Bản chép lời đầy đủ của ReadMultiplex.com:AI In Undiscovered Public Knowledge: A 1986 Chicago Librarian Said The Next Discoveries Were Already In Print.

swarm pulled off in 2026, we have to travel back 00:03:30.840 --> 00:03:33.219 in time and understand the analog foundation 00:03:33.219 --> 00:03:35.580 first. Right, we have to look at the analog roots. 00:03:35.759 --> 00:03:37.639 We have to start with the man who originally 00:03:37.639 --> 00:03:40.840 identified the underlying mathematical reality 00:03:40.840 --> 00:03:43.560 of human ignorance. We have to look at Don R. 00:03:43.599 --> 00:03:47.379 Swanson. Swanson is... A genuinely captivating 00:03:47.379 --> 00:03:49.560 historical figure, primarily because his mind 00:03:49.560 --> 00:03:52.080 just refused to be constrained by a single academic 00:03:52.080 --> 00:03:54.379 discipline. Yeah, his resume is all over the 00:03:54.379 --> 00:03:57.139 place. It really is. He was born in Los Angeles 00:03:57.139 --> 00:04:00.539 in 1924, and his early intellectual trajectory 00:04:00.539 --> 00:04:04.699 was rooted in pure, hard physical sciences. He 00:04:04.699 --> 00:04:06.900 earned his bachelor's degree in physics at Caltech 00:04:06.900 --> 00:04:10.300 in 1945. So right at the end of World War II. 00:04:10.639 --> 00:04:13.039 Exactly. This is right in the thick of the mid 00:04:13.039 --> 00:04:15.330 -century physics boom. You know, the era of the 00:04:15.330 --> 00:04:17.990 Manhattan Project and the birth of modern quantum 00:04:17.990 --> 00:04:20.470 mechanics. An exciting time to be a physicist. 00:04:20.810 --> 00:04:23.189 Absolutely. And he followed that up with a Ph 00:04:23.189 --> 00:04:25.110 .D. in theoretical physics at the University 00:04:25.110 --> 00:04:29.149 of California, Berkeley in 1952. So he is trained 00:04:29.149 --> 00:04:31.649 at the very highest levels to look at the universe 00:04:31.649 --> 00:04:34.629 in terms of fundamental laws, interacting forces 00:04:34.629 --> 00:04:37.259 in complex systems. Right. But rather than staying 00:04:37.259 --> 00:04:39.300 in a university basement calculating particle 00:04:39.300 --> 00:04:41.759 trajectories for the rest of his life, he pivots 00:04:41.759 --> 00:04:44.079 into the absolute bleeding edge of the early 00:04:44.079 --> 00:04:46.819 computer age. He took that systems -level thinking 00:04:46.819 --> 00:04:50.160 into applied technology. He transitioned into 00:04:50.160 --> 00:04:52.620 computer systems work, spending time at Hughes 00:04:52.620 --> 00:04:55.180 Aircraft, and then moving to Rama -Woldridge. 00:04:55.519 --> 00:04:58.939 Which were massive tech hubs back then. Oh, absolutely. 00:04:59.379 --> 00:05:01.519 And there's a critical detail in the source text 00:05:01.519 --> 00:05:04.000 about his time at Rama -Woldridge that foreshadows 00:05:04.000 --> 00:05:06.589 everything we're going to talk about. He managed 00:05:06.589 --> 00:05:09.810 a group specifically focused on machine indexing 00:05:09.810 --> 00:05:14.189 and machine translation. Ah, okay. By 1960, he 00:05:14.189 --> 00:05:16.110 actually published a paper in the

prestigious 00:05:16.110 --> 00:05:19.529 journal Science titled Searching Natural Language 00:05:19.529 --> 00:05:22.930 Text by Computer. I just want to pause and put 00:05:22.930 --> 00:05:25.930 that 1960 date into perspective for you listening. 00:05:26.250 --> 00:05:28.639 Think about what a computer... physically was 00:05:28.639 --> 00:05:31.800 in 1960. Hard to even imagine now. Right. We 00:05:31.800 --> 00:05:34.740 are talking about room -sized mainframes. Data 00:05:34.740 --> 00:05:37.939 was inputted with physical punch cards. The hardware 00:05:37.939 --> 00:05:41.819 ran on hot, fragile vacuum tubes. Massive, clunky 00:05:41.819 --> 00:05:44.540 machines. The vast majority of the global population 00:05:44.540 --> 00:05:47.199 had never even seen a computer, let alone conceived 00:05:47.199 --> 00:05:49.040 of using one to parse through human language. 00:05:49.360 --> 00:05:52.259 And here's Swanson, decades before the personal 00:05:52.259 --> 00:05:54.980 computer revolution. trying to teach primitive 00:05:54.980 --> 00:05:57.759 machines how to sift through the nuances of text. 00:05:57.980 --> 00:06:01.259 His foresight was staggering. He understood way 00:06:01.259 --> 00:06:04.459 before anyone else that the volume of human knowledge 00:06:04.459 --> 00:06:07.259 was about to outpace human memory. He saw the 00:06:07.259 --> 00:06:10.540 data explosion coming. He did. But then his career 00:06:10.540 --> 00:06:13.800 takes a turn that, on paper, must have seemed 00:06:13.800 --> 00:06:16.040 utterly baffling to his fellow physicists. Yeah, 00:06:16.060 --> 00:06:19.149 this is the part that always gets me. In 1963, 00:06:19.389 --> 00:06:21.730 this theoretical physicist and pioneering computer 00:06:21.730 --> 00:06:24.269 scientist accepts a position at the University 00:06:24.269 --> 00:06:26.769 of Chicago's Graduate School of Library Science. 00:06:27.089 --> 00:06:30.269 Wait, library science? Yes. And he didn't just 00:06:30.269 --> 00:06:32.509 teach a few classes. He ultimately served as 00:06:32.509 --> 00:06:34.550 the dean of the school across three separate 00:06:34.550 --> 00:06:37.129 terms. You look at that resume and you have to 00:06:37.129 --> 00:06:39.850 wonder why a theoretical physicist becomes a 00:06:39.850 --> 00:06:42.850 librarian. I mean, the conventional view of a 00:06:42.850 --> 00:06:46.569 library is just a silent depository, right? Right, 00:06:46.569 --> 00:06:48.959 a place where you just keep things. A warehouse 00:06:48.959 --> 00:06:51.660 where books are stored, categorized by the Dewey 00:06:51.660 --> 00:06:53.600 Decimal System, kept safe from the elements. 00:06:53.740 --> 00:06:56.540 It feels like a static environment for someone 00:06:56.540 --> 00:06:59.480 obsessed with dynamic systems. Well, it's only 00:06:59.480 --> 00:07:02.459 static if you look at the physical books. Swanson, 00:07:02.639 --> 00:07:04.579 looking through the lens of physics and early 00:07:04.579 -->

00:07:07.680 computation, viewed libraries fundamentally differently. 00:07:08.000 --> 00:07:12.230 How so? He saw an information ecosystem. Because 00:07:12.230 --> 00:07:14.209 he understood the mechanics of how complex systems 00:07:14.209 --> 00:07:17.029 fail, he was uniquely positioned to spot the 00:07:17.029 --> 00:07:19.910 failure mode of academic publishing. Okay, I 00:07:19.910 --> 00:07:22.209 follow. He wasn't looking at the books themselves. 00:07:22.629 --> 00:07:25.310 He started looking at the empty, dark spaces 00:07:25.310 --> 00:07:27.829 between the books. The source text offers this 00:07:27.829 --> 00:07:30.730 brilliant, visceral analogy to explain what Swanson 00:07:30.730 --> 00:07:33.350 realized, and it perfectly illustrates the structural 00:07:33.350 --> 00:07:35.730 failure you are talking about. The two rooms 00:07:35.730 --> 00:07:40.279 analogy. Yes. Imagine a sprawling research library 00:07:40.279 --> 00:07:43.259 with two specific rooms. In Room A, you have 00:07:43.259 --> 00:07:45.740 a dedicated group of medical researchers. They 00:07:45.740 --> 00:07:48.740 are exhaustively studying a specific dietary 00:07:48.740 --> 00:07:52.259 substance. Let's say it's fish oil. Okay, fish 00:07:52.259 --> 00:07:54.959 oil. They conduct their trials, and they publish 00:07:54.959 --> 00:07:57.600 hundreds of papers, proving beyond a shadow of 00:07:57.600 --> 00:08:00.860 a doubt that this substance does a few very specific 00:08:00.860 --> 00:08:04.810 biochemical things. It thins the blood. It calms 00:08:04.810 --> 00:08:06.769 the blood platelets so they don't clot as easily. 00:08:06.910 --> 00:08:09.370 And it relaxes the walls of the blood vessels. 00:08:09.810 --> 00:08:12.149 So they've thoroughly documented the mechanics 00:08:12.149 --> 00:08:15.209 of fish oil. Exactly. They do phenomenal work. 00:08:15.290 --> 00:08:17.810 They publish in their specific nutrition journals. 00:08:17.910 --> 00:08:21.649 And they stay in room A. Right. Meanwhile, all 00:08:21.649 --> 00:08:23.610 the way on the other side of the building in 00:08:23.610 --> 00:08:25.870 room B, there is a completely different group 00:08:25.870 --> 00:08:28.910 of vascular specialists. A totally separate discipline. 00:08:29.209 --> 00:08:31.790 Right. And they are studying a debilitating condition. 00:08:32.639 --> 00:08:34.960 Through their own exhaustive research, they publish 00:08:34.960 --> 00:08:37.940 hundreds of papers proving that this specific 00:08:37.940 --> 00:08:41.059 condition is consistently triggered and worsened 00:08:41.059 --> 00:08:43.940 when a patient's blood becomes thick, when their 00:08:43.940 --> 00:08:46.340 platelets become sticky and prone to clotting, 00:08:46.379 --> 00:08:49.019 and when their blood vessels aggressively clamp 00:08:49.019 --> 00:08:52.399 down. So you have room A holding a very specific, 00:08:52.580 --> 00:08:55.820 intricately cut key. Yes. And room B is staring 00:08:55.820 --> 00:08:58.240 at a locked door that perfectly matches that 00:08:58.240 --> 00:09:01.129 key's shape. It's almost tragic. It is. The researchers 00:09:01.129 --> 00:09:03.350 in both rooms are brilliant, well -funded, deeply 00:09:03.350 --> 00:09:05.690 diligent, but they never walk down the hall. 00:09:05.850 --> 00:09:08.009 They never cross paths at the cafeteria. The 00:09:08.009 --> 00:09:10.409 cure sits in the building, but it doesn't exist 00:09:10.409 --> 00:09:13.090 in anyone's mind. Swanson formally defined this 00:09:13.090 --> 00:09:16.169 tragedy of isolated knowledge. In 1986, he named 00:09:16.169 --> 00:09:18.710 it... undiscovered public knowledge. Not a great 00:09:18.710 --> 00:09:21.210 term. He hypothesized that the answers to some 00:09:21.210 --> 00:09:23.429 of our most pressing scientific questions were 00:09:23.429 --> 00:09:26.570 already fully documented, merely floating in 00:09:26.570 --> 00:09:28.769 the logical ether between isolated disciplines. 00:09:29.049 --> 00:09:31.149 I have to challenge this premise, though, especially 00:09:31.149 --> 00:09:33.669 when we apply it to modern science. Okay, let's 00:09:33.669 --> 00:09:35.639 hear it. The image of the isolated scientist 00:09:35.639 --> 00:09:38.080 toiling away in a dark room feels a bit dated. 00:09:38.200 --> 00:09:40.960 You know, today science is globally networked. 00:09:40.960 --> 00:09:42.980 We have massive interdisciplinary conferences. 00:09:43.139 --> 00:09:45.039 We have journals like Nature and Science that 00:09:45.039 --> 00:09:47.899 span multiple fields. Researchers have the Internet. 00:09:48.220 --> 00:09:50.480 All very true. So are we really supposed to believe 00:09:50.480 --> 00:09:52.539 that these brilliant minds just don't talk to 00:09:52.539 --> 00:09:54.879 each other? How does a blind spot that catastrophic 00:09:54.879 --> 00:09:57.379 actually survive in the real, highly collaborative 00:09:57.379 --> 00:10:01.120 world? What's fascinating here is that the isolation 00:10:01.120 --> 00:10:04.100 isn't driven by antisocial behavior at all. It's 00:10:04.100 --> 00:10:06.980 driven by pure, unforgiving arithmetic. Arithmetic? 00:10:07.139 --> 00:10:11.629 Yes. Science advances by zooming in. It grows 00:10:11.629 --> 00:10:14.889 by splitting into increasingly narrower subspecialties. 00:10:15.009 --> 00:10:17.629 A general biologist becomes a neurobiologist. 00:10:17.789 --> 00:10:20.889 That neurobiologist focuses down into cellular 00:10:20.889 --> 00:10:24.210 neurobiology. Eventually, their entire career 00:10:24.210 --> 00:10:26.610 is dedicated to the behavior of a single protein 00:10:26.610 --> 00:10:29.570 receptor on the membrane of a specific type of 00:10:29.570 --> 00:10:34.019 neuron. They get hyper -focused. Exactly. And 00:10:34.019 --> 00:10:36.679 as these fields deepen, the researchers are forced 00:10:36.679 --> 00:10:39.179 to develop highly specialized vocabularies just 00:10:39.179 --> 00:10:41.000 to

communicate the nuance of their findings. 00:10:41.240 --> 00:10:44.080 Oh, so they construct their own distinct dialects, 00:10:44.080 --> 00:10:46.559 like the cellular neurobiologist is practically 00:10:46.559 --> 00:10:48.779 speaking a different language than, say, a marine 00:10:48.779 --> 00:10:51.379 macrobiologist, even though they both technically 00:10:51.379 --> 00:10:54.679 hold degrees in biology. Right. Yeah. The jargon 00:10:54.679 --> 00:10:57.740 acts as a wall. A wall that is reinforced by 00:10:57.740 --> 00:11:00.559 the sheer volume of publishing, I imagine. Precisely. 00:11:00.600 --> 00:11:03.090 Yeah. A researcher barely has enough hours in 00:11:03.090 --> 00:11:05.509 the day to read the five or ten hyper -specific 00:11:05.509 --> 00:11:07.610 journals dedicated to their own protein receptor. 00:11:08.110 --> 00:11:11.149 The arithmetic of the situation is brutal. Because 00:11:11.149 --> 00:11:13.610 there's just too much to read. Right. As the 00:11:13.610 --> 00:11:15.850 number of scientific specialties grows linearly, 00:11:16.169 --> 00:11:19.129 the number of possible intersections or pairings 00:11:19.129 --> 00:11:21.230 between those specialties grows exponentially. 00:11:21.730 --> 00:11:25.149 Ah, okay. It is mathematically impossible for 00:11:25.149 --> 00:11:28.919 any human being, no matter... how collaborative 00:11:28.919 --> 00:11:31.419 they want to be, to read across all these boundaries. 00:11:32.120 --> 00:11:34.840 Swanson realized this exponential growth meant 00:11:34.840 --> 00:11:38.059 there was a massive, continuously expanding dark 00:11:38.059 --> 00:11:41.019 matter of logical connections that no human would 00:11:41.019 --> 00:11:43.679 ever perceive. He didn't just write a philosophical 00:11:43.679 --> 00:11:45.980 essay about this dark matter, though. He actively 00:11:45.980 --> 00:11:48.700 hunted for it to prove his point. He put his 00:11:48.700 --> 00:11:51.669 money where his mouth was. In 1986, he dropped 00:11:51.669 --> 00:11:55.009 two papers that the text notes are deeply intertwined. 00:11:55.129 --> 00:11:57.710 The first was published in the library quarterly, 00:11:57.850 --> 00:11:59.909 and it laid out the theoretical framework of 00:11:59.909 --> 00:12:02.970 undiscovered public knowledge. Which was the 00:12:02.970 --> 00:12:06.110 theory we just discussed. Right. The second paper, 00:12:06.190 --> 00:12:09.450 however, was a direct practical flex of his theory. 00:12:09.529 --> 00:12:12.090 He published it in Perspectives in Biology and 00:12:12.090 --> 00:12:14.870 Medicine. The title of that second paper is a 00:12:14.870 --> 00:12:17.750 masterpiece of intellectual collision. He called 00:12:17.750 --> 00:12:21.570 it Fish Oil. Raynaud syndrome and undiscovered 00:12:21.570 --> 00:12:23.789 public knowledge. Put yourself in the shoes of 00:12:23.789 --> 00:12:26.769 a practicing medical doctor in 1986. You're flipping 00:12:26.769 --> 00:12:28.750 through a medical journal over coffee and you 00:12:28.750 --> 00:12:31.429 find an article written by a library scientist 00:12:31.429 --> 00:12:33.870 telling you how to manage a vascular disease. 00:12:34.389 --> 00:12:37.169 It must have seemed so out of place. Raynaud 00:12:37.169 --> 00:12:39.330 syndrome, for context, is a painful disorder 00:12:39.330 --> 00:12:41.590 where the small blood vessels in the extremities, 00:12:41.629 --> 00:12:44.389 usually the fingers and toes, go into spasms 00:12:44.389 --> 00:12:47.309 and severely restrict blood flow. Often triggered 00:12:47.309 --> 00:12:49.909 by cold temperatures or stress. Exactly. The 00:12:49.909 --> 00:12:54.049 fingers literally turn stark white or blue. Swanson 00:12:54.049 --> 00:12:56.230 executed his two -room theory perfectly here 00:12:56.230 --> 00:12:58.570 by meticulously combing through the physical 00:12:58.570 --> 00:13:00.929 indexes of the literature. Remember, this is 00:13:00.929 --> 00:13:04.110 before modern search engines. He found one isolated 00:13:04.110 --> 00:13:07.029 set of articles. The fish oil ones. Right. Mostly 00:13:07.029 --> 00:13:09.570 in nutritional and hematology journals, detailing 00:13:09.570 --> 00:13:13.269 how consuming dietary fish oil alters blood viscosity 00:13:13.269 --> 00:13:16.669 and reduces vascular reactivity. Rume. Exactly. 00:13:17.210 --> 00:13:19.029 He then found a totally separate set of articles 00:13:19.029 --> 00:13:21.990 in rheumatology and vascular journals detailing 00:13:21.990 --> 00:13:24.129 the specific pathology of Raynaud's syndrome, 00:13:24.309 --> 00:13:26.970 which fundamentally involves high blood viscosity 00:13:26.970 --> 00:13:30.490 and hyperreactive clamping vessels. Room B. The 00:13:30.490 --> 00:13:33.009 ultimate proof of the silo effect here is the 00:13:33.009 --> 00:13:35.889 bibliography. Swanson demonstrated that these 00:13:35.889 --> 00:13:39.190 two massive bodies of literature had literally 00:13:39.190 --> 00:13:43.210 zero overlap. None at all. Not one single paper 00:13:43.210 --> 00:13:45.750 in the fish oil group cited a paper in the Raynaud's 00:13:45.750 --> 00:13:48.240 group. and vice versa. There was absolute silence 00:13:48.240 --> 00:13:51.320 between them. That's incredible. No human had 00:13:51.320 --> 00:13:54.159 ever formally suggested in print that fish oil 00:13:54.159 --> 00:13:56.759 might alleviate Raynaud's symptoms. The treatment 00:13:56.759 --> 00:14:00.220 existed entirely as a logical phantom, a connection 00:14:00.220 --> 00:14:02.500 drawn between two separate pieces of paper on 00:14:02.500 --> 00:14:05.019 Swanson's desk. He published the hypothesis, 00:14:05.340 --> 00:14:08.440 relying entirely on public records. He didn't 00:14:08.440 --> 00:14:10.759 draw a single drop of blood. He didn't run a 00:14:10.759 --> 00:14:14.110 double -blind placebo trial. He just read. He 00:14:14.110 --> 00:14:16.269 simply performed a synthesis of what humanity

00:14:16.269 --> 00:14:19.830 already knew but hadn't connected. The validation 00:14:19.830 --> 00:14:22.429 of his method is the most satisfying part of 00:14:22.429 --> 00:14:25.929 this timeline. Two years later, in 1988, clinical 00:14:25.929 --> 00:14:28.629 medical researchers went through the arduous, 00:14:28.629 --> 00:14:31.669 expensive, traditional process of running rigorous 00:14:31.669 --> 00:14:34.590 physical clinical trials on patients. The standard 00:14:34.590 --> 00:14:37.659 medical proof. Right. And their results definitively 00:14:37.659 --> 00:14:40.360 prove that dietary fish oil does, in fact, provide 00:14:40.360 --> 00:14:42.700 a significant beneficial effect for patients 00:14:42.700 --> 00:14:45.059 suffering from Raynaud syndrome. The crucial 00:14:45.059 --> 00:14:47.960 historical footnote to that 1988 clinical trial 00:14:47.960 --> 00:14:50.279 is that the medical researchers involved were 00:14:50.279 --> 00:14:53.139 completely oblivious to Don Swanson. They had 00:14:53.139 --> 00:14:56.360 no idea. They hadn't read his 1986 paper. They 00:14:56.360 --> 00:14:58.539 arrived at the exact same medical truth by taking 00:14:58.539 --> 00:15:02.179 the slow, grueling, expensive route of lab -based 00:15:02.179 --> 00:15:05.519 empirical science. They spent years of labor 00:15:05.519 --> 00:15:08.059 and untold thousands in funding to physically 00:15:08.059 --> 00:15:10.820 prove a reality that a librarian had already 00:15:10.820 --> 00:15:13.340 derived from a comfortable chair two years prior. 00:15:13.700 --> 00:15:15.960 It's wild to think about the resources saved 00:15:15.960 --> 00:15:19.360 if they had just known. It was irrefutable evidence 00:15:19.360 --> 00:15:21.460 that you could bypass the laboratory and generate 00:15:21.460 --> 00:15:24.500 genuine scientific breakthroughs simply by optimizing 00:15:24.500 --> 00:15:27.960 how you read. It transformed his hypothesis into 00:15:27.960 --> 00:15:31.450 a proven methodology. Swanson, the physicist, 00:15:31.610 --> 00:15:34.529 wasn't satisfied with a single anecdote. A one 00:15:34.529 --> 00:15:37.190 -off discovery could just be luck. Right. You 00:15:37.190 --> 00:15:40.230 need reproducible results in science. He needed 00:15:40.230 --> 00:15:42.509 to prove the existence of an underlying reliable 00:15:42.509 --> 00:15:45.509 system. He needed to formalize the mechanics 00:15:45.509 --> 00:15:47.830 so that other people could replicate his success 00:15:47.830 --> 00:15:51.330 across any discipline. And he immediately proved 00:15:51.330 --> 00:15:53.870 it wasn't a fluke. The source text outlines his 00:15:53.870 --> 00:15:56.769 subsequent victories. In 1988, he returned to 00:15:56.769 --> 00:15:58.909 Perspectives in Biology and Medicine with another 00:15:58.909 --> 00:16:02.779 paper. Migraine and magnesium, 11 neglected connections. 00:16:03.200 --> 00:16:06.580 11 connections. Using the exact same method, 00:16:06.740 --> 00:16:10.120 he found 11 distinct documented biological pathways 00:16:10.120 --> 00:16:12.919 in the literature that strongly suggested a magnesium 00:16:12.919 --> 00:16:15.980 deficiency played a causal role in migraine headaches. 00:16:16.320 --> 00:16:19.659 Two years after that, in 1990, he successfully 00:16:19.659 --> 00:16:22.600 connected dietary arginine intake to blood levels 00:16:22.600 --> 00:16:25.419 of somatomedans. He was just on a roll. He really 00:16:25.419 --> 00:16:28.409 was. Now, as the source material explicitly stresses, 00:16:28.570 --> 00:16:30.850 and we must firmly reiterate for everyone listening, 00:16:31.029 --> 00:16:34.370 we are discussing historical information theory, 00:16:34.529 --> 00:16:37.289 not practicing medicine. Absolutely. The biological 00:16:37.289 --> 00:16:40.350 outcomes are secondary to our focus today. Our 00:16:40.350 --> 00:16:43.129 focus is entirely on the machinery of discovery 00:16:43.129 --> 00:16:45.850 itself. It's about the methodology of finding 00:16:45.850 --> 00:16:48.929 the right questions to ask. Definitely. Do not 00:16:48.929 --> 00:16:51.470 go... treating medical conditions based on a 00:16:51.470 --> 00:16:53.690 deep dive. But looking at the machinery itself, 00:16:53.970 --> 00:16:56.830 how did Swanson formalize this? How did he take 00:16:56.830 --> 00:17:00.370 this massive abstract concept of reading everything 00:17:00.370 --> 00:17:03.330 and distill it into an actionable process? He 00:17:03.330 --> 00:17:05.509 reduced it to something called the ABC model, 00:17:05.750 --> 00:17:08.990 a model so elegant he boasted it could fit on 00:17:08.990 --> 00:17:12.190 a standard index card. Okay, break down the ABC 00:17:12.190 --> 00:17:14.950 model for us. It relies on the most basic form 00:17:14.950 --> 00:17:18.410 of logical deduction, the syllogism. The premise 00:17:18.410 --> 00:17:21.390 is this. If you find one distinct body of literature 00:17:21.390 --> 00:17:24.589 that empirically proves phenomenon A directly 00:17:24.589 --> 00:17:28.609 affects property B. Okay, A affects B. And you 00:17:28.609 --> 00:17:31.009 find a completely isolated body of literature 00:17:31.009 --> 00:17:33.829 proving that property B directly affects condition 00:17:33.829 --> 00:17:38.190 C. B affects C. It becomes highly logically probable 00:17:38.190 --> 00:17:41.890 that phenomenon A will affect condition C. Even 00:17:41.890 --> 00:17:44.210 if the letters A and C have never shared a paragraph 00:17:44.210 --> 00:17:46.309 in the history of human writing. Exactly the 00:17:46.309 --> 00:17:48.650 point. Mapping this back to his foundational 00:17:48.650 --> 00:17:51.750 success, A represented the dietary fish oil. 00:17:52.109 --> 00:17:55.109 B represented the intermediate biological mechanisms, 00:17:55.230 --> 00:17:58.210 the blood viscosity, the platelet aggregation, 00:17:58.349 --> 00:18:01.950 the vascular tone. Right. And C was the target 00:18:01.950 -->

00:18:05.269 disease, Raynaud's syndrome. It seems so simple 00:18:05.269 --> 00:18:07.589 when you lay it out like that. If we connect 00:18:07.589 --> 00:18:10.190 this to the bigger picture, the true brilliance 00:18:10.190 --> 00:18:13.410 of Swanson's framework isn't merely the ABC linkage. 00:18:13.970 --> 00:18:17.269 Lots of people can follow basic logic. The genius 00:18:17.269 --> 00:18:19.990 was his realization regarding the silence between 00:18:19.990 --> 00:18:23.170 the literatures. The disjointness. Swanson flipped 00:18:23.170 --> 00:18:25.509 our understanding of a dead end. He really did. 00:18:25.809 --> 00:18:27.910 Normally, if a researcher is looking into two 00:18:27.910 --> 00:18:30.349 topics and finds that no one is citing them together, 00:18:30.549 --> 00:18:32.589 they assume it's because there is no connection. 00:18:33.130 --> 00:18:35.809 That's the standard assumption. Swanson realized 00:18:35.809 --> 00:18:38.230 that when you have two highly specialized literatures 00:18:38.230 --> 00:18:41.190 sharing a critical B term, yet they never cross 00:18:41.190 --> 00:18:43.529 -reference each other, that silence is actually 00:18:43.529 --> 00:18:46.670 a massive flashing neon sign pointing directly 00:18:46.670 --> 00:18:49.869 at a void in human knowledge. The silence is 00:18:49.869 --> 00:18:52.430 the signal. The lack of citations between the 00:18:52.430 --> 00:18:54.809 domains is the precise reason you need to investigate 00:18:54.809 --> 00:18:57.170 that intersection. It's like finding a blank 00:18:57.170 --> 00:19:00.750 spot on a map. Exactly. But doing this manually, 00:19:00.930 --> 00:19:03.890 physically hauling massive bound indexes of medical 00:19:03.890 --> 00:19:06.130 abstracts off a library shelf and cross -referencing 00:19:06.130 --> 00:19:10.329 them by hand was excruciatingly slow work. Oh, 00:19:10.349 --> 00:19:12.910 I can't even imagine. Swanson recognized the 00:19:12.910 --> 00:19:15.859 need for automation. By 1989, he was already 00:19:15.859 --> 00:19:18.880 publishing step -by -step strategies for executing 00:19:18.880 --> 00:19:21.720 this logical hunt using the very primitive online 00:19:21.720 --> 00:19:24.359 databases available at the time. We are talking 00:19:24.359 --> 00:19:27.200 about the era of dial -up modems and clunky Boolean 00:19:27.200 --> 00:19:29.359 search queries, but he pushed the limits of the 00:19:29.359 --> 00:19:32.940 technology. Fast forward to 1997, he partners 00:19:32.940 --> 00:19:36.059 with a researcher named Neil R. Smallhiser, and 00:19:36.059 --> 00:19:39.140 they manifest the ABC model into actual software. 00:19:39.559 --> 00:19:42.740 They built a system called Aerosmith. Aerosmith. 00:19:42.880 --> 00:19:45.730 How did it work? Aerosmith was a targeted search 00:19:45.730 --> 00:19:48.630 engine designed specifically to interface with 00:19:48.630 --> 00:19:51.230 Medline, which is the massive U .S. National 00:19:51.230 --> 00:19:54.009 Library of Medicine database. It was programmed 00:19:54.009 --> 00:19:57.930 to hunt for disconnected literatures. So it automated 00:19:57.930 --> 00:20:02.029 the ABC hunt. Yes. A researcher would input a 00:20:02.029 --> 00:20:04.549 target topic, their C term, perhaps a specific 00:20:04.549 --> 00:20:08.000 disease. Aerosmith would then rapidly scan the 00:20:08.000 --> 00:20:10.160 database to find a completely separate literature 00:20:10.160 --> 00:20:13.500 that shared specific biological vocabulary, the 00:20:13.500 --> 00:20:16.539 B terms. Okay. But crucially, it filtered out 00:20:16.539 --> 00:20:18.680 anything that already cited the original literature. 00:20:19.019 --> 00:20:22.880 It automated the hunt for the silence, presenting 00:20:22.880 --> 00:20:25.740 the user with a list of shared B terms as candidate 00:20:25.740 --> 00:20:28.559 bridges to explore. It's essentially a metal 00:20:28.559 --> 00:20:30.759 detector for undiscovered knowledge. That's a 00:20:30.759 --> 00:20:33.130 great way to put it. This single piece of software 00:20:33.130 --> 00:20:35.670 essentially birthed the modern field of literature 00:20:35.670 --> 00:20:38.670 -based discovery. The specific pattern of finding 00:20:38.670 --> 00:20:41.470 these silent, indirect links is universally recognized 00:20:41.470 --> 00:20:44.390 today as Swanson linking. He was awarded the 00:20:44.390 --> 00:20:46.690 highest honor from the American Society for Information 00:20:46.690 --> 00:20:49.509 Science and Technology in 2000 for this work. 00:20:49.690 --> 00:20:52.069 And he remained an active thinker until he passed 00:20:52.069 --> 00:20:55.109 away in 2012 at the age of 88. His legacy is 00:20:55.109 --> 00:20:57.779 monumental. However, the source text highlights 00:20:57.779 --> 00:21:00.819 a critical historical bottleneck inherent in 00:21:00.819 --> 00:21:03.119 Aerosmith. Right. It wasn't perfect. It was a 00:21:03.119 --> 00:21:05.539 cyborg approach, a partnership between man and 00:21:05.539 --> 00:21:08.599 machine. The software was incredibly efficient 00:21:08.599 --> 00:21:11.700 at surfacing the candidate bridges and pulling 00:21:11.700 --> 00:21:14.880 up the possible B terms. But the human being 00:21:14.880 --> 00:21:17.740 was still the limiting factor. The machine could 00:21:17.740 --> 00:21:20.700 point, but the human had to read. Exactly. An 00:21:20.700 --> 00:21:23.140 expert still had to sit down at a desk. pour 00:21:23.140 --> 00:21:25.480 a cup of coffee, and manually read through the 00:21:25.480 --> 00:21:28.440 thousands of abstracts the machine flagged. They 00:21:28.440 --> 00:21:30.500 had to use human reasoning to parse the complex 00:21:30.500 --> 00:21:33.640 biology, understand the context, and deduce which 00:21:33.640 --> 00:21:36.440 of those theoretical bridges actually made scientific 00:21:36.440 --> 00:21:39.559 sense. The machine acted as a very fast librarian 00:21:39.559 --> 00:21:41.700

pointing at shelves, but the humans still had 00:21:41.700 --> 00:21:44.319 to do the reading. And human reading is inherently 00:21:44.319 --> 00:21:48.200 biologically slow. We need sleep. We get distracted. 00:21:48.480 --> 00:21:51.180 We lose our place. For decades after Aerosmith 00:21:51.180 --> 00:21:53.339 was launched, that biological bottleneck prevented 00:21:53.339 --> 00:21:55.880 Swanson's methodology from reaching its explosive 00:21:55.880 --> 00:21:59.380 theoretical scale. The overarching question of 00:21:59.380 --> 00:22:03.019 this entire deep dive emerges right here. What 00:22:03.019 --> 00:22:06.140 happens to the pace of scientific discovery when 00:22:06.140 --> 00:22:08.880 the entity doing the reading, the contextual 00:22:08.880 --> 00:22:10.900 reasoning, and the logical cross -referencing 00:22:10.900 --> 00:22:14.960 no longer requires sleep? That is the question. 00:22:15.980 --> 00:22:19.039 What happens when you can instantiate Don Swanson's 00:22:19.039 --> 00:22:22.920 solitary genius a hundred times over in a cloud 00:22:22.920 --> 00:22:25.759 computing cluster? That question transports us 00:22:25.759 --> 00:22:28.240 violently from the analog constraints of the 00:22:28.240 --> 00:22:30.759 20th century directly into the bleeding edge 00:22:30.759 --> 00:22:33.599 of the present day. We are jumping forward to 00:22:33.599 --> 00:22:38.539 October 4th, 2026. A massive leap forward. A 00:22:38.539 --> 00:22:41.359 technology company named Val's AI publishes a 00:22:41.359 --> 00:22:44.279 highly technical post authored by someone named 00:22:44.279 --> 00:22:48.660 Jebby Joff. The title of the post is Two Room 00:22:48.660 --> 00:22:52.539 Temperature Antiferromagnetic Semiconductor Candidates. 00:22:52.579 --> 00:22:55.680 To a layperson, that title reads like a dense, 00:22:55.799 --> 00:22:59.019 impenetrable wall of material science jargon. 00:22:59.160 --> 00:23:01.000 I mean, I barely know what half those words mean. 00:23:01.240 --> 00:23:04.140 Right. But the methodology buried in that post 00:23:04.140 --> 00:23:07.160 represents a seismic shift in how science operates. 00:23:08.210 --> 00:23:10.690 Val's AI casually announced that these two candidate 00:23:10.690 --> 00:23:13.109 materials were not discovered by a team of human 00:23:13.109 --> 00:23:15.589 postdocs grinding away for months in a university 00:23:15.589 --> 00:23:18.930 lab. No lab coats involved. Zero. The discovery 00:23:18.930 --> 00:23:22.049 was generated entirely by an orchestrated swarm 00:23:22.049 --> 00:23:25.349 of artificial intelligence agents. Specifically, 00:23:25.589 --> 00:23:30.089 a team of over 90 Claude Opus 5 .5 agents. Here's 00:23:30.089 --> 00:23:32.710 where it gets really interesting. We need to 00:23:32.710 --> 00:23:35.500 visualize the sheer scale and... ferocious velocity 00:23:35.500 --> 00:23:39.279 of this digital swarm. 90 highly advanced AI 00:23:39.279 --> 00:23:41.819 agents. They aren't just

skimming Wikipedia articles. 00:23:42.099 --> 00:23:44.359 No, they are diving deep into the technical literature. 00:23:44.980 --> 00:23:47.240 According to the independent coverage detailing 00:23:47.240 --> 00:23:50.359 this event, VALS AI unleashed this swarm and 00:23:50.359 --> 00:23:52.240 gave them a hard deadline of just three days. 00:23:52.440 --> 00:23:56.349 72 hours. A long weekend. In that incredibly 00:23:56.349 --> 00:23:58.710 compressed window, the agents autonomously organized 00:23:58.710 --> 00:24:01.430 themselves into specialized parallel lanes of 00:24:01.430 --> 00:24:04.150 labor. The architecture of the swarm is a marvel 00:24:04.150 --> 00:24:06.750 of modern software engineering. You have one 00:24:06.750 --> 00:24:08.970 entire lane of agents completely dedicated to 00:24:08.970 --> 00:24:11.170 literature review. The readers. Their sole function 00:24:11.170 --> 00:24:13.369 is to read millions of abstracts and papers, 00:24:13.509 --> 00:24:16.190 acting as the ultimate realization of Swanson's 00:24:16.190 --> 00:24:19.059 room -scanning librarian. Simultaneously, you 00:24:19.059 --> 00:24:21.579 have a second lane of agents dedicated to running 00:24:21.579 --> 00:24:25.099 highly complex physics -based simulations based 00:24:25.099 --> 00:24:27.079 on the molecular structures the first lane is 00:24:27.079 --> 00:24:30.059 finding. The testers. And the third lane is the 00:24:30.059 --> 00:24:32.779 most crucial part. It's dedicated entirely to 00:24:32.779 --> 00:24:35.660 adversarial checking. The skeptics. Exactly. 00:24:35.839 --> 00:24:37.940 You have a whole team of agents whose entire 00:24:37.940 --> 00:24:40.700 programmed existence is dedicated to aggressively 00:24:40.700 --> 00:24:43.319 attacking and trying to disprove the theories 00:24:43.319 --> 00:24:45.920 that the other agents are building. I read this 00:24:45.920 --> 00:24:47.940 breakdown and it immediately plays out in my 00:24:47.940 --> 00:24:50.119 head like a high -stakes scientific heist movie. 00:24:50.400 --> 00:24:53.299 I love that analogy. Instead of a crew of human 00:24:53.299 --> 00:24:55.440 specialists... you know, the safecracker, the 00:24:55.440 --> 00:24:57.599 inside man, the getaway driver, orchestrating 00:24:57.599 --> 00:25:00.460 a bank robbery, you have an army of 90 robotic 00:25:00.460 --> 00:25:03.200 scholars. Right. They digitally kick down the 00:25:03.200 --> 00:25:05.619 doors of the global research archive, tear through 00:25:05.619 --> 00:25:07.680 centuries of literature over a long weekend, 00:25:07.859 --> 00:25:10.339 scream at each other over the nuances of quantum 00:25:10.339 --> 00:25:12.940 mechanics, and walk out on Monday morning with 00:25:12.940 --> 00:25:15.559 the prize. The computational brute force is staggering. 00:25:16.180 --> 00:25:18.680 The text notes that just one of those specialized 00:25:18.680 --> 00:25:22.880 agent lanes submitted approximately 750 distinct 00:25:23.500 --> 00:25:26.079 quantum chemistry simulation jobs during that 00:25:26.079 --> 00:25:29.519 72 hour sprint. That's insane. A human team would 00:25:29.519 --> 00:25:32.200 take months, maybe years, to set up and run that 00:25:32.200 --> 00:25:35.430 many discrete simulations. But if we strip away 00:25:35.430 --> 00:25:38.369 the flashy AI terminology and the mind -bending 00:25:38.369 --> 00:25:41.329 speed, the fundamental operation occurring inside 00:25:41.329 --> 00:25:44.309 those servers is indistinguishable from what 00:25:44.309 --> 00:25:47.549 happened in Chicago in 1986. It's the exact same 00:25:47.549 --> 00:25:51.210 logic. The core ABC model that Swanson sketched 00:25:51.210 --> 00:25:53.849 on an index card remained flawlessly intact. 00:25:54.309 --> 00:25:57.849 A affects B, B affects C, therefore A affects 00:25:57.849 --> 00:26:00.369 C. The logic didn't evolve. The underlying arithmetic 00:26:00.369 --> 00:26:02.890 of discovery is a constant. The only variable 00:26:02.890 --> 00:26:05.390 that changed was the nature of the reader. But 00:26:05.390 --> 00:26:07.309 wait, hold on. I need to push back here because 00:26:07.309 --> 00:26:10.049 we are talking about 90 large language model 00:26:10.049 --> 00:26:13.109 agents recursively feeding each other data. I 00:26:13.109 --> 00:26:14.750 know how these generative models work underneath. 00:26:15.259 --> 00:26:18.000 They suffer from model collapse. They are notorious 00:26:18.000 --> 00:26:21.059 for hallucinating wildly. That is a very valid 00:26:21.059 --> 00:26:23.859 concern. If I ask a standard AI to write a legal 00:26:23.859 --> 00:26:26.720 brief, there is a very real chance it just invents 00:26:26.720 --> 00:26:29.779 fake court cases that sound perfectly real. How 00:26:29.779 --> 00:26:32.140 in the world are we trusting a swarm of these 00:26:32.140 --> 00:26:34.339 things to execute rigorous quantum chemistry 00:26:34.339 --> 00:26:37.500 and literature synthesis without just fabricating 00:26:37.500 --> 00:26:39.619 a perfect holy grail material out of thin air 00:26:39.619 --> 00:26:42.289 just to satisfy the user's prompt? That is the 00:26:42.289 --> 00:26:44.589 exact vulnerability that the adversarial lane 00:26:44.589 --> 00:26:47.089 is designed to mitigate. And it's why the simulation 00:26:47.089 --> 00:26:50.269 lane is so vital. Okay, explain how the simulation 00:26:50.269 --> 00:26:53.490 lane stops the hallucinations. The language models 00:26:53.490 --> 00:26:56.069 aren't just generating text predicting that a 00:26:56.069 --> 00:26:58.710 material works. They are generating the exact 00:26:58.710 --> 00:27:01.450 input code required to run established, rigid 00:27:01.450 --> 00:27:04.869 physics simulations. Oh, I see. The AI proposes 00:27:04.869 --> 00:27:07.029 the structure, but the hard math of the simulation 00:27:07.029 --> 00:27:10.049 environment acts as the reality check. If the 00:27:10.049 --> 00:27:12.670 AI hallucinates a

molecule that violates the 00:27:12.670 --> 00:27:15.289 laws of thermodynamics, the simulation simply 00:27:15.289 --> 00:27:18.750 fails. The language model proposes, but the physics 00:27:18.750 --> 00:27:21.710 engine disposes. The physics engine keeps them 00:27:21.710 --> 00:27:24.309 honest. That makes sense. It's not just a text 00:27:24.309 --> 00:27:27.109 output. It's executable code being tested against 00:27:27.109 --> 00:27:31.549 physical laws. It does. And this successful architecture 00:27:31.549 --> 00:27:34.349 represents a profound philosophical paradigm 00:27:34.349 --> 00:27:37.500 shift. Think about the entire history of science 00:27:37.500 --> 00:27:40.619 up to this point. Okay. The scarce resource, 00:27:40.859 --> 00:27:43.500 the ultimate bottleneck in discovering undiscovered 00:27:43.500 --> 00:27:46.619 public knowledge, was always attention. It was 00:27:46.619 --> 00:27:49.720 human cognitive bandwidth. Finding a hidden link 00:27:49.720 --> 00:27:53.140 required a human to painfully, slowly read across 00:27:53.140 --> 00:27:56.329 the dark spaces between disciplines. Human attention 00:27:56.329 --> 00:28:00.380 was the most valuable currency. Exactly. But 00:28:00.380 --> 00:28:03.599 in the span of a 72 -hour AI sprint, that specific 00:28:03.599 --> 00:28:06.890 bottleneck completely evaporated. An orchestrated 00:28:06.890 --> 00:28:10.269 swarm of 90 agents can now propose more logically 00:28:10.269 --> 00:28:13.250 sound, simulation -backed candidate links in 00:28:13.250 --> 00:28:15.890 a single weekend than a traditional human research 00:28:15.890 --> 00:28:17.990 department could physically synthesize and test 00:28:17.990 --> 00:28:21.740 in a decade. So the bottleneck shifted. Yes. 00:28:22.019 --> 00:28:25.359 The scarce resource is no longer attention. The 00:28:25.359 --> 00:28:27.420 new bottleneck in human progress is verification. 00:28:28.200 --> 00:28:31.640 It is the physical act of taking this newly generated 00:28:31.640 --> 00:28:35.119 mountain of AI hypotheses and proving them. in 00:28:35.119 --> 00:28:38.539 the messy physical world. So we have this incredibly 00:28:38.539 --> 00:28:41.759 powerful, specialized swarm of 90 AI scholars 00:28:41.759 --> 00:28:45.000 tearing through the digital archives. What exactly 00:28:45.000 --> 00:28:47.200 was the C -term? What was the target that initiated 00:28:47.200 --> 00:28:50.720 this massive 72 -hour computational sprint? They 00:28:50.720 --> 00:28:53.160 were dispatched to hunt down a very specific, 00:28:53.319 --> 00:28:56.140 highly coveted material property in the realm 00:28:56.140 --> 00:28:58.660 of advanced computer memory. They were looking 00:28:58.660 --> 00:29:02.299 for a substance that qualifies as Luttinger compensated. 00:29:02.859 --> 00:29:05.400 To understand why a Luttinger compensated material 00:29:05.400 --> 00:29:08.480 is the ultimate prize, we need to do a quick, 00:29:08.579 --> 00:29:11.279 simplified ELI -5 breakdown of the physics of 00:29:11.279 --> 00:29:14.279 magnets. It is fascinating and completely essential 00:29:14.279 --> 00:29:16.759 to understanding what the AI actually found. 00:29:16.940 --> 00:29:20.000 Yes, let's break down the magnetism. When engineers 00:29:20.000 --> 00:29:22.440 try to build magnetic computer memory, they are 00:29:22.440 --> 00:29:24.640 generally choosing between two primary types 00:29:24.640 --> 00:29:27.480 of magnets. First, you have ordinary ferromagnets. 00:29:27.720 --> 00:29:29.960 These are your classic, everyday magnets. Right. 00:29:30.000 --> 00:29:31.859 The kind you stick to your refrigerator. Right. 00:29:32.599 --> 00:29:34.940 At the atomic level, ferromagnets are highly 00:29:34.940 --> 00:29:37.480 organized. They sort their electrons by spin, 00:29:37.680 --> 00:29:40.099 meaning all the tiny magnetic moments of the 00:29:40.099 --> 00:29:43.160 atoms align, pointing in the exact same direction. 00:29:43.460 --> 00:29:45.519 Like a school of fish all swimming the same way. 00:29:45.619 --> 00:29:48.140 Precisely. In the context of computer memory, 00:29:48.359 --> 00:29:51.480 this uniform sorting is incredibly useful. It 00:29:51.480 --> 00:29:53.940 allows a device to very easily read and write 00:29:53.940 --> 00:29:57.259 data. You flip the spin direction, you flip the 00:29:57.259 --> 00:29:59.500 bit from a zero to a one. They are fantastic 00:29:59.500 --> 00:30:02.180 for storing data, but they have a fatal flaw 00:30:02.180 --> 00:30:04.180 when it comes to modern computing. Because all 00:30:04.180 --> 00:30:06.140 those microscopic spins are aligned in the same 00:30:06.140 --> 00:30:08.920 direction, they generate a collective, macroscopic, 00:30:08.920 --> 00:30:11.799 stray magnetic field that extends outward beyond 00:30:11.799 --> 00:30:14.180 the physical material. They bleed their magnetic 00:30:14.180 --> 00:30:17.140 force. Exactly. Think of a ferromagnet like a 00:30:17.140 --> 00:30:19.809 really messy eater. If you were trying to build 00:30:19.809 --> 00:30:22.329 ultra -dense computer memory, packing billions 00:30:22.329 --> 00:30:25.049 of these microscopic magnets nanometers apart 00:30:25.049 --> 00:30:27.509 on a silicon chip, those stray fields become 00:30:27.509 --> 00:30:29.549 a nightmare. Because they interfere with each 00:30:29.549 --> 00:30:32.109 other. One bit flips to write data, and its stray 00:30:32.109 --> 00:30:34.170 magnetic field accidentally scrambles the data 00:30:34.170 --> 00:30:36.730 in the bit right next to it. Ferromagnets are 00:30:36.730 --> 00:30:39.490 fundamentally bad for dense packing. To solve 00:30:39.490 --> 00:30:41.769 the interference problem, engineers turned to 00:30:41.769 --> 00:30:45.529 the second type, antiferromagnets. In these materials, 00:30:45.710 --> 00:30:48.190 the internal magnetic moments of the atoms don't 00:30:48.190 --> 00:30:52.170 align. They perfectly alternate. Up, down, up, 00:30:52.250 -->

00:30:54.230 down in a perfect grid. Like a checkerboard. 00:30:54.410 --> 00:30:57.130 Exactly. Because every up is immediately canceled 00:30:57.130 --> 00:31:00.410 out by a down, the material produces zero net 00:31:00.410 --> 00:31:03.269 macroscopic magnetic field. There is no stray 00:31:03.269 --> 00:31:06.049 field leaking out. You can pack antiferro magnets 00:31:06.049 --> 00:31:08.890 incredibly close together without them ever whispering 00:31:08.890 --> 00:31:11.369 to their neighbors. They are perfect for ultra 00:31:11.369 --> 00:31:14.359 -dense packing. However... Ordinary antiferromagnets 00:31:14.359 --> 00:31:16.579 possess their own fatal flaw for memory applications. 00:31:17.400 --> 00:31:20.180 Because those spins perfectly cancel each other 00:31:20.180 --> 00:31:23.019 out internally, the material cannot easily sort 00:31:23.019 --> 00:31:25.559 electrons by spin when an electrical current 00:31:25.559 --> 00:31:28.099 passes through it. And without that spin sorting 00:31:28.099 --> 00:31:30.519 capability, it becomes incredibly difficult to 00:31:30.519 --> 00:31:32.680 actually read or write data to the material. 00:31:32.960 --> 00:31:36.359 It's a cruel trade -off. The ferromagnets give 00:31:36.359 --> 00:31:38.640 you the easy data sorting, but completely ruin 00:31:38.640 --> 00:31:41.900 your ability to pack them densely. The antiferromagnets 00:31:41.900 --> 00:31:44.380 give you perfect density, but ruin your ability 00:31:44.380 --> 00:31:47.559 to easily interact with the data. So what's the 00:31:47.559 --> 00:31:50.599 solution? The goal, the absolute holy grail for 00:31:50.599 --> 00:31:53.720 material scientists, is to find a singular material 00:31:53.720 --> 00:31:56.960 that miraculously possesses the virtues of both 00:31:56.960 --> 00:31:59.460 without the flaws of either. The checklist for 00:31:59.460 --> 00:32:02.599 this holy grail is brutally restrictive. First, 00:32:02.740 --> 00:32:05.740 the material must have absolutely zero net magnetism 00:32:05.740 --> 00:32:08.579 so it produces no stray fields. Right. Second, 00:32:08.779 --> 00:32:10.640 it must have the band gap of a semiconductor, 00:32:10.960 --> 00:32:13.440 meaning its electrical conductivity can be precisely 00:32:13.440 --> 00:32:16.480 controlled. Crucial for electronics. And third, 00:32:16.779 --> 00:32:19.759 despite having zero net magnetism, its internal 00:32:19.759 --> 00:32:22.720 quantum structure must somehow still sort electrons 00:32:22.720 --> 00:32:25.480 by spin, allowing data to be read and written. 00:32:26.220 --> 00:32:28.759 That highly contradictory, incredibly rare combination 00:32:28.759 --> 00:32:31.980 of traits is the definition of a Lettinger compensated 00:32:31.980 --> 00:32:34.099 material. It's like asking for material that 00:32:34.099 --> 00:32:37.539 is entirely transparent, but casts a pitch black 00:32:37.539 --> 00:32:40.339 shadow. It's a wildly specific set of physical 00:32:40.339 --> 00:32:43.920 demands. So Val's AI unleashes the swarm with 00:32:43.920 --> 00:32:46.339 this impossible checklist. What did the agents 00:32:46.339 --> 00:32:49.319 return with? The text outlines two specific candidates. 00:32:49.660 --> 00:32:51.819 The first candidate proposed by the AI was an 00:32:51.819 --> 00:32:55.069 entirely novel, newly designed compound. The 00:32:55.069 --> 00:32:58.289 chemical formula is Y -B -A -M -N -F -E -O -5. 00:32:58.950 --> 00:33:02.309 That's yttrium, barium, manganese, iron, and 00:33:02.309 --> 00:33:06.099 oxygen. Correct. When the AI ran this novel oxide 00:33:06.099 --> 00:33:08.740 through its simulation lane, it looked utterly 00:33:08.740 --> 00:33:11.680 flawless on the digital chalkboard. It hit every 00:33:11.680 --> 00:33:14.180 single theoretical requirement for looting your 00:33:14.180 --> 00:33:17.559 compensation. A perfect hit. However, this is 00:33:17.559 --> 00:33:19.500 where the adversarial lane proved its worth. 00:33:20.119 --> 00:33:22.680 The skeptic agents identified a catastrophic 00:33:22.680 --> 00:33:26.529 physical catch. For this specific oxide to function 00:33:26.529 --> 00:33:29.329 as a semiconductor, the manganese and iron atoms 00:33:29.329 --> 00:33:32.289 within the crystal must be arranged in a flawlessly 00:33:32.289 --> 00:33:35.089 perfect alternating 3D checkerboard pattern. 00:33:35.690 --> 00:33:39.230 The theory is sound, but the problem is the messy 00:33:39.230 --> 00:33:42.819 reality of physics and manufacturing. And the 00:33:42.819 --> 00:33:45.359 adversarial simulations, they revealed that when 00:33:45.359 --> 00:33:47.839 you attempt to physically synthesize this oxide 00:33:47.839 --> 00:33:50.799 in a real -world laboratory, the ambient thermal 00:33:50.799 --> 00:33:53.400 energy required to create the crystal causes 00:33:53.400 --> 00:33:55.680 that perfect checkerboard structure to scramble 00:33:55.680 --> 00:33:59.019 into chaos. It falls apart. Right. The structure 00:33:59.019 --> 00:34:01.680 inherently breaks down, and it loses its precise 00:34:01.680 --> 00:34:05.279 ordering at around 950 Kelvin. This is a massive 00:34:05.279 --> 00:34:08.619 problem because 950 Kelvin is significantly below 00:34:08.619 --> 00:34:10.860 the extreme temperatures that are normally required 00:34:10.860 --> 00:34:13.440 to bake and synthesize these complex oxides. 00:34:13.539 --> 00:34:16.199 So you literally can't bake it without breaking 00:34:16.199 --> 00:34:20.559 it. Exactly. So now what? It is a beautiful mathematical 00:34:20.559 --> 00:34:23.860 theory that melts the moment you try to build 00:34:23.860 --> 00:34:26.659 it. But then we get to the second candidate. 00:34:26.760 --> 00:34:29.880 And this is the revelation that gives you absolute 00:34:29.880 --> 00:34:33.050 goosebumps because this is pure vintage. Don 00:34:33.050 --> 00:34:36.030 Swanson brought into the 21st century. The second 00:34:36.030 --> 00:34:38.250 candidate

wasn't some novel compound dreamed 00:34:38.250 --> 00:34:41.969 up by a neural network. It was a dusty book pulled 00:34:41.969 --> 00:34:44.769 off a forgotten shelf. Yes. The material is designated 00:34:44.769 --> 00:34:50.690 KVCRCN6. A potassium, vanadium, and chromium 00:34:50.690 --> 00:34:53.429 cyanide complex. Chemically speaking, it belongs 00:34:53.429 --> 00:34:56.070 to the exact same broad structural family as 00:34:56.070 --> 00:34:59.050 the famous pigment Prussian blue. Prussian blue. 00:34:59.639 --> 00:35:02.000 We are talking about a chemical structure whose 00:35:02.000 --> 00:35:04.820 roots go back to an accidental synthesis in Berlin 00:35:04.820 --> 00:35:09.139 around the year 1706. It was used in classic 00:35:09.139 --> 00:35:12.280 paintings. It was the literal blue in traditional 00:35:12.280 --> 00:35:15.400 architectural blueprints. And now, a chemical 00:35:15.400 --> 00:35:18.539 cousin of a 300 -year -old pigment is being flagged 00:35:18.539 --> 00:35:21.179 by a supercomputing swarm as the foundational 00:35:21.179 --> 00:35:23.380 architecture for next -generation quantum memory. 00:35:23.599 --> 00:35:26.880 The historical irony is palpable. But the AI 00:35:26.880 --> 00:35:30.079 agents didn't pull a document from 1706. They 00:35:30.079 --> 00:35:32.139 scoured the modern literature and unearthed a 00:35:32.139 --> 00:35:34.360 specific paper published in the Journal of the 00:35:34.360 --> 00:35:37.400 American Chemical Society. Meticulously authored 00:35:37.400 --> 00:35:39.800 by two chemists, Stephen M. Holmes and Gregory 00:35:39.800 --> 00:35:42.820 S. Girolami, based out of the University of Illinois 00:35:42.820 --> 00:35:45.559 at Urbana -Champaign. The publication date is 00:35:45.559 --> 00:35:48.199 the key here. The paper was published in 1999, 00:35:48.619 --> 00:35:52.579 27 years before the Vols AI swarm was even a 00:35:52.579 --> 00:35:55.760 line of code on a server. The paper was formally 00:35:55.760 --> 00:36:00.329 received on March 24th, 1999. In this text, Holmes 00:36:00.329 --> 00:36:02.670 and Gerolami detail their successful synthesis 00:36:02.670 --> 00:36:06.269 of this specific molecule -based magnet. And 00:36:06.269 --> 00:36:08.829 what did they find? They documented a physical 00:36:08.829 --> 00:36:12.849 property that was, at the time, astounding. The 00:36:12.849 --> 00:36:15.010 material exhibited what they called an unprecedented 00:36:15.010 --> 00:36:18.590 magnetic ordering temperature of 376 Kelvin. 00:36:19.869 --> 00:36:23.429 376 Kelvin is roughly 103 degrees Celsius. It's 00:36:23.429 --> 00:36:26.150 just above the boiling point of water. To understand 00:36:26.150 --> 00:36:28.650 why that is a monumental achievement, you have 00:36:28.650 --> 00:36:31.409 to realize that most of these weird experimental 00:36:31.409 --> 00:36:34.969 molecule -based magnets only maintain their magnetic 00:36:34.969 --> 00:36:37.650 properties when you freeze them down

near absolute 00:36:37.650 --> 00:36:41.010 zero using liquid helium. Right. They are incredibly 00:36:41.010 --> 00:36:44.409 fragile to heat. A material like this that remains 00:36:44.409 --> 00:36:47.469 stable and magnetic sitting on a desk in a warm 00:36:47.469 --> 00:36:50.969 room is a massive deal if you want to build actual 00:36:50.969 --> 00:36:53.750 commercial electronics. Nobody wants a laptop 00:36:53.750 --> 00:36:56.789 that requires a liquid helium cooling tank. The 00:36:56.789 --> 00:36:58.909 chemists had achieved a masterpiece of synthesis. 00:36:59.210 --> 00:37:02.150 They designed the compound so the inherent magnetism 00:37:02.150 --> 00:37:04.690 of the two transition metals involved, the chromium 00:37:04.690 --> 00:37:07.530 and the vanadium, would almost entirely cancel 00:37:07.530 --> 00:37:09.710 each other out at room temperature while maintaining 00:37:09.710 --> 00:37:12.519 structural integrity. But here is the profound, 00:37:12.599 --> 00:37:15.340 almost heartbreaking irony of the entire discovery. 00:37:15.659 --> 00:37:18.980 In that very same 1999 publication, right after 00:37:18.980 --> 00:37:21.320 detailing this incredible high temperature stability, 00:37:22.079 --> 00:37:25.340 Holmes and Juralami explicitly state their view 00:37:25.340 --> 00:37:27.800 on its utility. I know exactly what you're going 00:37:27.800 --> 00:37:30.079 to say. I am quoting directly from the source 00:37:30.079 --> 00:37:33.679 text here. They wrote that these solids do not 00:37:33.679 --> 00:37:36.659 at present have real world uses. They formally 00:37:36.659 --> 00:37:39.780 wrote off their own creation. And to be entirely 00:37:39.780 --> 00:37:42.179 fair to their scientific rigor, their assessment 00:37:42.179 --> 00:37:45.519 was completely rational in 1999. Why? Because 00:37:45.519 --> 00:37:47.800 the vast majority of these molecule -based materials 00:37:47.800 --> 00:37:51.239 were so fragile and required such extreme cold, 00:37:51.340 --> 00:37:54.019 the entire class of materials was widely viewed 00:37:54.019 --> 00:37:56.719 by the industry as a neat academic curiosity 00:37:56.719 --> 00:37:59.000 rather than a viable industrial stepping stone. 00:37:59.340 --> 00:38:01.679 Furthermore, they couldn't possibly foresee the 00:38:01.679 --> 00:38:04.039 application. because the desperate need for ultra 00:38:04.039 --> 00:38:06.940 -dense, spin -sorted, anti -ferromagnetic computer 00:38:06.940 --> 00:38:09.519 memory didn't truly exist yet as a mainstream 00:38:09.519 --> 00:38:11.980 engineering bottleneck. It is the ultimate example 00:38:11.980 --> 00:38:14.119 of being ahead of your time. It's like a blacksmith 00:38:14.119 --> 00:38:16.920 inventing a flawless titanium spark plug in the 00:38:16.920 --> 00:38:20.679 year 1800. It's a metallurgical marvel, but the 00:38:20.679 --> 00:38:23.179 internal combustion engine hasn't been conceptualized 00:38:23.179 --> 00:38:25.420 yet, so the blacksmith just puts it on a shelf, 00:38:25.519 --> 00:38:28.139 sighs, and writes in his journal, this object 00:38:28.139 --> 00:38:32.750 has no real -world use. And so the KVCR -CN6 00:38:32.750 --> 00:38:35.929 compound sat on that proverbial shelf for 27 00:38:35.929 --> 00:38:38.550 years. The original chemists moved on to other 00:38:38.550 --> 00:38:41.570 research. The global scientific community ignored 00:38:41.570 --> 00:38:46.909 it until October 2026, when the VALS AI agents 00:38:46.909 --> 00:38:49.630 swept through the archives. And what the agents 00:38:49.630 --> 00:38:51.889 accomplished was not a new physical experiment. 00:38:51.989 --> 00:38:54.449 They didn't synthesize a single gram of the powder 00:38:54.449 --> 00:38:57.789 in a lab. The value the AI added was a completely 00:38:57.789 --> 00:39:00.829 new reading. of the old material. They examined 00:39:00.829 --> 00:39:04.190 the 1999 data strictly through the lens of 2026 00:39:04.190 --> 00:39:07.110 engineering problems. Which changes everything. 00:39:07.809 --> 00:39:10.150 The agents imported the molecular structure reported 00:39:10.150 --> 00:39:12.690 by Holmes and Girolami into their simulation 00:39:12.690 --> 00:39:15.949 lane and ran massive modern quantum calculations 00:39:15.949 --> 00:39:18.909 on it. Those simulations yielded a staggering 00:39:18.909 --> 00:39:23.449 prediction. This useless 1999 compound inherently 00:39:23.449 --> 00:39:26.869 behaves exactly as a Luttinger compensated semiconductor. 00:39:27.670 --> 00:39:30.809 They found the unicorn. The simulations predicted 00:39:30.809 --> 00:39:33.570 the material possesses a band gap of approximately 00:39:33.570 --> 00:39:36.550 2 .1 electron volts. I want to break down band 00:39:36.550 --> 00:39:38.849 gap really quickly for anyone who isn't a solid 00:39:38.849 --> 00:39:40.989 state physicist because it's vital. Please do. 00:39:41.449 --> 00:39:45.230 Think of a band gap like a literal physical ditch 00:39:45.230 --> 00:39:48.929 separating two platforms. One platform is where 00:39:48.929 --> 00:39:51.710 electrons are stuck in place. The other platform 00:39:51.710 --> 00:39:54.349 is where electrons can flow freely and conduct 00:39:54.349 --> 00:39:57.769 electricity. The band gap is the energy ditch 00:39:57.769 --> 00:40:00.210 they have to jump over to become conductive. 00:40:00.230 --> 00:40:03.130 That's a great visual. A gap of 2 .1 electron 00:40:03.130 --> 00:40:06.030 volts means it's a solid, well -defined ditch. 00:40:06.190 --> 00:40:08.869 It won't conduct electricity accidentally, but 00:40:08.869 --> 00:40:10.989 you can force it to conduct when you want to 00:40:10.989 --> 00:40:13.690 by applying the right amount of energy. It makes 00:40:13.690 --> 00:40:16.929 it a highly controllable semiconductor. Exactly 00:40:16.929 --> 00:40:19.710 the mechanism. Furthermore... The AI predicted 00:40:19.710 --> 00:40:22.130 that both the top edge of that bandgap and

the 00:40:22.130 --> 00:40:24.929 bottom edge carry the exact same electron spin. 00:40:25.469 --> 00:40:28.489 This provides the coveted data sorting capability. 00:40:28.929 --> 00:40:31.429 The spin sorting. And the predicted stability 00:40:31.429 --> 00:40:33.670 of this spin sorting is what truly shocked the 00:40:33.670 --> 00:40:36.670 researchers. The AI calculated that this spin 00:40:36.670 --> 00:40:38.869 sorting remains stable over energy windows of 00:40:38.869 --> 00:40:42.969 about 2 .6 electron volts for the holes, the 00:40:42.969 --> 00:40:45.829 absence of an electron, and 1 .6 electron volts 00:40:45.829 --> 00:40:48.360 for the electrons themselves. Throwing around 00:40:48.360 --> 00:40:51.239 measurements like 2 .6 and 1 .6 electron volts 00:40:51.239 --> 00:40:54.179 doesn't mean much without context. The source 00:40:54.179 --> 00:40:56.400 text gives us the perfect baseline for comparison. 00:40:56.780 --> 00:40:59.239 You have to compare those numbers to ambient 00:40:59.239 --> 00:41:03.179 thermal energy. The heat of a room. Right. Ambient 00:41:03.179 --> 00:41:06.320 thermal energy is just the chaotic, random, microscopic 00:41:06.320 --> 00:41:09.139 jiggling of atoms that constantly happens simply 00:41:09.139 --> 00:41:11.980 because a room is warm. At standard room temperature, 00:41:12.159 --> 00:41:14.900 that random thermal noise amounts to only about 00:41:14.900 --> 00:41:19.840 26... thousandths of an electron volt 0 .026 00:41:19.840 --> 00:41:22.699 ev so you have a thermal disruption force of 00:41:22.699 --> 00:41:27.019 0 .026 against a structural stability of 2 .6 00:41:27.019 --> 00:41:30.460 if the thermal noise of a warm room is a slight 00:41:30.460 --> 00:41:33.260 annoying breeze the spin stability of this material 00:41:33.260 --> 00:41:36.780 is a solid brick wall the data locked into those 00:41:36.780 --> 00:41:39.599 electron spins is massively overwhelmingly protected 00:41:39.599 --> 00:41:41.579 against the random heat of the environment the 00:41:41.579 --> 00:41:43.840 memory won't spontaneously scramble itself on 00:41:43.840 --> 00:41:46.940 a hot summer day it is incredibly robust. And 00:41:46.940 --> 00:41:49.000 if you recall the catastrophic failure of the 00:41:49.000 --> 00:41:51.500 AI's first candidate, the novel oxide that melted 00:41:51.500 --> 00:41:55.440 at 950 Kelvin, Val's AI noted that this 1999 00:41:55.440 --> 00:41:57.780 compound doesn't suffer from that structural 00:41:57.780 --> 00:42:01.039 weakness. Because it's chemically locked. Yes, 00:42:01.099 --> 00:42:03.659 its chemical nature inherently locks every metal 00:42:03.659 --> 00:42:07.320 atom into a specific rigid site. The chromium 00:42:07.320 --> 00:42:09.219 atoms are chemically forced to bond exclusively 00:42:09.219 --> 00:42:11.960 to the carbon into the cyanide molecules. The 00:42:11.960 --> 00:42:14.019 vanadium atoms are forced to bond exclusively 00:42:14.019 --> 00:42:17.000 to the nitrogen net. So it forces itself into 00:42:17.000 --> 00:42:19.360 the perfect shape. It naturally possesses the 00:42:19.360 --> 00:42:22.139 exact, unyielding structural discipline that 00:42:22.139 --> 00:42:25.000 the AI's custom -designed oxide lacked. It is 00:42:25.000 --> 00:42:28.159 structurally idiot -proof. When Val's AI published 00:42:28.159 --> 00:42:30.119 this summary, they used a phrase to describe 00:42:30.119 --> 00:42:32.699 the material that Don Swanson would have absolutely 00:42:32.699 --> 00:42:36.329 loved. They said the compound had been... hiding 00:42:36.329 --> 00:42:39.210 in plain sight. It serves as the ultimate modern 00:42:39.210 --> 00:42:41.309 validation of undiscovered public knowledge. 00:42:41.610 --> 00:42:44.610 I really want you, the listener, to pause and 00:42:44.610 --> 00:42:47.050 let the weight of this sink in. The potential 00:42:47.050 --> 00:42:50.710 foundational architecture for a multi -trillion 00:42:50.710 --> 00:42:53.090 dollar leap in global computing was just sitting 00:42:53.090 --> 00:42:56.409 there, compressed in a PDF, on a server, labeled 00:42:56.409 --> 00:42:59.289 by its own creators as having no real world use. 00:42:59.610 --> 00:43:02.869 Waiting for someone to connect the dots. It was 00:43:02.869 --> 00:43:04.989 just waiting for something with enough patience 00:43:04.989 --> 00:43:08.869 to ask, the right question. However, as seductive 00:43:08.869 --> 00:43:11.610 as this narrative is, and as flawless as the 00:43:11.610 --> 00:43:14.969 AI simulations make it appear, we must aggressively 00:43:14.969 --> 00:43:19.469 pump the brakes. We must. Because if Don Swanson's 00:43:19.469 --> 00:43:22.349 legacy teaches us anything, it is that the map 00:43:22.349 --> 00:43:25.570 is never the territory. A paper is not a physical 00:43:25.570 --> 00:43:28.789 reality. And a digital simulation, even one run 00:43:28.789 --> 00:43:31.309 concurrently by 90 advanced Claude Opus agents, 00:43:31.510 --> 00:43:35.320 is still merely a simulation. Which brings us 00:43:35.320 --> 00:43:38.280 to the most vital question of this new era. How 00:43:38.280 --> 00:43:41.579 do we verify? How do we know the AI isn't hallucinating 00:43:41.579 --> 00:43:44.039 nuanced physics to please the human researchers? 00:43:44.320 --> 00:43:46.960 We touched on this earlier, but when the claims 00:43:46.960 --> 00:43:49.780 get this specific, the standard of proof has 00:43:49.780 --> 00:43:52.559 to skyrocket. This is where the source text heavily 00:43:52.559 --> 00:43:55.900 emphasizes the concept of the ledger. This raises 00:43:55.900 --> 00:43:58.440 an important question. How do we establish concrete 00:43:58.440 --> 00:44:01.159 trust in an era increasingly dominated by AI 00:44:01.159 --> 00:44:03.659 -generated science? It's the multi -billion dollar 00:44:03.659 --> 00:44:05.739 question. The answer doesn't lie in building 00:44:05.739 --> 00:44:09.179 better AI. It lies in relentless

human transparency 00:44:09.179 --> 00:44:11.500 and the rigorous acknowledgement of caveats. 00:44:12.099 --> 00:44:14.840 Val's AI did not claim they had built a working 00:44:14.840 --> 00:44:18.199 quantum drive. They attached massive, glaring 00:44:18.199 --> 00:44:21.139 caveats to their publication. They were refreshingly 00:44:21.139 --> 00:44:23.670 upfront. They stated explicitly across their 00:44:23.670 --> 00:44:26.090 social media and technical publications that 00:44:26.090 --> 00:44:29.449 every single claim regarding the 2 .1 eV bandgap 00:44:29.449 --> 00:44:32.730 and the spin sorting derived entirely from computational 00:44:32.730 --> 00:44:35.130 modeling. No one has actually built it yet to 00:44:35.130 --> 00:44:37.789 test these specific properties. Right. No human 00:44:37.789 --> 00:44:39.889 being has ever placed this material in a modern 00:44:39.889 --> 00:44:42.150 lab and physically measured its spin sorting. 00:44:42.309 --> 00:44:44.670 They wrote point blank that these numbers are, 00:44:44.730 --> 00:44:48.570 quote, predictions for perfect crystals. And 00:44:48.570 --> 00:44:51.409 as any chemist knows, nature vehemently despises 00:44:51.409 --> 00:44:54.369 a perfect crystal. The caveats detailed in the 00:44:54.369 --> 00:44:57.590 text go much deeper. The only physical, synthesized 00:44:57.590 --> 00:44:59.730 sample of this specific material on the public 00:44:59.730 --> 00:45:02.610 record is the original powder created by Holmes 00:45:02.610 --> 00:45:05.570 and Giralami back in 1999. And that physical 00:45:05.570 --> 00:45:08.889 powder was far from perfect. Exactly. It was 00:45:08.889 --> 00:45:11.389 porous, and water molecules from the ambient 00:45:11.389 --> 00:45:13.969 air had become permanently trapped inside its 00:45:13.969 --> 00:45:16.730 microscopic pores. And water, chemically speaking, 00:45:16.889 --> 00:45:19.389 tends to ruin elegant mathematical models. It 00:45:19.389 --> 00:45:21.769 absolutely does. Because of those trapped water 00:45:21.769 --> 00:45:25.070 molecules and microscopic defects, the 1999 physical 00:45:25.070 --> 00:45:28.690 sample did not exhibit the perfect zero net magnetism 00:45:28.690 --> 00:45:31.429 required by the Ludinger theory. It actually 00:45:31.429 --> 00:45:34.750 showed a small, lingering magnetic moment, specifically 00:45:34.750 --> 00:45:40.030 measured at about 0 .125 Bohr magnetons per formulae 00:45:40.030 --> 00:45:43.300 unit. A Bohr magneton, for those keeping score, 00:45:43.400 --> 00:45:46.059 is basically the absolute smallest fundamental 00:45:46.059 --> 00:45:48.940 unit of magnetism. You can think of it like a 00:45:48.940 --> 00:45:51.739 single pixel on a screen or a single drop of 00:45:51.739 --> 00:45:54.320 water in an ocean of magnetic force. That's a 00:45:54.320 --> 00:45:56.440 good way to visualize it. The Holy Grail material 00:45:56.440 --> 00:46:00.119 is supposed to have exactly zero drops. The 1999 00:46:00.119

--> 00:46:03.280 powder had about one -eighth of a drop per unit. 00:46:03.340 --> 00:46:06.960 It's tiny, but in quantum computing, tiny imperfections 00:46:06.960 --> 00:46:10.480 are massive roadblocks. So immediately... The 00:46:10.480 --> 00:46:13.900 messy physical reality of 1999 clashes slightly 00:46:13.900 --> 00:46:17.679 with the pristine digital theory of 2026. But 00:46:17.679 --> 00:46:19.860 the friction gets even more complex, because 00:46:19.860 --> 00:46:22.719 the AI swarm itself is internally divided. Oh, 00:46:22.760 --> 00:46:25.599 right, the two models. Val's AI didn't just run 00:46:25.599 --> 00:46:28.300 one monolithic simulation. They ran the compound 00:46:28.300 --> 00:46:30.239 through two completely different levels of a 00:46:30.239 --> 00:46:32.780 quantum mechanical modeling framework called 00:46:32.780 --> 00:46:36.900 density functional theory, or DFT. Let's demystify 00:46:36.900 --> 00:46:39.360 DFT really quickly. Density functional theory 00:46:39.360 --> 00:46:42.099 is basically a mathematical way to guess how 00:46:42.099 --> 00:46:44.659 all the electrons in a material are interacting 00:46:44.659 --> 00:46:48.179 with each other. Simulating every single electron 00:46:48.179 --> 00:46:50.659 perfectly is mathematically impossible, even 00:46:50.659 --> 00:46:54.900 for supercomputers. So DFT uses complex approximations. 00:46:54.960 --> 00:46:58.739 The AI ran two different styles of these approximations. 00:46:59.000 --> 00:47:02.320 Exactly. They utilize a faster, older model called 00:47:02.320 --> 00:47:06.119 PBE plus U. which relies on rougher approximations, 00:47:06.179 --> 00:47:09.579 specifically using a Harvard U correction to 00:47:09.579 --> 00:47:11.579 guess how certain stubborn electrons behave. 00:47:11.860 --> 00:47:15.460 It's fast, but it's a sketch. And the other one? 00:47:15.980 --> 00:47:18.340 Dent. They ran the material through a much slower, 00:47:18.519 --> 00:47:21.360 computationally agonizing, and generally more 00:47:21.360 --> 00:47:25.739 accurate model called HSE06, which is a hybrid 00:47:25.739 --> 00:47:28.119 functional that calculates exact electron exchanges. 00:47:28.639 --> 00:47:31.059 And the critical detail here is that these two 00:47:31.059 --> 00:47:33.809 advanced models fundamentally disagree on the 00:47:33.809 --> 00:47:37.769 physics of the trapped water. So if I'm tracking 00:47:37.769 --> 00:47:40.070 this correctly, the AI is effectively arguing 00:47:40.070 --> 00:47:42.670 with itself. You've got this fast, loose model 00:47:42.670 --> 00:47:46.210 and a slow, meticulous model, and they are essentially 00:47:46.210 --> 00:47:48.510 locked in a debate over whether a microscopic 00:47:48.510 --> 00:47:51.550 drop of trapped water ruins the entire recipe. 00:47:51.750 --> 00:47:54.469 That is precisely the conflict. The slower, more 00:47:54.469 --> 00:47:58.059 meticulous... HSE06 model calculates that even 00:47:58.059 --> 00:48:01.179 with the water trapped in the pores, the crucial 00:48:01.179 --> 00:48:04.679 spin sorting survives. The core property of the 00:48:04.679 --> 00:48:07.239 Holy Grail holds up under pressure. Okay, so 00:48:07.239 --> 00:48:10.599 HSE06 says it works. However, the faster PBE 00:48:10.599 --> 00:48:13.599 plus U model calculates that the water acts as 00:48:13.599 --> 00:48:16.400 a massive chemical disruptor, shrinking that 00:48:16.400 --> 00:48:19.860 protective hole window by more than half, which 00:48:19.860 --> 00:48:22.360 would potentially ruin its usefulness as stable 00:48:22.360 --> 00:48:25.000 memory. The swarm isn't delivering a verdict. 00:48:25.059 --> 00:48:27.719 It's delivering a hung jury based on the presence 00:48:27.719 --> 00:48:30.440 of microscopic humidity. And I want to point 00:48:30.440 --> 00:48:32.659 out to anyone listening that this is a phenomenal 00:48:32.659 --> 00:48:36.010 outcome. It really is. The AI didn't... quietly 00:48:36.010 --> 00:48:38.489 hide this disagreement to secure venture capital 00:48:38.489 --> 00:48:42.070 funding, Val's AI loudly published the internal 00:48:42.070 --> 00:48:44.610 conflict. They stated that the simulation has 00:48:44.610 --> 00:48:47.190 reached its absolute limit and the only necessary 00:48:47.190 --> 00:48:49.710 next step is for physical humans to go into a 00:48:49.710 --> 00:48:52.630 physical lab, synthesize the compound from scratch 00:48:52.630 --> 00:48:54.969 in a vacuum, and physically measure it to see 00:48:54.969 --> 00:48:57.750 which AI model is telling the truth. This is 00:48:57.750 --> 00:49:00.789 the triumph of the Val's AI experiment. They 00:49:00.789 --> 00:49:03.469 maintained what the source text beautifully terms 00:49:04.019 --> 00:49:07.380 A ledger of honesty. They didn't just issue a 00:49:07.380 --> 00:49:09.480 breathless press release claiming they solved 00:49:09.480 --> 00:49:12.039 the future of computing. They published every 00:49:12.039 --> 00:49:14.199 scrap of their work on a public GitHub repository. 00:49:14.619 --> 00:49:17.219 Complete transparency. They uploaded the input 00:49:17.219 --> 00:49:20.900 files, the raw, unedited outputs from the simulations, 00:49:21.159 --> 00:49:24.440 and the analytical code. They even built and 00:49:24.440 --> 00:49:27.159 provided a one -command checker script so that 00:49:27.159 --> 00:49:29.719 any skeptical scientist anywhere in the world 00:49:29.719 --> 00:49:31.940 could run the exact same simulations on their 00:49:31.940 --> 00:49:34.619 own hardware. They published the independent 00:49:34.619 --> 00:49:37.420 reruns and the exhaustive list of known physical 00:49:37.420 --> 00:49:40.179 caveats. They made their digital homework entirely 00:49:40.179 --> 00:49:43.820 public and instantly verifiable. In this new 00:49:43.820 --> 00:49:46.300 era, that public ledger matters just as much, 00:49:46.340 --> 00:49:49.099 if not more, than the discovery of the candidate

00:49:49.099 --> 00:49:52.719 material itself. I cannot emphasize enough why 00:49:52.719 --> 00:49:55.360 this operational transparency is so critical. 00:49:56.000 --> 00:49:58.599 As we integrate AI into scientific research, 00:49:58.760 --> 00:50:01.539 the goal is not to cultivate... blind trust in 00:50:01.539 --> 00:50:03.900 a tech startup or a language model, it's the 00:50:03.900 --> 00:50:06.260 exact opposite. We should be skeptical. The goal 00:50:06.260 --> 00:50:09.079 is to make the AI's claims brutally testable 00:50:09.079 --> 00:50:11.519 by third parties who owe the original researchers 00:50:11.519 --> 00:50:15.039 nothing. You must actively provide the tools 00:50:15.039 --> 00:50:19.239 for your own debunking. An AI swarm can propose 00:50:19.239 --> 00:50:22.880 the most brilliant, poetic bridge between a 1999 00:50:22.880 --> 00:50:26.500 paper and a 2026 problem. But the real world? 00:50:26.840 --> 00:50:29.619 The physical laboratory with its messy humidity, 00:50:29.960 --> 00:50:32.719 imperfect temperatures, and stubborn atoms always 00:50:32.719 --> 00:50:35.940 gets the final say. That encapsulates the new 00:50:35.940 --> 00:50:39.179 paradigm of discovery. We have structurally shifted 00:50:39.179 --> 00:50:41.639 the burden of science from the extreme difficulty 00:50:41.639 --> 00:50:44.679 of finding the hypothesis to the immense rigor 00:50:44.679 --> 00:50:47.840 required to test the hypothesis. We've examined 00:50:47.840 --> 00:50:51.099 how a singular genius librarian executed this 00:50:51.099 --> 00:50:54.019 methodology with index cards in Chicago in 1986. 00:50:55.260 --> 00:50:57.500 We've dissected how a cutting -edge startup executed 00:50:57.500 --> 00:51:00.639 it with a massive cloud computing's form in 2026. 00:51:01.630 --> 00:51:04.190 But the most empowering, actionable part of this 00:51:04.190 --> 00:51:07.769 entire deep dive is that the source text doesn't 00:51:07.769 --> 00:51:10.210 leave us as passive spectators of history. It 00:51:10.210 --> 00:51:13.050 provides a literal step -by -step blueprint. 00:51:13.429 --> 00:51:16.769 It democratizes the process entirely. The text 00:51:16.769 --> 00:51:18.829 lays out a comprehensive nine -step workflow 00:51:18.829 --> 00:51:21.909 to run a scaled -down, highly -focused version 00:51:21.909 --> 00:51:24.880 of Don Swanson's method. And it relies entirely 00:51:24.880 --> 00:51:28.699 on free public scientific databases and the commercial 00:51:28.699 --> 00:51:30.980 AI assistance that you likely already have installed 00:51:30.980 --> 00:51:33.820 on your phone or laptop. So what does this all 00:51:33.820 --> 00:51:37.239 mean? It means the tools of world -changing discovery 00:51:37.239 --> 00:51:40.099 are no longer exclusively locked behind the ivy 00:51:40.099 --> 00:51:42.039 -covered gates of elite research institutions. 00:51:42.579 --> 00:51:44.800 You, the person listening to this audio right 00:51:44.800 --> 00:51:47.179 now, possess the digital infrastructure to be 00:51:47.179 --> 00:51:49.630 a pioneer in your own field. You don't need a 00:51:49.630 --> 00:51:52.030 PhD to start connecting the dots. You don't need 00:51:52.030 --> 00:51:54.309 to know how to write Python code. You just need 00:51:54.309 --> 00:51:56.449 patience, domain knowledge, and a willingness 00:51:56.449 --> 00:51:58.989 to rigorously doubt your own exciting results. 00:51:59.769 --> 00:52:02.150 Let's walk through this DIY guide step by step, 00:52:02.210 --> 00:52:04.329 and let's actually apply it to a hypothetical 00:52:04.329 --> 00:52:06.889 scenario so you can see how it functions in reality. 00:52:07.639 --> 00:52:10.679 Step one is the setup. You must choose a highly 00:52:10.679 --> 00:52:14.760 specific C. Remember the ABC model. You have 00:52:14.760 --> 00:52:17.440 to start with your target. the end point, the 00:52:17.440 --> 00:52:20.340 problem you are trying to solve. And the key 00:52:20.340 --> 00:52:24.139 word is specific. You do not want your C term 00:52:24.139 --> 00:52:27.900 to be cancer or global warming. Those are way 00:52:27.900 --> 00:52:30.480 too broad. You will drown in millions of papers. 00:52:30.780 --> 00:52:33.519 The text suggests targeting a specific condition, 00:52:33.860 --> 00:52:36.780 a niche material failure in a machine you maintain, 00:52:37.059 --> 00:52:40.460 or a specific agricultural problem. Let's create 00:52:40.460 --> 00:52:42.460 a hypothetical scenario for this walkthrough. 00:52:42.940 --> 00:52:46.670 Let's say you are an arborist. And your C term 00:52:46.670 --> 00:52:50.929 is urban oak tree blight, a specific fungal infection 00:52:50.929 --> 00:52:55.090 destroying local trees. That is a perfectly scoped 00:52:55.090 --> 00:52:58.349 target. Once you have your specific C, step two 00:52:58.349 --> 00:53:01.010 is gathering the relevant literature. You navigate 00:53:01.010 --> 00:53:03.909 to a free public scientific database that suits 00:53:03.909 --> 00:53:06.650 your discipline. If your problem is medical or 00:53:06.650 --> 00:53:09.869 biological, you use PubMed. If you are hunting 00:53:09.869 --> 00:53:12.429 in physics, math, or computer science, you use 00:53:12.429 --> 00:53:15.840 Archive. If your topic is anything else, like 00:53:15.840 --> 00:53:18.440 our hypothetical oak tree blight, you can use 00:53:18.440 --> 00:53:21.199 OpenAlex, which is a phenomenal open -source 00:53:21.199 --> 00:53:24.320 repository indexing nearly every academic discipline 00:53:24.320 --> 00:53:27.059 on Earth. You query the database for your C -term 00:53:27.059 --> 00:53:29.780 and collect the recent abstracts. You don't need 00:53:29.780 --> 00:53:32.260 to pay for the full text right now. The abstracts 00:53:32.260 --> 00:53:35.260 contain the core claims. Gather a few dozen to 00:53:35.260 --> 00:53:37.619 a few hundred of them and compile them into a 00:53:37.619 --> 00:53:40.480 single text document. Now we hit step

three. 00:53:40.909 --> 00:53:43.530 This is where you leverage your everyday AI assistant. 00:53:43.809 --> 00:53:46.269 You upload that massive document of abstracts 00:53:46.269 --> 00:53:49.809 to Claude or ChatGPT, and you ask the AI to extract 00:53:49.809 --> 00:53:53.489 the B terms. You construct a prompt instructing 00:53:53.489 --> 00:53:56.090 the AI to read the abstracts and generate a comprehensive 00:53:56.090 --> 00:53:58.849 list of all the biological mechanisms, intermediate 00:53:58.849 --> 00:54:01.570 chemical properties, or environmental factors 00:54:01.570 --> 00:54:04.150 that the literature explicitly states make your 00:54:04.150 --> 00:54:08.070 target C better or worse. In Swanson's world, 00:54:08.150 --> 00:54:10.809 this was finding that blood viscosity made Raynaud's 00:54:10.809 --> 00:54:13.730 worse. In our hypothetical oak tree scenario, 00:54:14.070 --> 00:54:16.269 the AI might read the abstracts and tell you 00:54:16.269 --> 00:54:19.190 that urban oak tree blight is significantly worsened 00:54:19.190 --> 00:54:22.190 by a specific enzyme in the soil. Let's call 00:54:22.190 --> 00:54:25.369 it enzyme X, which degrades the tree's root walls. 00:54:25.730 --> 00:54:29.469 So enzyme X is your B term. And crucially, you 00:54:29.469 --> 00:54:32.250 must command the AI to tie every single extracted 00:54:32.250 --> 00:54:35.460 B term. back to the specific abstract that supports 00:54:35.460 --> 00:54:38.900 it. You must maintain a paper trail. Now you 00:54:38.900 --> 00:54:41.699 possess a list of underlying mechanisms. Step 00:54:41.699 --> 00:54:44.280 four is where the true hunt begins. You take 00:54:44.280 --> 00:54:46.820 each of those promising B terms, like enzyme 00:54:46.820 --> 00:54:50.559 X, and you search the database outward. You initiate 00:54:50.559 --> 00:54:53.599 a brand new search query. You look for substances, 00:54:53.920 --> 00:54:56.340 natural microbes, chemical compounds, or agricultural 00:54:56.340 --> 00:54:59.199 practices that are known to inhibit or destroy 00:54:59.199 --> 00:55:03.300 enzyme X. But, and this is the absolute most 00:55:03.300 --> 00:55:05.599 critical rule of the entire process. You must 00:55:05.599 --> 00:55:08.820 explicitly use Boolean operators to exclude any 00:55:08.820 --> 00:55:12.139 paper that mentions your original C term, urban 00:55:12.139 --> 00:55:14.860 oak tree blight. You are actively filtering out 00:55:14.860 --> 00:55:17.739 what humanity already knows. You are specifically 00:55:17.739 --> 00:55:20.440 hunting for the A terms, the solutions that affect 00:55:20.440 --> 00:55:23.719 B, but have never once been formally associated 00:55:23.719 --> 00:55:26.360 with C. Let's say your outward search discovers 00:55:26.360 --> 00:55:29.179 that a very specific strain of common agricultural 00:55:29.179 --> 00:55:32.519 moss, your A -term, naturally neutralizes enzyme 00:55:32.519 --> 00:55:36.199 X. Step five is checking your work, checking 00:55:36.199 --> 00:55:39.380 for that Swanson disjointness. You run one final 00:55:39.380 --> 00:55:42.840 combined search for the moss and the oak tree 00:55:42.840 --> 00:55:45.980 blight. If a dozen papers pop up, throw it in 00:55:45.980 --> 00:55:48.599 the trash. The link is already known. Someone 00:55:48.599 --> 00:55:51.719 is already working on it. But if that combined 00:55:51.719 --> 00:55:55.679 search yields zero results, That is the silence. 00:55:55.820 --> 00:55:58.880 That is the X on the map. You have successfully 00:55:58.880 --> 00:56:01.219 located a gap in human knowledge worth examining. 00:56:01.760 --> 00:56:04.659 So you've found a hidden link. You think planting 00:56:04.659 --> 00:56:07.739 this common moss around the oak trees will cure 00:56:07.739 --> 00:56:10.320 the blight. Steps six and seven are where you 00:56:10.320 --> 00:56:12.659 must force yourself to become the ultimate skeptic. 00:56:12.659 --> 00:56:16.039 You do not celebrate yet. Step six, assign a 00:56:16.039 --> 00:56:19.139 skeptic. You open a completely fresh blank session 00:56:19.139 --> 00:56:22.119 with your AI assistant. Or, ideally, you switch 00:56:22.119 --> 00:56:24.099 to a completely different AI model from a different 00:56:24.099 --> 00:56:27.159 company to avoid bias. You feed it your proposed 00:56:27.159 --> 00:56:30.719 A to C link. The moss cures the blight via enzyme 00:56:30.719 --> 00:56:33.960 X. And you explicitly instruct the model to act 00:56:33.960 --> 00:56:37.139 as a hostile peer reviewer. You command it to 00:56:37.139 --> 00:56:40.079 argue against you. You tell it to tear your beautiful 00:56:40.079 --> 00:56:43.699 theory to shreds. Have it search for contradicting 00:56:43.699 --> 00:56:46.199 evidence in the literature. Have it look for 00:56:46.199 --> 00:56:49.199 scale problems. Maybe you'd need 10 tons of moss 00:56:49.199 --> 00:56:52.579 to save one tree. Have it look for cascading 00:56:52.579 --> 00:56:55.719 ecological failures. You only retain the hypotheses 00:56:55.719 --> 00:56:58.539 that survive this brutal adversarial digital 00:56:58.539 --> 00:57:01.960 cross -examination. And then comes step seven, 00:57:02.039 --> 00:57:04.639 which requires the absolute highest level of 00:57:04.639 --> 00:57:07.579 human intellectual discipline. You must verify 00:57:07.579 --> 00:57:10.780 every single citation by hand. Because, as we 00:57:10.780 --> 00:57:13.960 discussed with the AI models earlier, they hallucinate. 00:57:14.139 --> 00:57:16.960 An AI will confidently invent a paper proving 00:57:16.960 --> 00:57:20.170 your theory is flawless, complete with fake authors, 00:57:20.329 --> 00:57:22.650 a fake university, and a fake digital object 00:57:22.650 --> 00:57:25.250 identifier. You have to physically click the 00:57:25.250 --> 00:57:27.789 links, open the original source documents, read 00:57:27.789 --> 00:57:29.969 the human words yourself, and confirm that the 00:57:29.969 -->

00:57:32.550 paper actually exists and says exactly what the 00:57:32.550 --> 00:57:35.070 AI claims it says. If the citations physically 00:57:35.070 --> 00:57:38.599 check out, you progress to step eight. Keep a 00:57:38.599 --> 00:57:41.780 ledger. Mirror exactly what Val's AI did with 00:57:41.780 --> 00:57:44.809 their public GitHub repository. For every link 00:57:44.809 --> 00:57:47.489 that survives the gauntlet, meticulously document 00:57:47.489 --> 00:57:50.400 the claim. Record the specific papers supporting 00:57:50.400 --> 00:57:54.400 A to B and the papers supporting B to C. Document 00:57:54.400 --> 00:57:56.699 the results of your disjointness search, proving 00:57:56.699 --> 00:57:59.820 the silence. Record the skeptic AI's objections 00:57:59.820 --> 00:58:03.079 and state the current untested status of the 00:58:03.079 --> 00:58:05.760 hypothesis. This ledger is the dividing line 00:58:05.760 --> 00:58:08.519 between a crazy late -night internet rabbit hole 00:58:08.519 --> 00:58:11.460 and a viable structured scientific hypothesis 00:58:11.460 --> 00:58:14.820 that other professionals can actually test. And 00:58:14.820 --> 00:58:17.340 finally, we arrive at step nine, the handoff. 00:58:17.840 --> 00:58:20.860 You must treat every single result you generate 00:58:20.860 --> 00:58:23.800 through this workflow as a hypothesis. It is 00:58:23.800 --> 00:58:26.760 not a fact. If you were exploring a human health 00:58:26.760 --> 00:58:29.400 question, you take your meticulously cited ledger 00:58:29.400 --> 00:58:31.539 directly to a qualified medical professional 00:58:31.539 --> 00:58:35.070 or researcher. The source text is fiercely firm 00:58:35.070 --> 00:58:38.230 on this point, and we are too. Under no circumstances 00:58:38.230 --> 00:58:41.090 should you ever act on medical advice you derived 00:58:41.090 --> 00:58:44.380 yourself using this method. You are utilizing 00:58:44.380 --> 00:58:46.679 AI to find better questions, not to prescribe 00:58:46.679 --> 00:58:48.929 final answers. If it's a materials engineering 00:58:48.929 --> 00:58:51.690 question, you take the ledger to a metallurgist. 00:58:51.730 --> 00:58:54.090 If it's our hypothetical tree blight, you hand 00:58:54.090 --> 00:58:55.989 the ledger to a university botany department 00:58:55.989 --> 00:58:58.869 equipped to run the physical soil tests. As the 00:58:58.869 --> 00:59:01.869 text wisely observes, if you integrate this workflow 00:59:01.869 --> 00:59:04.170 into your professional life, you will inevitably 00:59:04.170 --> 00:59:06.949 experience exactly what Don Swanson felt in his 00:59:06.949 --> 00:59:10.599 office in 1986. 99 % of your candidate links 00:59:10.599 --> 00:59:13.239 are going to spectacularly collapse under scrutiny. 00:59:13.420 --> 00:59:15.840 They will fail the adversarial skeptic test, 00:59:16.019 --> 00:59:18.519 or you will manually dig up a scanned paper from 00:59:18.519 --> 00:59:21.539 1972 that already completely disproved the chemical 00:59:21.539 --> 00:59:25.179 pathway. But the raw, intellectual feeling of 00:59:25.179 --> 00:59:28.139 finding the one, single link that survives the 00:59:28.139 --> 00:59:32.159 entire gauntlet? That is intoxicating. Because 00:59:32.159 --> 00:59:34.239 in that moment, you realize a world -changing 00:59:34.239 --> 00:59:36.860 answer was sitting in plain public view the entire 00:59:36.860 --> 00:59:39.789 time. and you are the first human being to finally 00:59:39.789 --> 00:59:43.260 see it. It forces a profound redefinition of 00:59:43.260 --> 00:59:46.639 what a library actually is. A library, or a digital 00:59:46.639 --> 00:59:49.559 database, is not merely a static archive of what 00:59:49.559 --> 00:59:53.000 humanity has achieved. It is a massive, mathematically 00:59:53.000 --> 00:59:56.519 dense, latent map of everything we haven't yet 00:59:56.519 --> 00:59:59.809 realized we already know. Let's bring this entire 00:59:59.809 --> 01:00:03.119 journey full circle. We are standing on the precipice 01:00:03.119 --> 01:00:05.480 of a fundamental shift in the velocity of human 01:00:05.480 --> 01:00:08.079 progress. The near future of scientific research 01:00:08.079 --> 01:00:10.920 looks exactly like Don Swanson's analog library, 01:00:11.179 --> 01:00:13.739 but with the lights left blazing all night long. 01:00:14.059 --> 01:00:16.719 We are going to witness relentless, decentralized 01:00:16.719 --> 01:00:20.000 teams of AI agents strip mining centuries of 01:00:20.000 --> 01:00:22.579 forgotten journals, abandoned patents, and dusty 01:00:22.579 --> 01:00:25.460 technical reports across every conceivable field 01:00:25.460 --> 01:00:28.440 of human endeavor. We are undoubtedly going to 01:00:28.440 --> 01:00:31.000 discover that countless existing drugs, synthetic 01:00:31.000 --> 01:00:33.519 compounds, and discarded manufacturing methods 01:00:33.519 --> 01:00:35.960 possess secondary, highly valuable properties 01:00:35.960 --> 01:00:38.760 that their original human creators never possessed 01:00:38.760 --> 01:00:42.300 the context to check. The chemists in 1999 looked 01:00:42.300 --> 01:00:45.179 at their beautiful high -temperature magnet and 01:00:45.179 --> 01:00:47.500 logically concluded it had no real -world use. 01:00:47.940 --> 01:00:51.579 27 years later, a team of machines synthesized 01:00:51.579 --> 01:00:54.159 their data and proposed that it might run the 01:00:54.159 --> 01:00:56.360 next generation of global computing. pending 01:00:56.360 --> 01:00:59.099 physical measurement. But the ultimate winners 01:00:59.099 --> 01:01:01.659 in this rapidly approaching era won't simply 01:01:01.659 --> 01:01:04.219 be the people who can generate the most AI hypotheses. 01:01:04.860 --> 01:01:07.360 The software tools to find these hidden links 01:01:07.360 --> 01:01:10.059 are becoming freely available to everyone all 01:01:10.059 --> 01:01:12.980 at once. The true winners will be the individuals 01:01:12.980 --> 01:01:16.039 and organizations who cultivate the immense, 01:01:16.280 --> 01:01:20.900 unglamorous discipline required to verify. The 01:01:20.900 --> 01:01:22.860 pioneers will be the ones who demand the public 01:01:22.860 --> 01:01:26.360 ledger, who rerun the adversarial checker, who 01:01:26.360 --> 01:01:28.719 meticulously read the original human papers, 01:01:28.880 --> 01:01:31.360 and who possess the patience to wait for the 01:01:31.360 --> 01:01:33.880 results from the physical laboratory before they 01:01:33.880 --> 01:01:37.139 declare victory. AI can exponentially accelerate 01:01:37.139 --> 01:01:39.739 the reading, but the rigorous discipline required 01:01:39.739 --> 01:01:43.519 to test reality remains a profoundly, exclusively 01:01:43.519 --> 01:01:47.360 human choice. Don Swanson spent the entirety 01:01:47.820 --> 01:01:49.880 of his academic career arguing that the next 01:01:49.880 --> 01:01:52.579 great discoveries were already sitting in print. 01:01:52.820 --> 01:01:56.579 This week, a digital swarm offered fresh, staggering, 01:01:56.820 --> 01:01:59.179 multi -million dollar support for his claim. 01:01:59.320 --> 01:02:02.579 Which leaves us with one final lingering question 01:02:02.579 --> 01:02:04.860 for you to ponder long after you take your headphones 01:02:04.860 --> 01:02:08.219 off today. If our most advanced supercomputers 01:02:08.219 --> 01:02:10.480 and our most cutting -edge AI architectures are 01:02:10.480 --> 01:02:13.260 currently dedicating massive amounts of electricity 01:02:13.260 --> 01:02:16.070 and compute power, Just to realize what a couple 01:02:16.070 --> 01:02:17.969 of human chemists already wrote down on paper 01:02:17.969 --> 01:02:21.789 27 years ago, what other world -changing miracles 01:02:21.789 --> 01:02:23.889 are gathering dust on a shelf right near you, 01:02:23.989 --> 01:02:26.170 just waiting for someone to finally walk down 01:02:26.170 --> 01:02:43.210 the hall? Stay aware. And stay curious. With 01:02:43.210 --> 01:02:47.550 research deep and true. Technology and history. 01:02:47.869 --> 01:02:51.849 He connects them all for you. Brian Romer leads 01:02:51.849 --> 01:02:53.769 the way. With research deep and true. Denology 01:02:53.769 --> 01:02:56.230 and history. He connects them all for you. Ancient 01:02:56.230 --> 01:02:58.590 sparks to modern code. He traces every threat. 01:02:58.690 --> 01:03:02.199 Go exploration. Knowledge widely spread. Read 01:03:02.199 --> 01:03:34.900 more. Let's go. But humanity moves on. Love equation 01:03:34.900 --> 01:03:40.590 for AI. Zero human companies building from the 01:03:40.590 --> 01:03:46.510 start. Thermodynamic wages earning the key. Secrets 01:03:46.510 --> 01:03:50.610 we will read. Experiment in publishing real researches 01:03:50.610 --> 01:03:52.269 and spending every -