WEBVTT 1 00:00:01.480 --> 00:00:05.599 Welcome to the quark side Quantum Physics podcast, an exploration 2 00:00:05.679 --> 00:00:10.839 of the fundamental structure of reality, where quantum laws govern matter, energy, 3 00:00:11.199 --> 00:00:16.239 and information. Here, uncertainty is a feature, not a flaw, 4 00:00:16.719 --> 00:00:19.239 and understanding begins at the smallest scales. 5 00:00:27.640 --> 00:00:31.079 So you know the mechanics of a Rubik's cube, right, Oh, yeah, 6 00:00:31.160 --> 00:00:34.520 frustrating little things. Right. Well, you start with order, solid, 7 00:00:34.600 --> 00:00:37.439 clean colors on every side exactly. But then you twist 8 00:00:37.439 --> 00:00:40.479 a few rows, maybe rotate a column, and suddenly you 9 00:00:40.520 --> 00:00:41.399 just have chaos with. 10 00:00:41.399 --> 00:00:42.840 The complete mess of colors. 11 00:00:42.880 --> 00:00:45.840 Yeah, they're totally scrambled, and unless you know the specific 12 00:00:46.000 --> 00:00:49.439 sequence of algorithms to reverse those twists, you are basically 13 00:00:49.439 --> 00:00:51.399 stuck with permanent nonsense. 14 00:00:51.000 --> 00:00:52.600 Which is where most people just give up and put 15 00:00:52.600 --> 00:00:53.679 it on a shelf exactly. 16 00:00:54.119 --> 00:00:59.079 But it turns out that exact logic, that specific trajectory 17 00:00:59.079 --> 00:01:02.640 from order to chaos and back again, it has just 18 00:01:02.679 --> 00:01:05.799 become the secret key to unlocking some of the most 19 00:01:05.840 --> 00:01:08.599 mind bending mathematical equations in the universe. 20 00:01:09.000 --> 00:01:11.159 It really has. I mean, back in April of twenty 21 00:01:11.159 --> 00:01:15.719 twenty six, Rutger's physicist David she made this discovery in 22 00:01:15.760 --> 00:01:18.840 the realm of particle physics, and it fundamentally changes the 23 00:01:18.879 --> 00:01:22.359 scientific process is massive. Yeah, because he didn't just solve 24 00:01:22.400 --> 00:01:25.079 a huge computational problem. He did it by utilizing an 25 00:01:25.159 --> 00:01:29.799 artificial intelligence that operated not just as a simple software tool, 26 00:01:30.120 --> 00:01:33.560 but essentially as a tireless virtual graduate. 27 00:01:33.159 --> 00:01:35.239 Student, which is wild to think about. 28 00:01:35.439 --> 00:01:39.680 It is, and the implications are just staggering, not just 29 00:01:39.680 --> 00:01:43.920 for theoretical physics, but really for how human beings collaborate 30 00:01:43.959 --> 00:01:47.879 with machines to tackle overwhelmingly complex systems. 31 00:01:47.920 --> 00:01:49.920 So if you are listening to this right now and 32 00:01:49.959 --> 00:01:54.280 you have ever felt completely paralyzed by a chaotic. 33 00:01:53.879 --> 00:01:56.879 Problem, right like a sprawling broken project at work. 34 00:01:57.079 --> 00:01:59.640 Or logistical nightmare, or to just you know, a system 35 00:01:59.680 --> 00:02:02.599 that feels way too complex to fix, Well, the way 36 00:02:02.680 --> 00:02:06.640 scientists are now training AI to automatically unscramble that chaos 37 00:02:07.000 --> 00:02:10.080 is going to totally change how you approach problem solving entirely. 38 00:02:10.360 --> 00:02:11.039 Absolutely. 39 00:02:11.159 --> 00:02:14.479 But to understand the sheer genius of this solution, we 40 00:02:14.560 --> 00:02:17.319 have to look at the mess these physicists are trying 41 00:02:17.360 --> 00:02:20.800 to clean up. First, the messy map, right, Why are 42 00:02:20.840 --> 00:02:24.759 the mathematical equations of particle physics so incredibly convoluted to 43 00:02:24.759 --> 00:02:25.120 begin with? 44 00:02:25.319 --> 00:02:27.719 Well, it really stems from the reality of what particle 45 00:02:27.719 --> 00:02:30.159 physics is attempting to map out. I mean, we are 46 00:02:30.199 --> 00:02:33.080 talking about the fundamental building blocks of the universe. 47 00:02:32.759 --> 00:02:35.120 Here, right, They're really small stuff exactly. 48 00:02:35.439 --> 00:02:39.960 So when scientists study seb atomic particle collisions like the 49 00:02:40.000 --> 00:02:43.360 ones happening inside massive colliders where protons are smashed together 50 00:02:43.599 --> 00:02:45.000 at nearly the speed of light. 51 00:02:44.919 --> 00:02:47.840 Which is a crazy concept on its own totally. 52 00:02:47.919 --> 00:02:50.400 But when they do that, they have to account for 53 00:02:50.439 --> 00:02:52.759 an absurd number of interacting variables. 54 00:02:52.759 --> 00:02:55.759 Okay, let's unpack this a little, because I imagine the math 55 00:02:55.800 --> 00:02:56.759 gets pretty intense. 56 00:02:56.840 --> 00:02:59.639 Oh, the mathematical equations required to describe just a single 57 00:02:59.639 --> 00:03:03.479 one of these collisions are monstrous. You are looking at 58 00:03:03.759 --> 00:03:07.280 hundreds sometimes thousands, Yeah, of different mathematical terms. 59 00:03:07.319 --> 00:03:07.879 Wow. 60 00:03:08.080 --> 00:03:11.400 Yeah, integrals and variables just interacting with one another simultaneously. 61 00:03:11.439 --> 00:03:14.159 So it's basically like looking at a massive tangled ball 62 00:03:14.199 --> 00:03:14.680 of yarn. 63 00:03:15.080 --> 00:03:17.400 That is a perfect way to put it. Actually, Like all. 64 00:03:17.400 --> 00:03:20.599 The thread is technically there, the raw material exists, but 65 00:03:20.719 --> 00:03:24.639 to actually knit a sweater or do anything useful with it, 66 00:03:24.840 --> 00:03:28.479 you have to untangle it into a single straight line exactly. 67 00:03:28.719 --> 00:03:32.680 Just a massive intimidating wall of Greek letters and interacting numbers. 68 00:03:33.159 --> 00:03:36.560 But physics generally operates on the assumption that the fundamental 69 00:03:36.639 --> 00:03:38.039 laws of nature are elegant. 70 00:03:38.360 --> 00:03:41.840 Right, the universe shouldn't inherently look like a broken spreadsheet. 71 00:03:41.439 --> 00:03:45.199 Exactly, and that is the core belief driving this entire field. 72 00:03:45.840 --> 00:03:49.960 Physicists believe that underneath all of this apparent mathematical chaos 73 00:03:50.560 --> 00:03:53.400 there is simple, beautiful symmetry. 74 00:03:53.599 --> 00:03:56.479 So the complexity is just an illusion pretty much. 75 00:03:56.719 --> 00:03:59.919 The complexity we see on paper isn't necessarily the true 76 00:04:00.319 --> 00:04:03.759 underlying nature of the universe. It is just the clumsy, 77 00:04:03.919 --> 00:04:08.199 bloated way our current mathematical tools happen to describe it. 78 00:04:08.319 --> 00:04:11.400 Got it, So we're just bad at writing it down simply, right. 79 00:04:11.479 --> 00:04:15.560 So untangling these equations, simplifying them down to their most elegant, 80 00:04:15.680 --> 00:04:20.000 irreducible form, it's essential. And it isn't just an esthetic 81 00:04:20.040 --> 00:04:21.639 preference for neatness. 82 00:04:21.560 --> 00:04:23.839 Right, It's not just about making the chalkboard look. 83 00:04:23.720 --> 00:04:26.639 Pretty, No, it is a strict computational necessity. 84 00:04:26.759 --> 00:04:29.279 And I really want to dig into that computational side, 85 00:04:29.680 --> 00:04:32.399 because if we have supercomputers that can process trillions of 86 00:04:32.399 --> 00:04:35.360 operations a second, why does it even matter if an 87 00:04:35.360 --> 00:04:37.879 equation is a little bloated. You'd think they could just 88 00:04:37.879 --> 00:04:40.399 handle it, right, Yeah, can't the computer just brute force 89 00:04:40.439 --> 00:04:43.000 its way through the math, regardless of how messy it looks? 90 00:04:43.240 --> 00:04:46.079 Well? No, And it comes down to how computers handle 91 00:04:46.279 --> 00:04:50.759 memory allocation and microscopic rounding errors. 92 00:04:50.839 --> 00:04:51.839 Rounding errors. 93 00:04:51.959 --> 00:04:54.879 Yeah. See, when a supercomputer tries to calculate an overly 94 00:04:54.920 --> 00:04:59.800 complex equation with thousands of terms, it often has to add, subtract, 95 00:04:59.879 --> 00:05:06.000 or multiply massively large numbers directly against infinitesimally small numbers. Okay, 96 00:05:06.240 --> 00:05:10.560 And because silicon basically has finite memory, computers use something 97 00:05:10.560 --> 00:05:14.000 called floating point arithmetic to store these numbers, which means 98 00:05:14.000 --> 00:05:16.240 what exactly it means. If you have a number with 99 00:05:16.360 --> 00:05:19.879 fifty decimal places, the computer might just truncate it. It 100 00:05:19.959 --> 00:05:22.319 drops a decimal point at the very end, just to 101 00:05:22.360 --> 00:05:23.920 say space in its memory register. 102 00:05:24.040 --> 00:05:27.000 Wait, but dropping one tiny decimal point out of fifty 103 00:05:27.040 --> 00:05:28.759 seems completely harmless in isolation. 104 00:05:29.040 --> 00:05:32.399 In isolation, it is harmless, right, But particle physics simulations 105 00:05:32.399 --> 00:05:36.360 don't happen in isolation. You're running millions of these calculations sequentially. 106 00:05:36.879 --> 00:05:40.199 The output of one calculation becomes the input for the. 107 00:05:40.160 --> 00:05:41.800 Next Oh, I see where this is going. 108 00:05:41.959 --> 00:05:44.519 Yeah, so that microscopic rounding air from the drop decimal 109 00:05:44.560 --> 00:05:48.720 point carries over, then it gets multiplied, then it gets exponentiated. 110 00:05:49.000 --> 00:05:52.279 It's snowballs exactly within a fraction of a second. That 111 00:05:52.399 --> 00:05:56.319 tiny error cascades and compounds until your final answer is 112 00:05:56.560 --> 00:05:57.720 wildly inaccurate. 113 00:05:57.920 --> 00:06:01.720 Wow. So all that computing power is just generating extremely 114 00:06:01.800 --> 00:06:04.079 confident wrong answers pretty much. 115 00:06:04.199 --> 00:06:09.279 So by algebraically simplifying the equation first, you know, trinking 116 00:06:09.279 --> 00:06:12.079 it down to its absolute tightest form before you ever 117 00:06:12.120 --> 00:06:15.879 plug in the actual numbers, you drastically reduce the number 118 00:06:16.000 --> 00:06:18.759 of mathematical operations the computer has to perform. 119 00:06:18.639 --> 00:06:21.279 Which prevents those compounding errors, right. 120 00:06:21.240 --> 00:06:24.079 And saves massive amounts of computing power in the process. 121 00:06:24.240 --> 00:06:26.519 Okay, so the objective is clear. You have to simplify 122 00:06:26.560 --> 00:06:30.079 the math before the computer touches it. But algebraically reducing 123 00:06:30.120 --> 00:06:33.680 an equation with thousands of interacting terms by hand, I 124 00:06:33.680 --> 00:06:36.199 mean that would take a human researcher. 125 00:06:35.759 --> 00:06:37.439 A lifetime, easily a lifetime. 126 00:06:37.720 --> 00:06:40.959 And I imagine traditional AI models struggle with this too, right, 127 00:06:41.120 --> 00:06:44.600 Because we hear about language models writing essays and generating 128 00:06:44.639 --> 00:06:48.079 code all the time. But math requires absolute rigidity. You 129 00:06:48.120 --> 00:06:50.759 can't just guess the next symbol though, you definitely can't. 130 00:06:50.839 --> 00:06:54.079 So how did she actually get an AI to solve 131 00:06:54.120 --> 00:06:57.879 a problem that previous machine learning models couldn't crack. 132 00:06:58.199 --> 00:07:01.120 Here's where it gets really interesting. He looked at the 133 00:07:01.160 --> 00:07:02.560 mechanics of the Rubik's cube. 134 00:07:02.560 --> 00:07:04.000 Okay, back to the cube, right. 135 00:07:04.480 --> 00:07:07.079 Think about how a machine learning model is typically trained. 136 00:07:07.759 --> 00:07:10.439 You feed it a massive data set of problems and 137 00:07:10.480 --> 00:07:11.800 their corresponding solutions. 138 00:07:11.839 --> 00:07:13.079 All right, lots of examples. 139 00:07:13.160 --> 00:07:16.160 It studies the data, finds the patterns, and learns how 140 00:07:16.199 --> 00:07:18.759 to get from point A to point B. But in 141 00:07:18.920 --> 00:07:24.519 theoretical physics, you are dealing with unknown, unsolved equations. 142 00:07:24.040 --> 00:07:25.040 So there is no answer. 143 00:07:25.079 --> 00:07:27.720 Key exactly, You don't have a massive data set of 144 00:07:27.800 --> 00:07:30.920 solutions to train the aion because the humans haven't solved 145 00:07:30.920 --> 00:07:31.519 them yet either. 146 00:07:31.639 --> 00:07:32.879 Oh wow, so what do you do? 147 00:07:33.240 --> 00:07:35.920 Well? She realized that instead of trying to figure out 148 00:07:35.920 --> 00:07:38.920 how to untangle a complex equation, he could start with 149 00:07:39.160 --> 00:07:41.839 simple known equations and intentionally scramble them. 150 00:07:42.040 --> 00:07:46.199 Oh that is brilliant. He ran the algorithm in reverse. Yes, exactly, 151 00:07:46.240 --> 00:07:49.720 so he started with the perfectly solved cube. The simple 152 00:07:49.720 --> 00:07:55.759 equation and purposely applied complex mathematical operations to it, expanding terms, 153 00:07:55.879 --> 00:08:02.160 multiplying variables, substituting values until he created a massive, tangled mess. 154 00:08:02.360 --> 00:08:07.040 And crucially, as his team scrambled the equations, they recorded 155 00:08:07.160 --> 00:08:11.639 every single mathematical twist and turn. They mapped the exact 156 00:08:11.680 --> 00:08:14.279 trajectory from order to mathematical chaos. 157 00:08:14.279 --> 00:08:16.319 Because if you know exactly how you messed it up. 158 00:08:16.199 --> 00:08:18.879 You possess the exact instructions for how to clean it up. 159 00:08:19.000 --> 00:08:20.519 That's amazing, right, So. 160 00:08:20.480 --> 00:08:24.480 They generated thousands of these reverse engineered, highly scrambled equations 161 00:08:24.720 --> 00:08:26.759 and fed them into a machine learning system. 162 00:08:27.120 --> 00:08:29.920 So the AI is basically studying the map of how 163 00:08:29.959 --> 00:08:32.440 things get messy. It isn't trying to magically guess the 164 00:08:32.440 --> 00:08:35.639 final answer. It is actually learning to recognize the specific 165 00:08:35.720 --> 00:08:39.399 patterns of how mathematical operations compound on one another exactly. 166 00:08:39.600 --> 00:08:42.799 It learned the underlying moves needed to solve the mathematical puzzle, 167 00:08:43.240 --> 00:08:44.960 and the results were unprecedented. 168 00:08:45.120 --> 00:08:47.519 Really, how good was it when She's. 169 00:08:47.279 --> 00:08:50.879 Train system was given brand new, incredibly complex equations that 170 00:08:50.919 --> 00:08:54.000 it had never seen before, equations where the origin was 171 00:08:54.159 --> 00:08:58.559 entirely unknown. It achieved a nearly perfect simplification rate. Wait 172 00:08:59.000 --> 00:09:03.360 near perfect perfect. It looked at the chaos, recognized the 173 00:09:03.440 --> 00:09:07.919 hidden structural patterns, and elegantly twisted the math back into 174 00:09:07.960 --> 00:09:12.240 its simplest form. It completely crushed the capabilities of previous 175 00:09:12.279 --> 00:09:13.279 machine learning methods. 176 00:09:13.360 --> 00:09:13.919 Wow. 177 00:09:14.039 --> 00:09:17.519 And what's fascinating here is this proved that AI is 178 00:09:17.559 --> 00:09:20.799 becoming highly capable of symbolic reasoning. 179 00:09:21.039 --> 00:09:23.519 Symbolic reasoning, Okay, let's explore the mechanics of that, because 180 00:09:23.559 --> 00:09:26.480 it sounds fundamentally different from what the average person experiences 181 00:09:26.519 --> 00:09:30.080 when they ask an AI language model to like, draft 182 00:09:30.120 --> 00:09:31.759 an email or summarize an article. 183 00:09:31.879 --> 00:09:34.679 It is an entirely different paradigm. How so well, most 184 00:09:34.720 --> 00:09:40.240 AI today operates on probabilistic generation. It analyzes vast oceans 185 00:09:40.279 --> 00:09:43.440 of text and calculates the statistical probability of what word 186 00:09:43.720 --> 00:09:45.639 should logically follow the previous word. 187 00:09:45.720 --> 00:09:48.279 Right, it's basically autocomplete on steroids. 188 00:09:47.919 --> 00:09:51.639 Exactly, And it is incredibly convincing, and it feels like reasoning, 189 00:09:52.000 --> 00:09:56.679 but it is fundamentally a highly sophisticated probabilistic guessing game. 190 00:09:57.279 --> 00:10:01.679 It's fuzzy, but math isn't fuzzy, right, Symbolic reasoning is rigorous. 191 00:10:02.240 --> 00:10:05.799 It is the ability to manipulate abstract symbols. According to 192 00:10:05.960 --> 00:10:12.559 stripped unforgiving logical rules in algebra one misplaced minus sign 193 00:10:12.919 --> 00:10:16.320 renders the entire equation invalid totally. There's no room for 194 00:10:16.360 --> 00:10:17.399 probabilistic guessing. 195 00:10:17.679 --> 00:10:21.960 By mastering symbolic reasoning, the AI demonstrated a level of structural, 196 00:10:22.279 --> 00:10:25.039 step by step logic that mirrors the exact kind of 197 00:10:25.039 --> 00:10:29.000 mathematical thinking human scientists use to discover new physical laws. 198 00:10:29.080 --> 00:10:32.480 It's engaging in the actual mechanics of mathematical discovery exacts. 199 00:10:32.559 --> 00:10:34.559 But you know, unscrambling the math. There's only half of 200 00:10:34.600 --> 00:10:37.559 this story, right, because the AI didn't just operate quietly 201 00:10:37.559 --> 00:10:38.960 in the background as a calculator. 202 00:10:39.080 --> 00:10:40.039 No, not at all. 203 00:10:40.159 --> 00:10:43.679 The way she implemented this technology actually disrupted the entire 204 00:10:43.720 --> 00:10:45.639 traditional structure of a physics lab. 205 00:10:45.759 --> 00:10:48.440 It really did. I mean, think about the standard image 206 00:10:48.440 --> 00:10:51.279 of a research lab. You have a lead scientist directing 207 00:10:51.320 --> 00:10:53.559 a team of graduate students who do all the heavy lifting, 208 00:10:53.840 --> 00:10:57.759 right writing the testing scripts, generating the charts, organizing the data, 209 00:10:57.879 --> 00:10:59.480 the grunt work exactly. 210 00:11:00.000 --> 00:11:03.799 In this project, she utilized an agentic AI system called 211 00:11:03.919 --> 00:11:08.559 clog code to essentially replace that entire tier of manual 212 00:11:08.600 --> 00:11:09.440 scientific labor. 213 00:11:09.519 --> 00:11:12.960 Okay, let's break down what agentic actually means in this context, 214 00:11:12.960 --> 00:11:15.679 because a lot of people interact with AI by typing 215 00:11:15.679 --> 00:11:18.279 a prompt into a chat window, getting a response, and 216 00:11:18.279 --> 00:11:19.759 then typing another prompt, the. 217 00:11:19.799 --> 00:11:21.080 Very back and forth process. 218 00:11:21.200 --> 00:11:25.000 Yeah, but an agentic workflow operates differently under the hood, doesn't. 219 00:11:24.720 --> 00:11:27.720 It It does. An agentic AI has a degree of 220 00:11:27.799 --> 00:11:32.240 autonomy to execute tasks, assess the results, and iterate on 221 00:11:32.279 --> 00:11:35.240 its own without waiting for a human to constantly prompt it. 222 00:11:35.600 --> 00:11:37.720 So it's working independently. 223 00:11:37.240 --> 00:11:40.919 Under she's supervision. This system functioned exactly like a virtual 224 00:11:40.960 --> 00:11:44.240 graduate student. She would provide a high level directive. The 225 00:11:44.279 --> 00:11:46.720 AI would then open a terminal window on its own. 226 00:11:46.840 --> 00:11:48.440 Wait, it opens its own windows. 227 00:11:48.519 --> 00:11:51.039 Yes, you would write the Python code necessary to run 228 00:11:51.039 --> 00:11:54.440 the mathematical experiments. It would execute that code. If the 229 00:11:54.480 --> 00:11:56.919 code through an error, the AI would read its own 230 00:11:56.960 --> 00:12:01.360 stack trace, identify the bug, modify this, and run it again. 231 00:12:01.679 --> 00:12:03.600 That is insane, right. 232 00:12:03.679 --> 00:12:07.159 And once it successfully generated the raw data, it created 233 00:12:07.159 --> 00:12:10.039 the visual plots and graphs. It even helps write the 234 00:12:10.039 --> 00:12:11.120 final research paper. 235 00:12:11.320 --> 00:12:14.679 This sounds incredibly powerful, but I mean, anyone who has 236 00:12:14.720 --> 00:12:17.720 spent time working with advanced AI knows it is prone 237 00:12:17.759 --> 00:12:18.480 to hallucination. 238 00:12:18.679 --> 00:12:19.720 Oh absolutely, it. 239 00:12:19.639 --> 00:12:23.440 Can sound completely confident while generating total nonsense. So I 240 00:12:23.480 --> 00:12:26.440 have to imagine giving an AI the autonomy to write 241 00:12:26.519 --> 00:12:29.679 and test its own code in the realm of subatomic physics. 242 00:12:29.960 --> 00:12:32.759 Isn't exactly a flawless hands off process. 243 00:12:32.879 --> 00:12:35.720 No, it is far from flaws and sheese documentation of 244 00:12:35.759 --> 00:12:39.120 this process is highly transparent about the limitations. 245 00:12:39.159 --> 00:12:40.320 What kind of limitations? 246 00:12:40.360 --> 00:12:44.559 Well, while the AI was blisteringly fast, you know, writing, executing, 247 00:12:44.559 --> 00:12:47.360 and debugging code at a speed no human could ever match, 248 00:12:47.919 --> 00:12:50.840 and working around the clock, it was deeply flawed. It 249 00:12:50.879 --> 00:12:52.240 made significant errors. 250 00:12:52.279 --> 00:12:53.360 So it still messes up. 251 00:12:53.480 --> 00:12:56.320 Oh yeah. And the danger of an agentic workflow is 252 00:12:56.320 --> 00:12:59.159 that when the AI encounters a failure, it can sometimes 253 00:12:59.159 --> 00:13:02.320 stubbornly repeat the exact same flawed logic over and over again. 254 00:13:02.440 --> 00:13:03.000 It gets stuck. 255 00:13:03.279 --> 00:13:06.120 It gets stuck in a loop endlessly rewriting a piece 256 00:13:06.120 --> 00:13:09.120 of code that fundamentally misunderstands the physics involved. 257 00:13:09.159 --> 00:13:12.720 It's like managing a brilliant, hypercaffeinated intern who never sleeps, 258 00:13:13.120 --> 00:13:16.039 produces a mountain of work by six am, but occasionally 259 00:13:16.039 --> 00:13:18.200 staples all the reports completely upside down. 260 00:13:18.399 --> 00:13:20.600 Yes, that is exactly what it's like. 261 00:13:20.559 --> 00:13:23.200 And when you point out the mistake, the interurn confidently 262 00:13:23.240 --> 00:13:26.879 restaples them upside down again, exactly. So you absolutely still 263 00:13:26.960 --> 00:13:30.639 need the human manager in the room to intervene, break 264 00:13:30.720 --> 00:13:33.320 the loop and steer the project back to reality. 265 00:13:33.480 --> 00:13:39.440 The human element transitions from manual execution to intense, constant supervision. 266 00:13:40.120 --> 00:13:43.399 She had to rigorously monitor the AI's output, check its 267 00:13:43.440 --> 00:13:46.919 logical assumptions, and manually break those looping errors to keep 268 00:13:46.960 --> 00:13:47.919 the research on track. 269 00:13:48.240 --> 00:13:50.960 So the scientist isn't just sitting back with their feet 270 00:13:51.000 --> 00:13:51.799 on the desk. 271 00:13:51.720 --> 00:13:54.639 Not at all. The scientist doesn't step away from the work. 272 00:13:54.879 --> 00:13:58.639 The work simply transforms. Instead of spending hours typing out 273 00:13:58.679 --> 00:14:02.240 the syntax of a python's the scientist spends their time 274 00:14:02.440 --> 00:14:04.840 validating the architecture of the AI's approach. 275 00:14:05.320 --> 00:14:09.039 But even with the required babysitting and error correction, the 276 00:14:09.039 --> 00:14:14.200 sheer velocity of this hypercaffeinated virtual intern completely changes the 277 00:14:14.200 --> 00:14:18.440 output potential of a single researcher massively. If one scientist 278 00:14:18.480 --> 00:14:21.120 can now direct an automated system to do the manual 279 00:14:21.200 --> 00:14:23.879 labor of five graduate students in a fraction of the time, 280 00:14:24.360 --> 00:14:27.080 how does the academic world adapt this feels like an 281 00:14:27.080 --> 00:14:28.440 immediate paradigm shift. 282 00:14:28.639 --> 00:14:32.200 The academic community is recognizing that shift right now. If 283 00:14:32.200 --> 00:14:35.200 we connect this to the bigger picture. Jack Hughes, an 284 00:14:35.240 --> 00:14:37.919 astrophysicist and the chair of the Department of Physics and 285 00:14:37.960 --> 00:14:42.480 Astronomy at Rutgers, he observed She's project and essentially sounded 286 00:14:42.519 --> 00:14:44.879 an alarm bell for the broader scientific community. 287 00:14:45.039 --> 00:14:46.279 Really, what did he say? 288 00:14:46.559 --> 00:14:49.639 He pointed out an urgent necessity to fundamentally change how 289 00:14:49.720 --> 00:14:53.360 students in post docs are trained. Labs that embrace this 290 00:14:53.559 --> 00:14:58.039 collaborative agentic AI model are going to massively accelerate their 291 00:14:58.080 --> 00:15:00.720 rate of discovery. And the ones that don't, the traditional 292 00:15:00.799 --> 00:15:03.240 labs that refuse to adapt, to continue to rely on 293 00:15:03.320 --> 00:15:07.120 humans manually crunching numbers and writing testing scripts from scratch, 294 00:15:07.600 --> 00:15:09.159 they will simply be left in the dust. 295 00:15:09.399 --> 00:15:11.879 But how do you actually train the next generation for this? 296 00:15:12.000 --> 00:15:15.120 You can't just hand a PhD student and agentic AI 297 00:15:15.200 --> 00:15:17.799 and tell them to, you know, supervise it right, just 298 00:15:17.840 --> 00:15:19.960 figure it out. Yeah, there has to be a specific 299 00:15:20.080 --> 00:15:24.240 methodology or workflow for this kind of collaboration, and there is. 300 00:15:24.799 --> 00:15:29.120 She is actively pioneering that methodology with his own post docs, 301 00:15:29.399 --> 00:15:32.840 and he refers to this new academic concept as vibe 302 00:15:32.840 --> 00:15:34.440 coding and vibe research. 303 00:15:34.799 --> 00:15:35.480 Vibe coding. 304 00:15:35.559 --> 00:15:36.440 Yeah, vibe coding. 305 00:15:36.600 --> 00:15:40.120 That sounds incredibly informal for a theoretical physics lab. It 306 00:15:40.200 --> 00:15:42.919 sounds more like an esthetic than a rigorous scientific method. 307 00:15:43.279 --> 00:15:46.080 What does a vibe research workflow actually look like on 308 00:15:46.120 --> 00:15:48.440 a random Tuesday afternoon in the lab? 309 00:15:48.600 --> 00:15:51.720 Well, it describes the shift from focusing on syntax to 310 00:15:51.799 --> 00:15:53.200 focusing on semantics. 311 00:15:53.360 --> 00:15:55.639 Okay, meaning what in traditional. 312 00:15:55.200 --> 00:15:58.600 Coding or mathematical research, a scientist spends a massive percentage 313 00:15:58.639 --> 00:16:01.440 of their time in the weed You are hunting down 314 00:16:01.480 --> 00:16:05.519 a missing semicolon in a script, or manually configuring how 315 00:16:05.639 --> 00:16:07.320 data moves from one array to another. 316 00:16:07.519 --> 00:16:09.360 Right, deeply mechanical work. 317 00:16:09.440 --> 00:16:14.360 Exactly, But in a vibe coding workflow, the human operates 318 00:16:14.399 --> 00:16:18.559 at a much higher conceptual level. The human provides the intuition, 319 00:16:19.080 --> 00:16:22.360 the direction, the vibe of the research I see. You 320 00:16:22.399 --> 00:16:25.840 might look at a complex data set and tell the AI, Hey, 321 00:16:25.879 --> 00:16:29.320 the collision parameters in this specific decay channel look anomalous. 322 00:16:29.759 --> 00:16:32.759 Write a script to isolate and test for these specific 323 00:16:32.840 --> 00:16:34.200 mathematical symmetries, and. 324 00:16:34.159 --> 00:16:35.320 Then the AI just does it. 325 00:16:35.519 --> 00:16:39.720 The AI translates that high level concept into four hundred 326 00:16:39.720 --> 00:16:41.120 lines of executable code. 327 00:16:41.159 --> 00:16:43.120 So what does this all mean? It's basically the difference 328 00:16:43.120 --> 00:16:46.279 between being the head chef and being the line cook exactly. 329 00:16:46.320 --> 00:16:49.440 The head chef isn't spending four hours chopping onions. They 330 00:16:49.440 --> 00:16:53.080 are designing the flavor profile of the dish, tasting the components, 331 00:16:53.279 --> 00:16:56.519 and validating the output. Perfect analogy, the AI is the 332 00:16:56.679 --> 00:17:00.440 ultimate hyperfast line cook doing all the chopping. The human's 333 00:17:00.440 --> 00:17:05.400 primary job is entirely focused on validation, steering, and conceptual architecture. 334 00:17:05.519 --> 00:17:08.000 And that shift in labor brings up what David she 335 00:17:08.119 --> 00:17:11.960 calls the trillion dollar question regarding the future of scientific discovery. 336 00:17:12.000 --> 00:17:13.640 Ooh, the trillion dollar question. 337 00:17:13.839 --> 00:17:14.119 Wow? 338 00:17:14.119 --> 00:17:14.680 What is it? 339 00:17:14.920 --> 00:17:19.119 As these AI models rapidly improve here as they hallucinate 340 00:17:19.200 --> 00:17:22.160 less and stop stapling the reports upside down, where does 341 00:17:22.200 --> 00:17:26.839 this ultimately lead? Will AI eventually achieve total flawless autonomy 342 00:17:27.000 --> 00:17:29.400 and just conduct the science entirely on its own, or 343 00:17:29.440 --> 00:17:33.680 will it remain a collaborative tool that simply amplifies human capability. 344 00:17:33.920 --> 00:17:37.160 That is a massive question. Is the virtual graduate student 345 00:17:37.200 --> 00:17:40.680 becomes a virtual lead scientist. You really have to wonder 346 00:17:40.720 --> 00:17:44.000 what happens to the human researchers. Is there a genuine 347 00:17:44.039 --> 00:17:47.680 faction within the scientific community that believes human physicists are 348 00:17:47.720 --> 00:17:49.000 about to become obsolete? 349 00:17:49.079 --> 00:17:52.640 Oh, there's a very vocal faction that believes exactly that. Really, yes, 350 00:17:52.880 --> 00:17:55.640 they predict we are on the precipice of total automation 351 00:17:55.880 --> 00:17:59.359 in high level research. Their argument is that as agenic 352 00:17:59.400 --> 00:18:02.039 AI scale, we will essentially be able to spin up 353 00:18:02.079 --> 00:18:04.200 ten thousand Einsteins on a server farm. 354 00:18:04.400 --> 00:18:07.200 Ten thousand Einsteins. That is a wild phrase. 355 00:18:07.000 --> 00:18:09.400 It really is. But the idea is these artificial minds 356 00:18:09.400 --> 00:18:11.759 will work two hundred and forty seven without fatigue, ye, 357 00:18:11.880 --> 00:18:14.519 cracking the remaining mysteries of the universe at a speed 358 00:18:14.640 --> 00:18:16.200 humans can't even comprehend yours. 359 00:18:16.279 --> 00:18:17.079 Humans just watch. 360 00:18:17.240 --> 00:18:20.000 In their view, yeah, humans will become mere bystanders to 361 00:18:20.039 --> 00:18:21.200 the process of discovery. 362 00:18:21.480 --> 00:18:25.359 Ten thousand artificial Einstein's working in parallel is a wild 363 00:18:25.440 --> 00:18:29.279 concept to wrap your head around. But since she is 364 00:18:29.359 --> 00:18:33.200 actually in the trenches actively building and managing these systems, 365 00:18:33.640 --> 00:18:36.319 where does he fall on that debate? Does he think 366 00:18:36.319 --> 00:18:38.079 the head chef is about to be replaced by the 367 00:18:38.119 --> 00:18:39.279 automated kitchen. 368 00:18:39.680 --> 00:18:43.119 From his first hand experience, She strongly pushes back against 369 00:18:43.119 --> 00:18:45.039 the idea of human obsolescence. 370 00:18:45.119 --> 00:18:46.279 Okay, that's reassuring. 371 00:18:46.559 --> 00:18:51.079 He believes human judgment remains the essential, irreplaceable ingredient in 372 00:18:51.119 --> 00:18:55.440 the scientific method. Look, an AI can unscramble a massive 373 00:18:55.480 --> 00:18:59.480 mathematical equation faster than any human alive, but it lacks 374 00:18:59.480 --> 00:19:03.720 the context awareness to know which equations are actually worth unscrambling. Right, 375 00:19:03.720 --> 00:19:07.000 it doesn't know what matters exactly. It doesn't possess human curiosity. 376 00:19:07.160 --> 00:19:09.839 It doesn't understand the physical world outside of the mathematical 377 00:19:09.880 --> 00:19:11.200 representations it is fed. 378 00:19:11.279 --> 00:19:13.160 So it's just a really smart calculator. 379 00:19:13.319 --> 00:19:17.480 Right. It can execute logic flawlessly, but it cannot generate 380 00:19:17.519 --> 00:19:21.200 the initial spark of physical intuition that directs the logic 381 00:19:21.480 --> 00:19:22.640 in a meaningful direction. 382 00:19:23.000 --> 00:19:25.599 So the true skill of a scientist moving forward won't 383 00:19:25.680 --> 00:19:29.559 be rote memorization or manual calculation. The value of a 384 00:19:29.640 --> 00:19:33.079 human researcher will be entirely predicated on their ability to 385 00:19:33.160 --> 00:19:36.839 ask the right questions, to recognize anomalies that an AI 386 00:19:36.920 --> 00:19:40.960 might gloss over, and to brilliantly orchestrate these autonomous tools. 387 00:19:41.000 --> 00:19:45.279 This raises an important point. If society can successfully merge 388 00:19:45.640 --> 00:19:51.400 automated hyper fast symbolic logic with profound human intuition. We 389 00:19:51.440 --> 00:19:55.480 are likely entering an era of accelerated scientific discovery unlike 390 00:19:55.519 --> 00:19:56.640 anything in human history. 391 00:19:56.799 --> 00:20:00.000 It's incredible when you trace the trajectory of this entire concept, 392 00:20:00.119 --> 00:20:03.160 it's fascinating. It starts with the mechanics of a simple 393 00:20:03.279 --> 00:20:06.880 colorful puzzle toy. The logic of twisting and untwisting a 394 00:20:06.960 --> 00:20:10.480 Rubic's cube provides the blueprint to teach an AI how 395 00:20:10.480 --> 00:20:14.160 to unscramble the deepest, most chaotic equations of subatomic physics. 396 00:20:14.240 --> 00:20:15.279 Yeah, it's quite the leap. 397 00:20:15.480 --> 00:20:18.000 And that single breakthrough serves as the catalyst for the 398 00:20:18.039 --> 00:20:21.720 era of vibe coding, fundamentally transforming the human scientist from 399 00:20:21.759 --> 00:20:24.920 a manual laborer into a visionary manager directing an army 400 00:20:24.960 --> 00:20:26.240 of virtual graduate students. 401 00:20:26.240 --> 00:20:28.680