WEBVTT 00:00:00.000 --> 00:00:02.000 Are we really looking at an age of autonomous 00:00:02.000 --> 00:00:05.480 genius just around the corner? Or is the next 00:00:05.480 --> 00:00:08.199 decade actually going to be more about grinding 00:00:08.199 --> 00:00:10.099 through debugging? Yeah, that's the core tension, 00:00:10.259 --> 00:00:13.060 isn't it? I mean, since ChatGPT arrived, something 00:00:13.060 --> 00:00:16.140 like $14 trillion in valuation has flooded the 00:00:16.140 --> 00:00:18.480 tech market. Right. And that huge number, it's 00:00:18.480 --> 00:00:20.500 really built on this promise, this idea that 00:00:20.500 --> 00:00:23.420 AI agents are going to automate, well, almost 00:00:23.420 --> 00:00:26.739 everything. Exactly. But what if? What if that 00:00:26.739 --> 00:00:30.620 valuation is resting on tools that... maybe aren't 00:00:30.620 --> 00:00:33.020 quite ready. You know, tools that one of the 00:00:33.020 --> 00:00:35.880 sharpest minds in AI just called, pretty bluntly, 00:00:35.899 --> 00:00:38.600 slop. So today we're going to do a deep dive 00:00:38.600 --> 00:00:41.579 into that, into this recent reality check that 00:00:41.579 --> 00:00:44.100 seems to be hitting the AI world. We really want 00:00:44.100 --> 00:00:45.979 to focus on the technical honesty here. We'll 00:00:45.979 --> 00:00:48.460 kick off with Andrej Karpathy's critique, which 00:00:48.460 --> 00:00:51.060 honestly deflates a lot of that hype. Then we'll 00:00:51.060 --> 00:00:53.240 pivot to what's happening right now. Applications, 00:00:53.240 --> 00:00:56.119 the good, the bad. And maybe the con scary, like 00:00:56.119 --> 00:00:57.979 those viral deep fakes popping up everywhere. 00:00:59.229 --> 00:01:01.609 Definitely that. But we won't just stay there. 00:01:01.729 --> 00:01:04.890 We're also going to cover a truly mind -bending 00:01:04.890 --> 00:01:07.530 scientific development. Think artificial neurons 00:01:07.530 --> 00:01:12.390 finally learning to whisper to actual brain cells. 00:01:12.890 --> 00:01:15.209 Okay, so our goal here is to give you a clear 00:01:15.209 --> 00:01:18.209 view. Where does AI really stand today? We want 00:01:18.209 --> 00:01:20.590 to help separate the, let's say, the overhyped 00:01:20.590 --> 00:01:23.489 promises from the real, tangible, scientific 00:01:23.489 --> 00:01:25.730 steps forward. Let's get into it. Unpack this 00:01:25.730 --> 00:01:28.150 dichotomy. Okay, first stop has to be Andres 00:01:28.150 --> 00:01:31.170 Karpathy. And, you know, he's no outsider throwing 00:01:31.170 --> 00:01:34.609 stones. Ex -OpenAI co -founder, led AI at Tesla. 00:01:34.709 --> 00:01:36.530 Right, he's deep inside the field. Yeah. And 00:01:36.530 --> 00:01:39.109 he basically stood up and told everyone pumping 00:01:39.109 --> 00:01:42.859 the year of the AI agent. idea to maybe take 00:01:42.859 --> 00:01:45.159 a breath. Especially with that $14 trillion number 00:01:45.159 --> 00:01:47.260 hanging in the air, he thinks reliable, truly 00:01:47.260 --> 00:01:49.939 autonomous agents are, what, 10 years away? At 00:01:49.939 --> 00:01:51.420 least 10 years. That's his estimate. And yeah, 00:01:51.480 --> 00:01:53.560 10 years changes everything for strategies based 00:01:53.560 --> 00:01:55.599 on near -term automation. So what were his main 00:01:55.599 --> 00:01:57.560 points? What's the bottleneck? Well, he didn't 00:01:57.560 --> 00:02:00.840 mince words. Agentic coding, you know, the dream 00:02:00.840 --> 00:02:03.519 of AI running complex projects on its own. He 00:02:03.519 --> 00:02:06.120 called it SLOP. SLOP. And reinforcement learning, 00:02:06.260 --> 00:02:09.259 RL, which is... Pretty crucial for teaching agents 00:02:09.259 --> 00:02:13.560 complex behaviors. His word was terrible. You 00:02:13.560 --> 00:02:16.199 mentioned prompt drift earlier when we were prepping. 00:02:16.259 --> 00:02:18.080 For listeners, maybe explain what that means 00:02:18.080 --> 00:02:21.360 when an AI is trying a long task. Sure. So prompt 00:02:21.360 --> 00:02:24.740 drift. It's like the AI starts forgetting the 00:02:24.740 --> 00:02:26.680 original goal or the constraints you gave it 00:02:26.680 --> 00:02:28.780 as it works through steps. I wrestle with it 00:02:28.780 --> 00:02:31.259 myself, honestly. Just yesterday, trying to get 00:02:31.259 --> 00:02:33.300 a system to plan a trip. Three parts, right? 00:02:33.379 --> 00:02:36.289 Yeah. By step two, it... Totally forgot I said 00:02:36.289 --> 00:02:39.050 no airports. So in Carpeth, he calls it slop. 00:02:39.370 --> 00:02:41.610 Yeah, I kind of get it. That unreliability is 00:02:41.610 --> 00:02:44.050 real. Beat. And then he said something even more 00:02:44.050 --> 00:02:46.229 counterintuitive, I thought. That perfect memory 00:02:46.229 --> 00:02:49.150 in AI is actually a bug. Why is remembering everything 00:02:49.150 --> 00:02:51.650 bad? It sounds weird, right? But think about 00:02:51.650 --> 00:02:54.650 it. We don't want AI to just memorize facts like 00:02:54.650 --> 00:02:56.710 a database. That's overfitting. We want it to 00:02:56.710 --> 00:02:58.830 generalize to learn the underlying patterns or 00:02:58.830 --> 00:03:01.610 rules so it can handle new situations it hasn't 00:03:01.610 --> 00:03:04.610 seen before. Perfect memory can make it brittle. 00:03:04.939 --> 00:03:07.219 Okay, that makes sense. Learn the rules, not 00:03:07.219 --> 00:03:09.180 just the answers. That's some needed technical 00:03:09.180 --> 00:03:11.860 honesty. Of course, you know, Elon Musk immediately 00:03:11.860 --> 00:03:14.719 shot back, challenged Karpathy to a face -off 00:03:14.719 --> 00:03:17.900 with Grok 5, but that banter doesn't really address 00:03:17.900 --> 00:03:20.759 the core issue Karpathy raised, does it? The 00:03:20.759 --> 00:03:23.719 reliability gap. No, it doesn't. And that gap 00:03:23.719 --> 00:03:27.020 is the crux of it. Getting from, say, 99 % accuracy 00:03:27.020 --> 00:03:31.080 to 99 .9%, that last little bit can be just as 00:03:31.080 --> 00:03:33.560 hard, sometimes harder, than getting from 0 % 00:03:33.560 --> 00:03:36.699 to 90%. Right, because that last 0 .9 % covers 00:03:36.699 --> 00:03:39.460 all the weird edge cases, the unexpected stuff. 00:03:39.599 --> 00:03:41.639 The first 90 % is the more predictable part. 00:03:41.860 --> 00:03:43.780 Exactly. And for real -world automation, you 00:03:43.780 --> 00:03:46.460 need that super high reliability. You need to 00:03:46.460 --> 00:03:48.740 conquer those edge cases. So, probe in question 00:03:48.740 --> 00:03:50.560 time. If agents are hitting this reliability 00:03:50.560 --> 00:03:53.520 wall, is the main problem the underlying AI architecture 00:03:53.520 --> 00:03:56.039 itself, or is it just about feeding them more 00:03:56.039 --> 00:03:58.199 and better data? Karpathy's point suggests it's 00:03:58.199 --> 00:04:00.139 more fundamental. The big challenge is reliably 00:04:00.139 --> 00:04:02.740 scaling that final accuracy point. It's about 00:04:02.740 --> 00:04:04.659 the architecture handling novelty correctly. 00:04:04.979 --> 00:04:07.340 Okay, so it's more than just data. Right. It's 00:04:07.340 --> 00:04:09.599 about dependable performance in unpredictable 00:04:09.599 --> 00:04:13.240 scenarios. OK, shifting gears a bit, let's talk 00:04:13.240 --> 00:04:15.800 about what specialized AI is doing effectively 00:04:15.800 --> 00:04:19.420 right now. Anthropic, for example, just launched 00:04:19.420 --> 00:04:22.980 Claude for Life Sciences. Ah, so really niching 00:04:22.980 --> 00:04:25.279 down, not just general chat anymore. Exactly. 00:04:25.360 --> 00:04:27.839 This isn't just a chat bot. It offers like full 00:04:27.839 --> 00:04:30.540 stack research support. It integrates with actual 00:04:30.540 --> 00:04:33.759 lab tools. That's deep specialization. Impressive. 00:04:33.899 --> 00:04:36.240 And on the creative side, we saw Anthropic's 00:04:36.240 --> 00:04:39.920 head of dev relations create this insane PDF 00:04:39.920 --> 00:04:42.459 flipbook, apparently entirely through code prompts. 00:04:42.620 --> 00:04:45.879 It shows these things can handle complex, structured, 00:04:45.939 --> 00:04:48.220 creative work. That's the positive side. But 00:04:48.220 --> 00:04:51.839 that same tower. Well, it brings risks. The sources 00:04:51.839 --> 00:04:54.620 flagged this rise of viral AI slop. You mentioned 00:04:54.620 --> 00:04:56.980 it earlier. Videos like the one with dogs supposedly 00:04:56.980 --> 00:04:59.519 rescuing babies. Yeah, that one hit like 6 .3 00:04:59.519 --> 00:05:01.740 million views. And the scary part is most people 00:05:01.740 --> 00:05:03.560 watching apparently couldn't tell it was fake. 00:05:04.319 --> 00:05:07.720 That ability to fool us is getting really good 00:05:07.720 --> 00:05:11.160 really fast. It changes how we have to approach 00:05:11.160 --> 00:05:13.879 online content. And speaking of caution, there's 00:05:13.879 --> 00:05:17.360 this privacy aspect too. Meta Facebook has a 00:05:17.360 --> 00:05:20.319 new opt -in button. Oh? Yeah. It potentially 00:05:20.319 --> 00:05:22.959 lets their AI look at your unpublished photos. 00:05:23.180 --> 00:05:25.139 Okay, that definitely needs a big flashing warning 00:05:25.139 --> 00:05:28.000 sign. What's the risk there? Well, the risk is 00:05:28.000 --> 00:05:31.500 the AI starts learning things about you, connecting 00:05:31.500 --> 00:05:34.439 your face or your private life to data points 00:05:34.439 --> 00:05:36.699 you never, ever intended to make public. It just, 00:05:36.759 --> 00:05:38.500 you know, increases your digital vulnerability 00:05:38.500 --> 00:05:41.339 over time. Yeah. User awareness is key here. 00:05:41.699 --> 00:05:44.170 You need to know what you're opting into. Good 00:05:44.170 --> 00:05:46.209 warning. On the flip side, though, there are 00:05:46.209 --> 00:05:48.810 genuinely useful new tools. Perplexity -added 00:05:48.810 --> 00:05:51.149 language learning features, flashcards, translations, 00:05:51.490 --> 00:05:54.029 feedback. Looks handy. Yeah, practical stuff. 00:05:54.230 --> 00:05:56.129 And in medicine, this company, Open Evidence, 00:05:56.209 --> 00:05:58.810 think chat GPT specifically for doctors. They 00:05:58.810 --> 00:06:01.769 just raised $200 million. Wow. And they're apparently 00:06:01.769 --> 00:06:04.029 handling 15 million consultations a month already. 00:06:04.170 --> 00:06:06.189 That's a significant scale in a critical field. 00:06:06.470 --> 00:06:09.250 Okay, another probing question. Maybe for me 00:06:09.250 --> 00:06:12.810 this time, from you. Given how easily this fake 00:06:12.810 --> 00:06:15.829 media goes viral, how do we even begin to adjust 00:06:15.829 --> 00:06:18.290 our online skepticism? That's a tough one. Yeah, 00:06:18.370 --> 00:06:20.569 I think we almost have to default to disbelief 00:06:20.569 --> 00:06:23.660 now. Assume virality doesn't mean truth. Maybe 00:06:23.660 --> 00:06:25.699 the opposite sometimes. All right, let's talk 00:06:25.699 --> 00:06:27.939 practical workflows. If you're running your own 00:06:27.939 --> 00:06:30.339 show, a solopreneur, the sources had some interesting 00:06:30.339 --> 00:06:32.819 ideas about building a kind of virtual team using 00:06:32.819 --> 00:06:35.680 AI tools. Yeah, moving beyond just single -use 00:06:35.680 --> 00:06:38.060 apps. And for the more advanced users, there's 00:06:38.060 --> 00:06:40.459 this concept they call build your AI twin, basically 00:06:40.459 --> 00:06:43.759 a digital you for making videos. Okay, AI twin. 00:06:43.939 --> 00:06:46.899 The sources mentioned a three -tool method. Can 00:06:46.899 --> 00:06:49.439 you break down the principle behind that functionally? 00:06:49.579 --> 00:06:52.240 What are the Lego blocks, so to speak? Sure. 00:06:52.360 --> 00:06:54.939 It's like an assembly line. Tool one would handle 00:06:54.939 --> 00:06:57.779 your script, the logic, the words. Tool two is 00:06:57.779 --> 00:07:00.339 the visual engine, the part that generates the 00:07:00.339 --> 00:07:03.300 photorealistic video of you, syncing lip movements, 00:07:03.399 --> 00:07:06.680 expressions. Think deep fake tech, but you control 00:07:06.680 --> 00:07:09.439 it. Got it. And tool three handles cloning your 00:07:09.439 --> 00:07:11.550 voice. and maybe putting it all together for 00:07:11.550 --> 00:07:13.790 distribution. So yeah, stacking those blocks 00:07:13.790 --> 00:07:16.930 of data script, visuals, voice to create an avatar 00:07:16.930 --> 00:07:19.730 that looks and sounds like you. Pretty wild automation 00:07:19.730 --> 00:07:21.790 potential there. Which brings us back to agents. 00:07:22.269 --> 00:07:24.589 ChatGPT agent mode was mentioned. The idea is 00:07:24.589 --> 00:07:26.389 you give it a research task and it comes back 00:07:26.389 --> 00:07:28.230 with a finished report. And this is where Carpe 00:07:28.230 --> 00:07:30.529 these warnings echo, right? The sources gave 00:07:30.529 --> 00:07:32.370 four examples of how you could set this up now. 00:07:33.699 --> 00:07:35.939 automating complex knowledge work like that? 00:07:36.060 --> 00:07:38.339 Yeah. Well, the risk of the AI making stuff up 00:07:38.339 --> 00:07:40.759 or just getting it wrong is still high. Definitely. 00:07:40.920 --> 00:07:43.720 So thinking back to Karpathy's reliability warnings, 00:07:43.980 --> 00:07:46.199 if you are going to use something like agent 00:07:46.199 --> 00:07:48.699 mode for a serious project, what's the absolute 00:07:48.699 --> 00:07:51.660 number one safety measure? Constant human oversight, 00:07:51.879 --> 00:07:54.519 period. You absolutely have to review the output 00:07:54.519 --> 00:07:57.540 carefully. Don't trust, verify. Human in the 00:07:57.540 --> 00:08:00.379 loop is still essential. Okay. All right, let's 00:08:00.379 --> 00:08:02.470 do some quick hits. Rapid fire updates. Google 00:08:02.470 --> 00:08:05.069 put its Maps tool right into the Gemini API, 00:08:05.370 --> 00:08:08.529 so AI can now leverage real -world location data 00:08:08.529 --> 00:08:11.110 more easily. Interesting integration. Adobe launched 00:08:11.110 --> 00:08:13.889 a Foundry service, lets companies build their 00:08:13.889 --> 00:08:17.269 own custom gen AI models trained on their specific 00:08:17.269 --> 00:08:20.509 data. More bespoke AI. Makes sense. And market 00:08:20.509 --> 00:08:24.029 moves. Meta AI's app downloads reportedly jumped 00:08:24.029 --> 00:08:26.529 quite a bit after they rolled out that Vibes 00:08:26.529 --> 00:08:29.029 AI feed. Trying to get that engagement. And Anthropic 00:08:29.029 --> 00:08:31.389 made its coding assistant available as a web 00:08:31.389 --> 00:08:33.820 app. making it easier for people to access. Okay. 00:08:33.919 --> 00:08:36.019 And policy. There was something serious there, 00:08:36.039 --> 00:08:37.820 too. Yeah, a significant one. The Department 00:08:37.820 --> 00:08:40.720 of Homeland Security asked OpenAI to provide 00:08:40.720 --> 00:08:44.460 user data, basically, unmask users in a child 00:08:44.460 --> 00:08:47.769 abuse investigation. Wow. Beat. That really highlights 00:08:47.769 --> 00:08:50.210 the friction, doesn't it, between privacy expectations 00:08:50.210 --> 00:08:53.169 and law enforcement demands on these huge platforms. 00:08:53.309 --> 00:08:55.350 Absolutely. A major tension point that's only 00:08:55.350 --> 00:08:57.289 going to grow. Okay. Let's completely shift gears 00:08:57.289 --> 00:09:00.289 now, away from software agents and policy debates, 00:09:00.470 --> 00:09:03.429 towards something truly fundamental, this UMass 00:09:03.429 --> 00:09:06.389 Amherst breakthrough, artificial neurons talking 00:09:06.389 --> 00:09:09.490 to brain cells. Yes. This is really cool science. 00:09:09.710 --> 00:09:13.009 And the key word you used before was perfect. 00:09:13.800 --> 00:09:16.519 Right, because previous attempts were more like 00:09:16.519 --> 00:09:20.159 shouting. Exactly. Older artificial neurons needed 00:09:20.159 --> 00:09:23.240 way more voltage, like 10 times more. And they 00:09:23.240 --> 00:09:27.039 used maybe 100 times more power just blasting 00:09:27.039 --> 00:09:29.730 the biological cells, relatively speaking. But 00:09:29.730 --> 00:09:32.750 this new one gets down to 0 .1 volts. Around 00:09:32.750 --> 00:09:35.769 0 .1 volts, yeah, which is roughly the same energy 00:09:35.769 --> 00:09:38.210 level biological neurons use to communicate with 00:09:38.210 --> 00:09:40.490 each other. That's the game changer. It's finally 00:09:40.490 --> 00:09:42.629 speaking the same language, electrically speaking. 00:09:42.809 --> 00:09:45.490 That efficiency difference is huge. How did they 00:09:45.490 --> 00:09:48.440 achieve that gentle touch? What's the tech? The 00:09:48.440 --> 00:09:50.600 magic ingredient seems to be these things called 00:09:50.600 --> 00:09:54.139 protein nanowires. Protein nanowires. Yeah, fascinating 00:09:54.139 --> 00:09:57.379 stuff. They're actually grown by bacteria. Specific 00:09:57.379 --> 00:10:00.019 kinds of bacteria produce these tiny wires that 00:10:00.019 --> 00:10:02.399 naturally conduct electricity. Bacteria are building 00:10:02.399 --> 00:10:04.820 components for neural interfaces. Kind of mind 00:10:04.820 --> 00:10:06.779 -blowing, right? Yeah. And these nanowires have 00:10:06.779 --> 00:10:09.440 three huge advantages for interfacing with biology. 00:10:09.720 --> 00:10:12.120 Okay, lay them out for us. First, they're stable 00:10:12.120 --> 00:10:14.419 in wet, salty environments, like, you know, inside 00:10:14.419 --> 00:10:17.470 the body. That's critical. Second, they operated 00:10:17.470 --> 00:10:20.309 that ultra -low voltage, the 0 .1V we mentioned 00:10:20.309 --> 00:10:23.330 matching neurons. And third, because of that 00:10:23.330 --> 00:10:25.809 low voltage, they are incredibly energy efficient, 00:10:25.970 --> 00:10:28.970 like 100 times less power needed than the old 00:10:28.970 --> 00:10:31.590 approaches. Okay, hold on. Grown by bacteria, 00:10:31.830 --> 00:10:34.700 does that mean we could like... brew these things, 00:10:34.919 --> 00:10:37.200 is this potentially scalable in a way silicon 00:10:37.200 --> 00:10:40.000 isn't for biointerfaces? That's exactly the long 00:10:40.000 --> 00:10:42.039 -term potential people are excited about. This 00:10:42.039 --> 00:10:44.419 isn't just fiddly lab work. It's just a pathway 00:10:44.419 --> 00:10:47.139 to maybe producing these components using biological 00:10:47.139 --> 00:10:50.919 processes. Scale, reproducibility, whoa. I mean, 00:10:50.940 --> 00:10:53.419 just imagine scaling that gentle, precise communication 00:10:53.419 --> 00:10:56.659 without frying the tissue. That's the real moment 00:10:56.659 --> 00:10:59.340 of wonder here. Wow, okay. So that biological 00:10:59.340 --> 00:11:02.080 compatibility, that gentle whisper. It means 00:11:02.080 --> 00:11:04.019 for the first time, an artificial neuron can 00:11:04.019 --> 00:11:06.440 have a real two -way conversation with a biological 00:11:06.440 --> 00:11:09.100 one, not just stimulate it crudely. Precisely. 00:11:09.100 --> 00:11:12.419 It's communication, not just activation. And 00:11:12.419 --> 00:11:14.980 that opens up entirely new possibilities for 00:11:14.980 --> 00:11:17.960 neural interfaces. Think wearables, maybe even 00:11:17.960 --> 00:11:20.159 implants someday, that can interact with our 00:11:20.159 --> 00:11:22.600 nervous system much more naturally, much more 00:11:22.600 --> 00:11:26.200 seamlessly. Huge potential for therapies. OK, 00:11:26.299 --> 00:11:29.220 final probing question on this. Does this breakthrough, 00:11:29.399 --> 00:11:32.399 this biocompatibility autol, does it suggest 00:11:32.399 --> 00:11:35.539 that maybe neuro AI, the hardware interface side, 00:11:35.720 --> 00:11:39.059 could actually leapfrog the pure software agents 00:11:39.059 --> 00:11:41.820 Carpathy was critiquing? It's a fascinating thought. 00:11:41.980 --> 00:11:44.220 This foundational work certainly suggests neuro 00:11:44.220 --> 00:11:46.639 interfaces have very strong potential. They might 00:11:46.639 --> 00:11:49.240 bypass some software reliability hurdles. Definitely 00:11:49.240 --> 00:11:51.279 something to watch. So what a journey today. 00:11:51.419 --> 00:11:53.299 We've really seen two sides of the coin. On one 00:11:53.299 --> 00:11:56.490 side, you've got this. Massive $14 trillion AI 00:11:56.490 --> 00:11:59.370 hype machine. Yeah, built on agentic coding that, 00:11:59.429 --> 00:12:01.769 according to key insiders, is still kind of struggling. 00:12:01.870 --> 00:12:04.610 Still maybe slopping places. Requiring that constant 00:12:04.610 --> 00:12:06.990 human oversight for the immediate future. We're 00:12:06.990 --> 00:12:09.309 in the debugging grind, maybe. But then on the 00:12:09.309 --> 00:12:11.769 other side, there's this genuinely revolutionary 00:12:11.769 --> 00:12:16.110 science. The 0 .1 volt neuron whisperer. Foundational 00:12:16.110 --> 00:12:19.809 stuff. Totally different level. That UMass Amherst 00:12:19.809 --> 00:12:22.929 work points toward a future that might rely more 00:12:22.929 --> 00:12:26.570 on physical, deeply biocompatible interfaces, 00:12:26.850 --> 00:12:29.850 technology that talks with our biology, not just 00:12:29.850 --> 00:12:32.590 at our screens. It really makes you think. So 00:12:32.590 --> 00:12:34.009 here's a final thought, something for you, the 00:12:34.009 --> 00:12:36.350 listener, to chew on. If Karpathy's timeline 00:12:36.350 --> 00:12:39.500 is even close to right, If software AI faces 00:12:39.500 --> 00:12:42.080 a tough decade -long slog to get truly reliable, 00:12:42.379 --> 00:12:45.200 should the industry, the investment, maybe start 00:12:45.200 --> 00:12:47.740 shifting away from purely autonomous software 00:12:47.740 --> 00:12:50.759 dreams and more towards these biocompatible hardware 00:12:50.759 --> 00:12:53.620 breakthroughs like the whispering neuron? Is 00:12:53.620 --> 00:12:56.120 the real next frontier physical, not just digital? 00:12:56.240 --> 00:12:58.620 A really interesting question to ponder. Indeed. 00:12:58.700 --> 00:13:00.840 Well, thank you for joining us on this deep dive, 00:13:01.000 --> 00:13:03.360 navigating the hype and the breakthroughs. We 00:13:03.360 --> 00:13:05.039 definitely encourage you to dig into the source 00:13:05.039 --> 00:13:06.220 materials we touched on today.