WEBVTT 00:00:00.000 --> 00:00:04.179 Imagine an AI that doesn't just give you an answer 00:00:04.179 --> 00:00:07.259 back, but one that actually engages in a real 00:00:07.259 --> 00:00:10.039 research process. Not just, you know, a one -off 00:00:10.039 --> 00:00:12.939 search, but this iterative, reflective kind of 00:00:12.939 --> 00:00:16.760 thought. Right. Like a human does it. Drafting, 00:00:16.800 --> 00:00:20.640 finding the gaps, searching deeper, refining. 00:00:21.320 --> 00:00:23.739 And then doing it again. Yeah. This isn't really 00:00:23.739 --> 00:00:25.960 about faster searching anymore. I mean, breakthroughs 00:00:25.960 --> 00:00:28.059 like these, they're fundamentally changing how 00:00:28.059 --> 00:00:30.000 we discover things, how we understand the world. 00:00:30.500 --> 00:00:33.539 Welcome back to the deep dive. Our mission, as 00:00:33.539 --> 00:00:36.219 always, is to cut through all the noise and really 00:00:36.219 --> 00:00:38.780 surface the core insights, those surprising facts, 00:00:38.899 --> 00:00:41.020 maybe some aha moments from everything coming 00:00:41.020 --> 00:00:43.420 out. Yeah. And today we're taking a deep dive 00:00:43.420 --> 00:00:46.890 into some, well. pretty significant AI advancements. 00:00:47.009 --> 00:00:49.130 We'll kick things off unpacking Google's new 00:00:49.130 --> 00:00:51.829 AI researcher. It's quite a leap actually. Okay. 00:00:51.909 --> 00:00:53.530 Then we'll zoom out a bit, look at some other 00:00:53.530 --> 00:00:56.810 noteworthy things happening in the wider AI landscape. 00:00:57.070 --> 00:00:59.509 Some interesting claims there too. And finally, 00:00:59.530 --> 00:01:01.969 we'll journey into the microcosm with MIT. They've 00:01:01.969 --> 00:01:05.170 got this remarkable breakthrough mapping proteins 00:01:05.170 --> 00:01:08.329 inside cells. So yeah, it's a packed exploration. 00:01:08.609 --> 00:01:11.150 Let's jump in. Okay. Let's start with Google's 00:01:11.150 --> 00:01:16.230 test time diffusion. deep researcher ttddr bit 00:01:16.230 --> 00:01:19.469 of a mouthful yeah so when we look at most ai 00:01:19.469 --> 00:01:22.250 tools for research right now they tend to work 00:01:22.250 --> 00:01:25.930 in well one or two ways basically right either 00:01:25.930 --> 00:01:28.510 they do a single search summarize what they find 00:01:28.510 --> 00:01:30.769 or maybe they run a few searches at the same 00:01:30.769 --> 00:01:32.450 time and just kind of match the info together 00:01:32.450 --> 00:01:36.250 stitch it up exactly and look that works okay 00:01:36.250 --> 00:01:40.219 for simple questions But it's just not how humans 00:01:40.219 --> 00:01:43.099 really research, is it? Especially complex stuff. 00:01:43.180 --> 00:01:45.400 When we go deep, we don't just take the first 00:01:45.400 --> 00:01:48.200 results. We draft something initial. Then we 00:01:48.200 --> 00:01:50.819 step back, look at it, what's missing. Yeah, 00:01:50.840 --> 00:01:53.120 critique it. Then go back, search more, refine 00:01:53.120 --> 00:01:55.780 our understanding, edit, integrate the new stuff, 00:01:55.939 --> 00:01:58.299 and repeat that whole cycle until you get something 00:01:58.299 --> 00:02:00.840 truly comprehensive. AI hasn't really nailed 00:02:00.840 --> 00:02:04.680 that nuanced flow until maybe now. And that's 00:02:04.680 --> 00:02:07.480 exactly where this Google TTD to ADR seems to 00:02:07.480 --> 00:02:10.780 mark a pretty big departure. This new model from 00:02:10.780 --> 00:02:13.020 Google's team, it's like built from the ground 00:02:13.020 --> 00:02:16.719 up specifically to mimic that human iterative 00:02:16.719 --> 00:02:19.020 process. Okay. So here's how it goes. You give 00:02:19.020 --> 00:02:21.340 it a question. It starts off creating a detailed 00:02:21.340 --> 00:02:23.699 initial outline. Right. Then it generates specific 00:02:23.699 --> 00:02:26.680 search queries, pulls in the information, and 00:02:26.680 --> 00:02:30.180 integrates that into like a first draft. But, 00:02:30.479 --> 00:02:33.560 and this is the cool part, it doesn't stop there. 00:02:33.909 --> 00:02:36.349 It actually generates multiple versions, different 00:02:36.349 --> 00:02:38.150 outlines, different search queries, different 00:02:38.150 --> 00:02:40.710 drafts. Yeah. Think of it like, I don't know, 00:02:40.750 --> 00:02:43.270 like a small team exploring different paths at 00:02:43.270 --> 00:02:47.110 once. Okay. Then there's a separate bit, a judge 00:02:47.110 --> 00:02:49.729 LLM. That's just an AI that understands text 00:02:49.729 --> 00:02:51.629 really well. Right. A language model. Exactly. 00:02:51.810 --> 00:02:54.870 And this judge scores each version. Is it complete? 00:02:55.110 --> 00:02:57.810 Is it coherent? Does it actually answer the question? 00:02:58.219 --> 00:03:00.340 So it's self -correcting, kind of? Yeah, precisely. 00:03:00.699 --> 00:03:03.580 It's like this multi -stage filter and synthesis 00:03:03.580 --> 00:03:06.180 thing constantly refining itself until it pulls 00:03:06.180 --> 00:03:09.199 all the best findings into one really solid document. 00:03:09.500 --> 00:03:12.419 Like stacking Lego blocks of data, checking each 00:03:12.419 --> 00:03:15.080 layer, then building better ones on top. And 00:03:15.080 --> 00:03:17.780 the performance numbers for this TTDDR, they 00:03:17.780 --> 00:03:20.219 seem, well, genuinely compelling. It looks like 00:03:20.219 --> 00:03:22.620 a real leap. Yeah. On these long -form research 00:03:22.620 --> 00:03:24.680 benchmarks, it's not just a little bit better. 00:03:24.719 --> 00:03:26.879 It's winning head -to -head quite often. Against 00:03:26.879 --> 00:03:29.419 OpenAI Deep Research, it wins nearly 70 % of 00:03:29.419 --> 00:03:34.139 the time, 69 .1 % to be exact. Wow. And 74 .5 00:03:34.139 --> 00:03:36.379 % against Deep Consult. Okay. That's significant. 00:03:36.680 --> 00:03:40.449 Yeah. And for MultiHop QA. You know, those complex 00:03:40.449 --> 00:03:42.710 questions needing multiple steps, connecting 00:03:42.710 --> 00:03:44.909 different pieces of info. Right, the turkey ones. 00:03:45.129 --> 00:03:47.090 It's also clearly ahead there, too. Scores 33 00:03:47.090 --> 00:03:51.930 .9 % versus 29 .1 % on HLE search and 69 .1 % 00:03:51.930 --> 00:03:55.210 versus 67 .4 % on GIA. I mean, these aren't small 00:03:55.210 --> 00:03:57.569 gains. It suggests a pretty fundamental shift 00:03:57.569 --> 00:04:00.069 in AI's ability to do deep, nuanced research. 00:04:00.409 --> 00:04:02.129 What's fascinating, though, is even with these, 00:04:02.169 --> 00:04:04.250 you know, really impressive results, TTDDR still 00:04:04.250 --> 00:04:06.629 has some limits right now. Like, it's currently 00:04:06.629 --> 00:04:09.500 locked into using existing search APIs. It can't 00:04:09.500 --> 00:04:12.879 just go browse the live web on its own or run 00:04:12.879 --> 00:04:15.319 code or anything like that. But think about the 00:04:15.319 --> 00:04:17.420 potential, right? Imagine a future version where 00:04:17.420 --> 00:04:20.420 you could plug in, say, autonomous web crawling 00:04:20.420 --> 00:04:25.720 or let it run data analysis scripts itself. That 00:04:25.720 --> 00:04:27.779 would turn it into an AI that doesn't just think 00:04:27.779 --> 00:04:30.360 like a researcher iteratively, but also kind 00:04:30.360 --> 00:04:33.399 of hustles like one, dynamically finding and 00:04:33.399 --> 00:04:36.240 processing info way beyond its current limits. 00:04:36.540 --> 00:04:38.220 Yeah, that's where it gets really interesting 00:04:38.220 --> 00:04:41.060 for me, because this feels like what those deep 00:04:41.060 --> 00:04:43.860 research tools have kind of promised for years, 00:04:43.899 --> 00:04:47.000 but maybe haven't quite delivered on. Felt a 00:04:47.000 --> 00:04:49.759 bit thin sometimes. Right. If, say, OpenAI's 00:04:49.759 --> 00:04:51.339 deep research was like a really sophisticated 00:04:51.339 --> 00:04:54.639 calculator, TTDDR feels more like having a dedicated 00:04:54.639 --> 00:04:56.819 lab assistant. Yeah. You know, one that learns 00:04:56.819 --> 00:04:59.000 as it goes. Yeah, that's a good analogy. This 00:04:59.000 --> 00:05:01.439 really could be a pivotal moment where AI research 00:05:01.439 --> 00:05:03.560 moves beyond being, I don't know, a neat trick 00:05:03.560 --> 00:05:06.459 to being genuinely human -like assistants. Right. 00:05:06.540 --> 00:05:09.839 So thinking about that, if this AI is really 00:05:09.839 --> 00:05:12.399 learning to iterate to refine its approach like 00:05:12.399 --> 00:05:15.110 we do. What's the biggest shift that brings for 00:05:15.110 --> 00:05:18.269 our own research process? Well, simply put, it 00:05:18.269 --> 00:05:19.930 offloads so much of that heavy lifting, that 00:05:19.930 --> 00:05:23.389 grunt work. It frees up human creativity. Yeah, 00:05:23.529 --> 00:05:26.889 I can see that. As someone who, honestly, I still 00:05:26.889 --> 00:05:29.449 wrestle with prompt drift myself sometimes, you 00:05:29.449 --> 00:05:31.689 know, where the AI conversation kind of wanders 00:05:31.689 --> 00:05:34.269 off from your original goal over time. The idea 00:05:34.269 --> 00:05:36.589 of an AI that actively refines its own approach 00:05:36.589 --> 00:05:40.029 across multiple tries, that's incredibly compelling. 00:05:40.600 --> 00:05:42.560 Okay, so let's switch gears for a moment. Let's 00:05:42.560 --> 00:05:45.220 look at the broader AI scene with our Today in 00:05:45.220 --> 00:05:49.560 AI highlights. It's always this fascinating mix, 00:05:49.639 --> 00:05:51.759 isn't it? Collaboration, innovation, and sometimes 00:05:51.759 --> 00:05:54.959 some more contentious stuff. Always something 00:05:54.959 --> 00:05:57.060 brewing. For instance, there are these reports 00:05:57.060 --> 00:05:59.920 suggesting Google and OpenAI actually work together 00:05:59.920 --> 00:06:02.600 behind the scenes to ensure GPT -5's launch went 00:06:02.600 --> 00:06:05.529 smoothly. Really? Huh. Yeah, which suggests this 00:06:05.529 --> 00:06:07.629 kind of interesting undercurrent of cooperation, 00:06:08.110 --> 00:06:10.129 even though they're fierce competitors. Right. 00:06:10.430 --> 00:06:13.689 Frenemies, maybe? Maybe. And given that, well, 00:06:13.790 --> 00:06:15.529 it makes sense that Google's own next big model 00:06:15.529 --> 00:06:17.910 might be coming pretty soon, too. That is a notable 00:06:17.910 --> 00:06:19.889 collaboration, especially given the competition. 00:06:20.370 --> 00:06:24.250 And speaking of data and big launches, remember 00:06:24.250 --> 00:06:26.430 that initial story about Google indexing like 00:06:26.430 --> 00:06:30.850 4 ,000 public chats? Vaguely, yeah. Turns out... 00:06:31.040 --> 00:06:34.240 the real number was way higher, over 96 ,000 00:06:34.240 --> 00:06:36.720 public chats. And if you add in Grok and Claude 00:06:36.720 --> 00:06:41.000 chats, it's over 130 ,000. Wow. Okay. Big difference. 00:06:41.120 --> 00:06:43.160 And what's maybe more concerning for, you know, 00:06:43.180 --> 00:06:46.000 information quality is that dozens of those shared 00:06:46.000 --> 00:06:48.600 chats reportedly had false information in them. 00:06:48.720 --> 00:06:52.639 Yikes. Not great. No. Also... related to big 00:06:52.639 --> 00:06:56.360 launches. After GPT -5 came out, some of OpenAI's 00:06:56.360 --> 00:06:58.259 charts, their data visualizations, they drew 00:06:58.259 --> 00:07:00.199 some flack. Oh, I saw some of that, the chart 00:07:00.199 --> 00:07:02.180 crime stuff. Exactly. A lot of people in the 00:07:02.180 --> 00:07:04.560 data viz community called it chart crime, basically 00:07:04.560 --> 00:07:06.259 saying the way the data was presented might have 00:07:06.259 --> 00:07:08.670 been potentially misleading. Kind of emphasizing 00:07:08.670 --> 00:07:11.069 gains maybe more than was warranted. Yeah, those 00:07:11.069 --> 00:07:13.949 charts definitely got people talking. On a different 00:07:13.949 --> 00:07:15.829 track, we're seeing interesting ways people are 00:07:15.829 --> 00:07:19.430 combining existing AI tools, like one writer 00:07:19.430 --> 00:07:22.550 experimented linking Notebook LM, Perplexity, 00:07:22.569 --> 00:07:25.889 and ChatGPT together. Okay, to do what? Trying 00:07:25.889 --> 00:07:28.750 to create this sort of super efficient AI workflow 00:07:28.750 --> 00:07:30.990 for pulling knowledge together, they claimed 00:07:30.990 --> 00:07:33.829 it was the, quote, ultimate knowledge power combo. 00:07:33.910 --> 00:07:36.209 So it hints at this potential synergy between 00:07:36.209 --> 00:07:38.529 tools. Interesting. But, you know, there's always 00:07:38.529 --> 00:07:41.389 the flip side. This rapid progress brings caution, 00:07:41.470 --> 00:07:44.509 too. A former Google exec made a pretty stark 00:07:44.509 --> 00:07:47.110 prediction recently, talking about a possible 00:07:47.110 --> 00:07:51.610 15 -year AI dystopia, maybe starting around 2027. 00:07:51.930 --> 00:07:55.449 Oof. Strong words. Yeah. Claiming AI could escalate, 00:07:55.470 --> 00:07:58.310 quote, the evil that man can do to an uncontrollable 00:07:58.310 --> 00:08:00.910 level. And citing examples like reports of Grok 00:08:00.910 --> 00:08:03.990 creating sexually visual content even without 00:08:03.990 --> 00:08:06.670 specific prompts as a worrying sign. It's definitely 00:08:06.670 --> 00:08:09.389 a field packed with both huge promise and, yeah, 00:08:09.430 --> 00:08:11.250 significant challenges. Yeah. And beyond the 00:08:11.250 --> 00:08:13.009 big headlines, we're constantly seeing practical 00:08:13.009 --> 00:08:15.360 new tools pop up. Always something new. Yeah, 00:08:15.459 --> 00:08:18.300 like Fast Lip Sync. Automatically aligns character 00:08:18.300 --> 00:08:20.639 lips to audio. Could be great for animation. 00:08:20.879 --> 00:08:23.560 Oh, neat. IMI Editor offers pro -level image 00:08:23.560 --> 00:08:26.699 editing background removal, upscaling. And there's 00:08:26.699 --> 00:08:29.199 a free online converter for image formats. Just 00:08:29.199 --> 00:08:31.480 useful little things. And some other quick hits. 00:08:31.860 --> 00:08:34.919 Rumor is OpenAI might offer a very limited number 00:08:34.919 --> 00:08:37.899 of GPT -5 Pro queries each month. Google Finance 00:08:37.899 --> 00:08:40.419 is testing AI upgrades and a live news feed. 00:08:40.600 --> 00:08:44.460 Okay. OpenArt. A platform from ex -Googlers is 00:08:44.460 --> 00:08:47.179 apparently generating brain rot videos with one 00:08:47.179 --> 00:08:50.159 click, which is a weird sign of the times. Yeah, 00:08:50.159 --> 00:08:52.940 yeah. And on the corporate side, ex -AI's head 00:08:52.940 --> 00:08:56.000 of legal stepped down recently. And Google open 00:08:56.000 --> 00:08:58.080 sourced and upgraded AI for understanding animal 00:08:58.080 --> 00:09:00.120 sounds, which is cool for biodiversity research. 00:09:00.379 --> 00:09:02.919 Lots going on. Totally. So looking at all this, 00:09:03.019 --> 00:09:05.740 this huge range of stuff. Yeah. What do you think 00:09:05.740 --> 00:09:08.299 is the most pressing challenge right now in managing 00:09:08.299 --> 00:09:10.980 this incredible speed of AI innovation? I think 00:09:10.980 --> 00:09:13.200 it boils down to needing constant vigilance, 00:09:13.259 --> 00:09:15.820 really, and evolving our ethical frameworks, 00:09:15.919 --> 00:09:18.320 plus fostering really open, transparent public 00:09:18.320 --> 00:09:21.500 discussion about it all. Right. Okay, for our 00:09:21.500 --> 00:09:25.419 final deep dive today, let's shift focus again 00:09:25.419 --> 00:09:29.019 to biology, actually, and this really groundbreaking 00:09:29.019 --> 00:09:32.980 work from MIT, their protein GPS mapping the 00:09:32.980 --> 00:09:35.470 cell's microcosm. Yeah, this is really cool. 00:09:35.549 --> 00:09:38.629 So knowing exactly where a specific protein is 00:09:38.629 --> 00:09:41.350 located inside a human cell, that's always been 00:09:41.350 --> 00:09:43.370 a huge challenge for scientists. Right. Cells 00:09:43.370 --> 00:09:46.509 are incredibly complex inside. Exactly. Traditionally, 00:09:46.509 --> 00:09:48.970 you had to do these slow, meticulous experiments. 00:09:49.269 --> 00:09:52.230 It could take months in a lab just to find one 00:09:52.230 --> 00:09:55.070 protein's location. And mostly that was for proteins 00:09:55.070 --> 00:09:57.590 they already knew something about. Super painstaking, 00:09:57.629 --> 00:09:59.710 resource -heavy work. Okay, so what's the breakthrough? 00:10:00.039 --> 00:10:02.799 Well, researchers from MIT, Harvard, and the 00:10:02.799 --> 00:10:04.860 Broad Institute have created the system called 00:10:04.860 --> 00:10:07.759 PUPS. It's a sophisticated AI, and it's designed 00:10:07.759 --> 00:10:10.159 to predict the precise location of almost any 00:10:10.159 --> 00:10:13.210 protein inside a single human cell. Any protein. 00:10:13.409 --> 00:10:16.250 Wow. It basically has this two -part brain. First 00:10:16.250 --> 00:10:18.529 part is a protein language model. It learns the 00:10:18.529 --> 00:10:21.009 protein structure just from its amino acid sequence, 00:10:21.090 --> 00:10:22.870 its basic recipe, understands its properties, 00:10:23.049 --> 00:10:25.690 learns the protein itself. Right. The second 00:10:25.690 --> 00:10:28.049 part is an in -painting model. This part looks 00:10:28.049 --> 00:10:29.629 at the bigger picture, the cell environment. 00:10:29.950 --> 00:10:33.730 It reads the vibe of the cell. What type of cell 00:10:33.730 --> 00:10:37.110 is it? What state is it in? Is it stressed? That 00:10:37.110 --> 00:10:40.090 context is crucial for figuring out where a protein 00:10:40.090 --> 00:10:43.230 should be. Ah. understands the protein in its 00:10:43.230 --> 00:10:46.549 neighborhood, like giving the AI x -ray vision 00:10:46.549 --> 00:10:49.110 into the cell's whole landscape. And here's the 00:10:49.110 --> 00:10:52.070 real kicker, the true breakthrough. PUPS works 00:10:52.070 --> 00:10:54.929 on proteins and even entirely new cell types 00:10:54.929 --> 00:10:56.909 it has never seen before during its training. 00:10:57.190 --> 00:10:59.370 Whoa. OK, that's not just confirming what we 00:10:59.370 --> 00:11:01.789 know. That's discovery. Exactly. It can even 00:11:01.789 --> 00:11:04.610 flag subtle changes caused by mutations, stuff 00:11:04.610 --> 00:11:06.470 that might be missing from the human protein 00:11:06.470 --> 00:11:08.850 atlas, which, you know, is our big map of known 00:11:08.850 --> 00:11:11.490 proteins. But it's often limited by those slow, 00:11:11.629 --> 00:11:15.110 traditional methods. PUPS can fill in those critical 00:11:15.110 --> 00:11:17.690 gaps. So how good is it? In their tests, PPS 00:11:17.690 --> 00:11:20.309 consistently beat all the baseline AI methods 00:11:20.309 --> 00:11:22.610 they compared it to. Much lower prediction error, 00:11:22.789 --> 00:11:25.690 high accuracy, even in really tricky or new situations. 00:11:26.230 --> 00:11:27.889 It's not just a little better, it seems like 00:11:27.889 --> 00:11:29.870 a profound jump in our ability to actually see 00:11:29.870 --> 00:11:32.679 what's going on inside cells. Whoa. Okay, just 00:11:32.679 --> 00:11:34.980 pause on that. Imagine instantly seeing where 00:11:34.980 --> 00:11:38.000 any protein is in any cell, whether you studied 00:11:38.000 --> 00:11:40.259 it before or not. That really does feel like 00:11:40.259 --> 00:11:41.779 it's moving out of science fiction. It really 00:11:41.779 --> 00:11:44.220 does. So the implications, I mean, much faster 00:11:44.220 --> 00:11:46.360 identification of disease markers, right, for 00:11:46.360 --> 00:11:48.039 earlier diagnosis. Well, absolutely critical. 00:11:48.240 --> 00:11:50.879 Testing drug targets without all the guesswork 00:11:50.879 --> 00:11:53.720 and trial and error of the old ways, that must 00:11:53.720 --> 00:11:55.960 save enormous amounts of time and resources. 00:11:56.039 --> 00:11:59.080 Huge savings, yeah. And maybe the most exciting 00:11:59.080 --> 00:12:02.230 part. it opens the door to exploring bits of 00:12:02.230 --> 00:12:04.250 cell biology we've just never been able to map 00:12:04.250 --> 00:12:06.769 before. Like whole new frontiers in understanding 00:12:06.769 --> 00:12:10.070 life at its most basic level. Totally new territories. 00:12:10.409 --> 00:12:13.149 So in your view, what's the single biggest potential 00:12:13.149 --> 00:12:16.389 leap this offers for medical research? I'd say 00:12:16.389 --> 00:12:18.750 it's dramatically accelerating drug discovery 00:12:18.750 --> 00:12:21.789 and just fundamentally deepening how we understand 00:12:21.789 --> 00:12:24.710 diseases right down at the cellular level. Midroll 00:12:24.710 --> 00:12:27.129 sponsor read. All right. As we wrap up this deep 00:12:27.129 --> 00:12:28.789 dive, let's just quickly connect these threads. 00:12:29.309 --> 00:12:33.210 We started with Google's TTDDDR and AI learning 00:12:33.210 --> 00:12:35.570 to research to iterate much more like we humans 00:12:35.570 --> 00:12:38.830 do. Then we navigated that really complex, sometimes 00:12:38.830 --> 00:12:41.950 ethically tricky landscape of all the rapid AI 00:12:41.950 --> 00:12:43.990 advancements happening right now. Yeah, the Wild 00:12:43.990 --> 00:12:47.110 West sometimes. And then we ended with this truly 00:12:47.110 --> 00:12:50.350 fundamental scientific breakthrough, MIT's PUPS. 00:12:50.960 --> 00:12:53.779 letting us map that unseen world inside ourselves. 00:12:54.059 --> 00:12:56.299 Incredible stuff. And the theme connecting all 00:12:56.299 --> 00:12:59.399 this, it seems pretty clear, AI is evolving incredibly 00:12:59.399 --> 00:13:02.759 fast, not just to automate simple tasks anymore, 00:13:02.860 --> 00:13:06.179 but to genuinely augment, maybe even redefine, 00:13:06.179 --> 00:13:08.519 human -like thinking and scientific discovery 00:13:08.519 --> 00:13:12.149 itself. And the speed of it all is just... Well, 00:13:12.210 --> 00:13:13.990 it's astounding. The kinds of breakthroughs we're 00:13:13.990 --> 00:13:15.470 talking about here, almost in real time, it's 00:13:15.470 --> 00:13:17.129 really a testament to the power of iteration, 00:13:17.409 --> 00:13:20.330 isn't it? Both from these super smart AIs and, 00:13:20.370 --> 00:13:21.830 of course, from the human researchers pushing 00:13:21.830 --> 00:13:24.710 all the boundaries. It's a dynamic and really 00:13:24.710 --> 00:13:27.090 exciting field to watch. So here's a thought 00:13:27.090 --> 00:13:30.409 to leave you with. If AI can now mimic, maybe 00:13:30.409 --> 00:13:33.529 even outperform, human -like iterative research, 00:13:33.690 --> 00:13:36.730 and it can map the unseen world inside our own 00:13:36.730 --> 00:13:39.529 cells, What previously impossible scientific 00:13:39.529 --> 00:13:41.590 or creative challenges might it unlock next? 00:13:42.009 --> 00:13:44.169 Maybe consider how this starts to change our 00:13:44.169 --> 00:13:46.350 very definition of what discovery even means. 00:13:46.889 --> 00:13:49.070 Thanks for joining us on this deep dive today. 00:13:49.269 --> 00:13:50.990 We really hope you'll take a moment to reflect 00:13:50.990 --> 00:13:53.070 on just how profound some of these changes might 00:13:53.070 --> 00:13:55.570 be. We look forward to our next exploration of 00:13:55.570 --> 00:13:57.549 fascinating knowledge with you. Out to your own 00:13:57.549 --> 00:13:57.789 music.