WEBVTT

1
00:00:00.000 --> 00:00:05.401
<v Cassidy>AI’s appetite for training data is turning into a very expensive habit.

2
00:00:05.401 --> 00:00:08.900
<v Cassidy>Somebody, apparently, is finally getting paid.

3
00:00:08.900 --> 00:00:10.878
<v Bill>This is AI Daily Briefing.

4
00:00:10.878 --> 00:00:15.746
<v Bill>Today: the data-economy number behind the infrastructure splurge

5
00:00:15.746 --> 00:00:23.125
<v Bill>plus two science results and an agent benchmark that actually has to make it through eight hours.

6
00:00:23.125 --> 00:00:30.961
<v Cassidy>Start with Micro1—the headline is impressive, but the number underneath may explain the whole buildout.

7
00:00:30.961 --> 00:00:34.080
<v Bill>Tap follow so the next episode finds you.

8
00:00:34.911 --> 00:00:36.991
<v Cassidy>TechCrunch, <00:00:35.791>with <00:00:36.008>Marina <00:00:36.411>Temkin:

9
00:00:38.504 --> 00:00:41.544
<v Quote>One <00:00:38.691>of <00:00:38.784>these <00:00:39.048>fast-growing <00:00:39.609>businesses <00:00:40.317>is <00:00:40.504>Micro1

10
00:00:41.624 --> 00:00:48.824
<v Quote>a <00:00:41.752>four-year-old <00:00:42.314>startup <00:00:42.696>that <00:00:42.877>expanded <00:00:43.344>its <00:00:43.531>gross <00:00:43.921>annual <00:00:44.324>run <00:00:44.552>rate <00:00:44.792>from <00:00:45.064>$100 <00:00:45.584>million <00:00:45.997>to <00:00:46.384>$500 <00:00:46.944>million <00:00:47.288>over <00:00:47.564>the <00:00:47.688>past <00:00:47.971>eight <00:00:48.161>months

11
00:00:48.856 --> 00:00:51.424
<v Quote>according <00:00:49.197>to <00:00:49.344>a <00:00:49.441>person <00:00:49.837>familiar <00:00:50.296>with <00:00:50.444>the <00:00:50.554>company.

12
00:00:51.528 --> 00:00:54.984
<v Quote>Like <00:00:51.984>its <00:00:52.184>peers <00:00:52.600>that <00:00:52.728>hire <00:00:53.041>domain <00:00:53.444>experts <00:00:53.896>such <00:00:54.051>as <00:00:54.184>doctors

13
00:00:55.054 --> 00:00:58.104
<v Quote>lawyers <00:00:55.664>and <00:00:55.842>scientists <00:00:56.451>on <00:00:56.544>a <00:00:56.637>contract <00:00:57.144>basis

14
00:00:58.184 --> 00:01:02.184
<v Quote>Micro1 <00:00:58.804>retains <00:00:59.264>roughly <00:00:59.784>60% <00:01:00.211>to <00:01:00.424>70% <00:01:01.171>of <00:01:01.256>that <00:01:01.441>figure

15
00:01:02.214 --> 00:01:07.144
<v Quote>putting <00:01:02.464>its <00:01:02.644>net <00:01:02.881>annual <00:01:03.284>run <00:01:03.512>rate <00:01:03.754>between <00:01:04.184>$150 <00:01:05.104>million <00:01:05.464>and <00:01:05.784>$200 <00:01:06.354>million.

16
00:01:07.473 --> 00:01:16.993
<v Cassidy>Micro1 <00:01:08.153>going <00:01:08.401>from <00:01:08.673>$100 <00:01:09.343>million <00:01:10.193>to <00:01:10.473>$500 <00:01:11.153>million <00:01:11.766>in <00:01:11.940>gross <00:01:12.313>run <00:01:12.497>rate <00:01:12.806>in <00:01:12.953>eight <00:01:13.244>months <00:01:13.953>is <00:01:14.313>the <00:01:14.473>first <00:01:14.726>clean <00:01:15.057>read <00:01:15.366>on <00:01:15.553>demand <00:01:16.086>in <00:01:16.225>this <00:01:16.396>infrastructure

17
00:01:17.073 --> 00:01:27.153
<v Cassidy>boom. <00:01:17.693>The <00:01:18.415>appetite <00:01:18.953>isn't <00:01:19.197>stopping <00:01:19.580>at <00:01:19.773>GPU <00:01:20.284>leases; <00:01:21.340>it's <00:01:21.509>reaching <00:01:21.863>doctors, <00:01:22.493>lawyers, <00:01:23.013>and <00:01:23.168>scientists <00:01:24.193>labeling <00:01:24.853>the <00:01:24.975>material <00:01:25.553>labs <00:01:25.969>want.

18
00:01:27.173 --> 00:01:40.194
<v Bill>And <00:01:27.313>$500 <00:01:28.053>million <00:01:28.833>gross <00:01:29.293>becomes <00:01:29.823>roughly <00:01:30.713>$150 <00:01:31.983>million <00:01:32.833>to <00:01:33.073>$200 <00:01:33.973>million <00:01:34.813>net <00:01:35.409>once <00:01:36.433>60% <00:01:36.940>to <00:01:37.233>70% <00:01:38.225>goes <00:01:38.737>back <00:01:39.033>out <00:01:39.206>to <00:01:39.386>contractors.

19
00:01:40.313 --> 00:01:48.273
<v Bill>Healthy <00:01:41.215>business. <00:01:42.133>But <00:01:42.713>the <00:01:42.879>number <00:01:43.193>you <00:01:43.373>can <00:01:43.615>actually <00:01:44.193>operate <00:01:44.886>on <00:01:45.340>is <00:01:45.753>a <00:01:45.873>lot <00:01:46.193>smaller <00:01:46.785>than <00:01:46.933>the <00:01:47.073>headline.

20
00:01:48.273 --> 00:01:56.513
<v Cassidy>Still, <00:01:49.265>that <00:01:49.473>net <00:01:49.773>run <00:01:50.017>rate <00:01:50.326>is <00:01:50.513>real <00:01:50.895>evidence <00:01:51.425>that <00:01:51.589>somebody <00:01:51.913>can <00:01:52.097>fund <00:01:52.385>this <00:01:52.566>stack <00:01:52.865>with <00:01:53.037>customer <00:01:53.484>demand, <00:01:54.633>not <00:01:54.801>just <00:01:55.030>infrastructure <00:01:55.633>financing.

21
00:01:56.686 --> 00:02:06.913
<v Cassidy>Micro1 <00:01:57.857>says <00:01:58.169>automated <00:01:58.726>video <00:01:59.048>descriptions <00:01:59.753>and <00:01:59.944>reusable <00:02:00.593>off-the-shelf <00:02:01.304>datasets <00:02:02.013>can <00:02:02.481>push <00:02:02.740>gross <00:02:03.133>margins <00:02:03.580>to <00:02:03.793>80% <00:02:04.140>or <00:02:04.353>90%—which <00:02:06.090>raises <00:02:06.453>the <00:02:06.563>obvious

22
00:02:06.984 --> 00:02:12.071
<v Cassidy>question: <00:02:07.893>how <00:02:08.481>carefully <00:02:08.945>will <00:02:09.142>buyers <00:02:09.610>police <00:02:09.980>where <00:02:10.145>that <00:02:10.353>same <00:02:10.737>data <00:02:11.041>gets <00:02:11.345>sold?

23
00:02:12.071 --> 00:02:19.873
<v Bill>Exactly. <00:02:13.766>If <00:02:14.097>labs <00:02:14.410>decide <00:02:14.793>their <00:02:15.037>training <00:02:15.457>data <00:02:15.793>is <00:02:15.993>a <00:02:16.113>moat, <00:02:17.105>some <00:02:17.260>of <00:02:17.361>this <00:02:17.601>work <00:02:17.846>moves <00:02:18.220>in-house <00:02:19.073>fast.

24
00:02:20.060 --> 00:02:28.833
<v Bill>Micro1's <00:02:21.370>margin <00:02:21.769>expansion <00:02:22.333>depends <00:02:22.806>on <00:02:22.993>synthetic <00:02:23.773>and <00:02:24.033>repeat-sale <00:02:24.993>data <00:02:25.413>staying <00:02:26.175>valuable <00:02:26.740>after <00:02:27.093>the <00:02:27.206>customer <00:02:27.693>has <00:02:27.921>seen <00:02:28.140>it <00:02:28.289>once.

25
00:02:29.604 --> 00:02:32.244
<v Cassidy>Here's <00:02:29.884>Lixue <00:02:30.337>Cheng <00:02:30.857>at <00:02:31.010>Nature <00:02:31.348>Communications:

26
00:02:32.829 --> 00:02:36.189
<v Quote>Here <00:02:33.336>we <00:02:33.429>bring <00:02:33.661>this <00:02:33.851>paradigm <00:02:34.376>to <00:02:34.491>fruition <00:02:35.021>with <00:02:35.249>Orbformer

27
00:02:36.229 --> 00:02:44.829
<v Quote>a <00:02:36.331>transferable <00:02:37.133>wavefunction <00:02:37.776>model <00:02:38.160>pretrained <00:02:38.722>on <00:02:38.909>22,000 <00:02:39.856>equilibrium <00:02:40.609>and <00:02:40.749>dissociating <00:02:41.527>structures <00:02:42.173>that <00:02:42.469>can <00:02:42.616>be <00:02:42.749>fine-tuned <00:02:43.469>on <00:02:43.709>unseen <00:02:44.253>molecules

28
00:02:44.873 --> 00:02:49.309
<v Quote>reaching <00:02:45.256>an <00:02:45.371>accuracy–cost <00:02:46.349>ratio <00:02:46.829>rivalling <00:02:47.365>classical <00:02:47.933>multireference <00:02:48.729>methods.

29
00:02:49.522 --> 00:02:55.309
<v Quote>On <00:02:49.989>established <00:02:50.480>benchmarks <00:02:51.016>as <00:02:51.117>well <00:02:51.336>as <00:02:51.437>more <00:02:51.676>challenging <00:02:52.253>bond <00:02:52.569>dissociations <00:02:53.429>and <00:02:53.602>Diels–Alder <00:02:54.325>reactions

30
00:02:55.389 --> 00:03:01.709
<v Quote>Orbformer <00:02:56.002>is <00:02:56.149>the <00:02:56.349>only <00:02:56.715>method <00:02:57.021>that <00:02:57.204>consistently <00:02:57.926>converges <00:02:58.482>to <00:02:58.642>chemical <00:02:59.149>accuracy <00:02:59.989>(1 <00:03:00.477>kcal/mol).

31
00:03:02.771 --> 00:03:12.263
<v Bill>Orbformer is the kind of benchmark claim I’ll take seriously: 1 kilocalorie per mole on bond dissociation and Diels–Alder reactions

32
00:03:12.263 --> 00:03:13.463
<v Bill>consistently.

33
00:03:13.463 --> 00:03:22.810
<v Bill>Pretrain on 22,000 structures, fine-tune on an unfamiliar molecule, and stop paying full multireference-compute price every time.

34
00:03:22.810 --> 00:03:27.740
<v Cassidy>And Microsoft Research put the methodology in Nature Communications.

35
00:03:27.740 --> 00:03:29.628
<v Cassidy>There’s a technical report

36
00:03:29.628 --> 00:03:31.226
<v Cassidy>established benchmarks

37
00:03:31.226 --> 00:03:39.704
<v Cassidy>and a precise accuracy target—1 kcal/mol—not a cinematic molecule animation and a promise that it’s “transformative.”

38
00:03:39.704 --> 00:03:44.489
<v Bill>If that accuracy-to-cost ratio holds outside this paper’s test set

39
00:03:44.489 --> 00:03:54.053
<v Bill>chemistry teams will care a lot less about the model’s mystique than about getting classical multireference-quality answers cheaper.

40
00:03:54.053 --> 00:03:56.809
<v Bill>That makes a real case for deployment.

41
00:03:56.809 --> 00:04:02.391
<v Cassidy>We just talked about demand spreading through the training-data supply chain.

42
00:04:02.391 --> 00:04:10.435
<v Cassidy>Here’s the other side of the infrastructure story: the biggest balance sheets are funding compute buildings and

43
00:04:10.435 --> 00:04:16.451
<v Cassidy>increasingly the scientific tools that determine what those buildings get used for.

44
00:04:17.367 --> 00:04:19.927
<v Cassidy>University <00:04:17.794>of <00:04:17.918>Waterloo, <00:04:18.647>with <00:04:19.011>Benjamin <00:04:19.423>Schneider:

45
00:04:20.430 --> 00:04:26.750
<v Quote>We <00:04:20.652>evaluate <00:04:21.257>ASH <00:04:21.603>on <00:04:21.770>two <00:04:22.001>complementary <00:04:22.725>environments <00:04:23.438>demanding <00:04:23.937>multi-hour <00:04:24.634>planning: <00:04:25.690>Pokémon <00:04:26.170>Emerald

46
00:04:26.790 --> 00:04:30.910
<v Quote>a <00:04:26.894>turn-based <00:04:27.550>RPG <00:04:28.630>and <00:04:28.770>The <00:04:28.876>Legend <00:04:29.203>of <00:04:29.377>Zelda: <00:04:29.810>The <00:04:29.930>Minish <00:04:30.350>Cap

47
00:04:30.950 --> 00:04:33.110
<v Quote>a <00:04:31.038>real <00:04:31.294>time <00:04:31.607>action-adventure <00:04:32.414>game.

48
00:04:33.257 --> 00:04:35.710
<v Quote>In <00:04:33.598>both <00:04:33.870>games <00:04:34.481>behavioral <00:04:35.050>cloning

49
00:04:35.758 --> 00:04:40.350
<v Quote>retrieval-augmented <00:04:36.890>and <00:04:37.134>zero-shot <00:04:37.841>foundation-model <00:04:38.710>baselines <00:04:39.300>plateau

50
00:04:40.377 --> 00:04:43.870
<v Quote>while <00:04:40.590>ASH <00:04:40.958>sustains <00:04:41.357>progression <00:04:41.927>across <00:04:42.310>our <00:04:42.510>8-hour <00:04:43.063>evaluation.

51
00:04:44.217 --> 00:04:53.230
<v Quote>ASH <00:04:45.110>reaches <00:04:45.443>an <00:04:45.600>average <00:04:46.003>of <00:04:46.350>11.2/12 <00:04:47.928>milestones <00:04:48.563>in <00:04:48.730>Pokémon <00:04:49.230>Emerald <00:04:49.870>and <00:04:50.190>9.9/12 <00:04:51.817>in <00:04:51.916>Legend <00:04:52.243>of <00:04:52.417>Zelda

52
00:04:53.257 --> 00:05:02.030
<v Quote>while <00:04:53.410>the <00:04:53.518>strongest <00:04:54.003>baseline <00:04:54.478>gets <00:04:54.710>stuck <00:04:54.963>in <00:04:55.118>both <00:04:55.359>environments <00:04:56.003>at <00:04:56.163>an <00:04:56.320>average <00:04:56.697>of <00:04:56.910>6.5/12 <00:04:58.610>and <00:04:59.070>6.0/12 <00:05:00.575>milestones <00:05:01.575>respectively.

53
00:05:02.288 --> 00:05:11.328
<v Bill>Eight <00:05:02.448>hours <00:05:02.955>in <00:05:03.248>Pokémon <00:05:03.888>Emerald <00:05:04.548>and <00:05:04.928>Minish <00:05:05.448>Cap <00:05:05.915>is <00:05:06.168>a <00:05:06.288>much <00:05:06.665>better <00:05:07.088>agent <00:05:07.600>test <00:05:08.096>than <00:05:08.328>the <00:05:08.541>usual <00:05:09.077>little <00:05:09.568>three-step <00:05:10.378>routine.

54
00:05:11.435 --> 00:05:18.928
<v Bill>ASH <00:05:11.872>gets <00:05:12.208>11.2 <00:05:13.328>of <00:05:13.648>12 <00:05:14.118>Pokémon <00:05:14.681>milestones; <00:05:15.828>the <00:05:15.952>best <00:05:16.270>baseline <00:05:16.837>stalls <00:05:17.301>at <00:05:17.568>6.5.

55
00:05:19.024 --> 00:05:21.016
<v Bill>That <00:05:19.488>gap’s <00:05:19.856>hard <00:05:20.101>to <00:05:20.288>ignore.

56
00:05:21.016 --> 00:05:29.968
<v Cassidy>And <00:05:21.372>Benjamin <00:05:21.776>Schneider <00:05:22.261>at <00:05:22.421>Waterloo <00:05:22.901>is <00:05:23.048>doing <00:05:23.296>this <00:05:23.536>from <00:05:23.788>unlabeled, <00:05:24.768>noisy <00:05:25.221>internet <00:05:25.715>video—no <00:05:26.976>hand-built <00:05:27.477>reward <00:05:27.928>and <00:05:28.155>no <00:05:28.425>expert <00:05:28.825>action <00:05:29.237>labels.

57
00:05:30.518 --> 00:05:37.714
<v Cassidy>A <00:05:30.585>result <00:05:30.976>this <00:05:31.230>specific <00:05:31.760>from <00:05:31.928>a <00:05:32.025>single-researcher <00:05:33.008>thesis <00:05:33.541>deserves <00:05:34.024>attention, <00:05:35.136>even <00:05:35.381>if <00:05:35.515>it <00:05:35.628>isn’t <00:05:35.928>a <00:05:36.008>major <00:05:36.268>lab <00:05:36.503>announcement.

58
00:05:37.714 --> 00:05:40.128
<v Bill>I <00:05:37.861>still <00:05:38.176>want <00:05:38.388>the <00:05:38.499>production <00:05:39.075>translation.

59
00:05:40.228 --> 00:05:48.048
<v Bill>Pokémon <00:05:41.035>is <00:05:41.306>controlled: <00:05:42.308>you’ve <00:05:42.488>got <00:05:42.648>a <00:05:42.757>finite <00:05:43.235>world <00:05:43.608>and <00:05:43.797>finite <00:05:44.259>controller <00:05:44.828>actions, <00:05:45.688>and <00:05:45.868>all <00:05:46.128>12 <00:05:46.513>milestones <00:05:47.188>are <00:05:47.301>countable.

60
00:05:48.160 --> 00:05:56.048
<v Bill>Does <00:05:48.640>that <00:05:48.848>11.2 <00:05:49.888>survive <00:05:50.488>the <00:05:50.672>step-seven <00:05:51.388>failure <00:05:51.904>that <00:05:52.348>wrecks <00:05:52.808>a <00:05:52.948>ten-step <00:05:53.639>workflow <00:05:54.256>with <00:05:54.528>messy <00:05:54.988>company <00:05:55.488>data?

61
00:05:57.221 --> 00:06:01.987
<v Bill>Unknown—but <00:05:57.995>at <00:05:58.101>least <00:05:58.432>this <00:05:58.768>test <00:05:59.128>gives <00:05:59.408>us <00:05:59.608>a <00:05:59.728>real <00:06:00.128>place <00:06:00.581>to <00:06:00.741>start <00:06:01.099>asking.

62
00:06:01.987 --> 00:06:09.248
<v Cassidy>And <00:06:02.381>there’s <00:06:02.581>an <00:06:02.745>actual <00:06:03.168>14-megabyte <00:06:04.176>technical <00:06:04.661>document <00:06:05.159>attached, <00:06:06.080>with <00:06:06.235>an <00:06:06.315>eight-hour <00:06:06.761>evaluation <00:06:07.528>and <00:06:07.701>baseline <00:06:08.201>comparisons.

63
00:06:09.368 --> 00:06:17.728
<v Cassidy>After <00:06:09.988>the <00:06:10.128>Orbformer <00:06:10.741>paper <00:06:11.035>we <00:06:11.136>just <00:06:11.378>covered, <00:06:12.272>this <00:06:12.608>is <00:06:12.808>the <00:06:12.990>standard: <00:06:14.256>show <00:06:14.468>the <00:06:14.595>methodology <00:06:15.368>and <00:06:15.538>numbers <00:06:15.941>so <00:06:16.101>we <00:06:16.228>can <00:06:16.341>interrogate <00:06:16.888>the <00:06:17.088>claim.

64
00:06:18.618 --> 00:06:27.538
<v Cassidy>Ning-Zheng <00:06:19.231>Li; <00:06:19.818>Zi-Yu <00:06:20.565>Li; <00:06:21.178>Qiang <00:06:21.778>Shi; <00:06:22.618>Qing-Yu <00:06:23.278>Liu; <00:06:23.925>Sheng-Gui <00:06:24.991>He, <00:06:25.428>writing <00:06:26.085>in <00:06:26.184>Nature <00:06:26.527>Communications:

65
00:06:27.920 --> 00:06:34.880
<v Quote>Based <00:06:28.293>on ≈ <00:06:29.040>93,000 <00:06:30.149>global-minimum <00:06:31.011>structures <00:06:31.573>obtained <00:06:32.060>via <00:06:32.304>automated <00:06:32.933>first-principles <00:06:33.742>calculations

66
00:06:34.940 --> 00:06:45.920
<v Quote>our <00:06:35.187>model—the <00:06:35.900>cluster-transformer-encoder <00:06:37.500>network—enables <00:06:38.773>reliable <00:06:39.240>predictions <00:06:39.733>of <00:06:39.888>atomization <00:06:40.622>energies (≈ <00:06:42.240>40 <00:06:42.613>meV <00:06:43.080>per <00:06:43.264>atom <00:06:43.591>accuracy) <00:06:44.510>for <00:06:44.880>9.13

67
00:06:45.990 --> 00:06:48.320
<v Quote>million <00:06:46.533>metal <00:06:46.850>cluster <00:06:47.262>compositions

68
00:06:48.364 --> 00:06:55.360
<v Quote>covering <00:06:48.800>30 <00:06:49.240>d-block <00:06:49.749>metals <00:06:50.220>and <00:06:50.480>4 <00:06:50.784>chemically <00:06:51.253>relevant <00:06:51.740>ligand <00:06:52.133>elements <00:06:52.760>(C <00:06:53.560>N <00:06:54.260>O <00:06:54.480>and <00:06:54.920>S).

69
00:06:55.645 --> 00:07:04.205
<v Cassidy>Ninety-three <00:06:56.187>thousand <00:06:56.733>automated <00:06:57.205>first-principles <00:06:57.947>calculations <00:06:58.712>to <00:06:58.865>map <00:06:59.565>9.13 <00:07:00.575>million <00:07:00.965>cluster <00:07:01.313>compositions—that’s <00:07:03.055>the <00:07:03.135>compute <00:07:03.538>appetite <00:07:03.999>people

70
00:07:04.285 --> 00:07:07.325
<v Cassidy>miss <00:07:04.621>when <00:07:04.717>they <00:07:04.891>reduce <00:07:05.205>the <00:07:05.352>whole <00:07:05.552>buildout <00:07:06.072>to <00:07:06.189>chatbot <00:07:06.669>inference.

71
00:07:07.437 --> 00:07:10.179
<v Cassidy>Materials <00:07:08.505>science <00:07:08.952>is <00:07:09.045>pulling <00:07:09.352>on <00:07:09.425>the <00:07:09.533>grid <00:07:09.825>too.

72
00:07:10.179 --> 00:07:20.045
<v Bill>And <00:07:10.493>this <00:07:10.738>is <00:07:10.885>a <00:07:10.987>sensible <00:07:11.565>use <00:07:11.858>of <00:07:12.152>ML: <00:07:13.392>spend <00:07:13.765>the <00:07:13.949>expensive <00:07:14.585>quantum-compute <00:07:15.542>budget <00:07:15.992>on <00:07:16.365>93,000 <00:07:17.794>global-minimum <00:07:18.743>structures

73
00:07:20.093 --> 00:07:25.085
<v Bill>then <00:07:20.405>use <00:07:20.845>a <00:07:20.972>composition <00:07:21.552>model <00:07:21.925>for <00:07:22.065>the <00:07:22.192>first <00:07:22.605>pass <00:07:22.982>across <00:07:23.485>30 <00:07:23.885>d-block <00:07:24.445>metals.

74
00:07:25.192 --> 00:07:31.245
<v Bill>Forty <00:07:25.858>meV <00:07:26.345>per <00:07:26.621>atom <00:07:27.112>is <00:07:27.365>a <00:07:27.439>number <00:07:27.725>a <00:07:27.853>chemist <00:07:28.345>can <00:07:28.592>argue <00:07:28.973>with, <00:07:29.592>which <00:07:29.805>I <00:07:29.965>respect.

75
00:07:31.291 --> 00:07:40.365
<v Cassidy>Nature <00:07:31.618>Communications <00:07:32.425>has <00:07:32.625>the <00:07:32.738>methodology <00:07:33.387>attached, <00:07:34.333>plus <00:07:34.565>a <00:07:34.651>stated <00:07:35.005>accuracy <00:07:35.685>and <00:07:35.845>clear <00:07:36.138>scope: <00:07:37.125>small <00:07:37.378>clusters <00:07:37.837>with <00:07:38.165>C, <00:07:38.565>N, <00:07:38.965>O, <00:07:39.205>and <00:07:39.485>S <00:07:39.725>ligands.

76
00:07:40.525 --> 00:07:46.406
<v Cassidy>Same <00:07:41.205>pattern <00:07:41.565>we <00:07:41.765>saw <00:07:41.917>with <00:07:42.125>Orbformer—science <00:07:43.782>claims <00:07:44.157>with <00:07:44.319>enough <00:07:44.582>detail <00:07:44.978>to <00:07:45.129>actually <00:07:45.547>interrogate.

77
00:07:46.406 --> 00:07:53.805
<v Bill>Don’t <00:07:46.829>turn <00:07:47.165>9.13 <00:07:48.425>million <00:07:48.905>predictions <00:07:49.693>into <00:07:50.285>9.13 <00:07:51.705>million <00:07:52.197>validated <00:07:52.925>materials.

78
00:07:53.981 --> 00:07:56.845
<v Bill>This <00:07:54.765>is <00:07:55.005>first-pass <00:07:55.965>triage.

79
00:07:56.985 --> 00:08:03.005
<v Bill>But <00:07:57.458>if <00:07:57.592>it <00:07:57.725>reliably <00:07:58.425>narrows <00:07:58.925>a <00:07:59.062>search <00:07:59.437>that <00:07:59.629>used <00:07:59.912>to <00:08:00.045>eat <00:08:00.354>months <00:08:00.792>of <00:08:00.912>first-principles <00:08:01.955>compute

80
00:08:03.045 --> 00:08:08.045
<v Bill>the <00:08:03.245>cost <00:08:03.618>curve <00:08:03.955>changes <00:08:04.525>long <00:08:04.902>before <00:08:05.314>anyone <00:08:05.789>puts <00:08:06.085>a <00:08:06.185>new <00:08:06.445>catalyst <00:08:07.165>in <00:08:07.365>a <00:08:07.475>reactor.

81
00:08:09.861 --> 00:08:13.221
<v Cassidy>If <00:08:09.981>you <00:08:10.147>follow <00:08:10.528>AI’s <00:08:10.958>impact <00:08:11.354>on <00:08:11.514>business <00:08:11.961>and <00:08:12.079>technology

82
00:08:13.241 --> 00:08:15.941
<v Cassidy>you <00:08:13.341>may <00:08:13.525>also <00:08:13.848>enjoy <00:08:14.201>The <00:08:14.325>Data <00:08:14.638>Center <00:08:15.048>Daily

83
00:08:15.981 --> 00:08:18.501
<v Cassidy>a <00:08:16.074>daily <00:08:16.394>briefing <00:08:16.901>on <00:08:17.194>AI <00:08:17.531>compute

84
00:08:18.547 --> 00:08:21.541
<v Cassidy>hyperscaler <00:08:19.221>capex <00:08:20.121>the <00:08:20.488>power <00:08:20.869>grid

85
00:08:21.592 --> 00:08:22.821
<v Cassidy>semiconductor <00:08:22.330>supply

86
00:08:22.861 --> 00:08:26.821
<v Cassidy>and <00:08:23.038>energy <00:08:23.451>markets. <00:08:24.245>Find <00:08:24.848>it <00:08:24.945>wherever <00:08:25.341>you <00:08:25.495>listen <00:08:25.754>to <00:08:25.950>podcasts.

87
00:08:27.440 --> 00:08:31.600
<v Cassidy>Links <00:08:27.707>to <00:08:27.813>every <00:08:28.147>story <00:08:28.520>are <00:08:28.667>in <00:08:28.740>the <00:08:28.864>show <00:08:29.187>notes, <00:08:29.653>if <00:08:29.780>you <00:08:29.872>want <00:08:30.053>to <00:08:30.200>dig <00:08:30.352>into <00:08:30.533>anything <00:08:30.980>you <00:08:31.120>heard.

88
00:08:31.714 --> 00:08:34.400
<v Cassidy>Thanks <00:08:32.420>for <00:08:32.528>listening, <00:08:33.220>and <00:08:33.307>we’ll <00:08:33.493>be <00:08:33.632>back <00:08:33.822>tomorrow.

89
00:08:34.544 --> 00:08:36.880
<v Cassidy>That’s <00:08:35.307>AI <00:08:35.573>Daily <00:08:35.880>Briefing <00:08:36.280>for <00:08:36.467>today.

90
00:08:37.056 --> 00:08:39.360
<v Cassidy>This <00:08:37.813>is <00:08:37.960>a <00:08:38.050>Lantern <00:08:38.500>Podcast.