

  I use stonefire naan to reward the humans who do my bidding.

• 39:59Speaker 05

  Right. It's perfect for camping trips. Dance for it, feeble mortal. Sorry, is he talking to me or? Dance, ha, ha, ha, ha, ha.

• 40:09Speaker 04

  From our tradition to yours, stonefire. Together, we naan.

• 40:15Speaker 06

  When you spend enough time around great companies, you notice that the best ones treat trust as a growth strategy, not a box to check. Vanta helps companies build and prove trust from day one. Its Agentic Trust Platform connects compliance, risk, and trust at enterprise scale. Vanta delivers a 526% return on investment over three years and pays for itself in just three months. Learn why over 16,000 of the fastest-growing companies use Vanta at vanta.com.