WEBVTT 00:00:00.000 --> 00:00:02.919 Okay, so picture this. Google's newest robot 00:00:02.919 --> 00:00:06.120 checks the local weather online, it sees rain, 00:00:06.320 --> 00:00:10.339 and then it decides all by itself to grab an 00:00:10.339 --> 00:00:12.720 umbrella and pack it in a bag for a trip. Wow. 00:00:13.220 --> 00:00:15.679 That's, well, that's some pretty advanced decision 00:00:15.679 --> 00:00:18.679 making, right? Using live web data in real time. 00:00:18.879 --> 00:00:21.519 It really is a genuinely impressive step for, 00:00:21.519 --> 00:00:24.120 you know, physical machines. Maybe not a total 00:00:24.120 --> 00:00:26.879 revolution, but definitely a big stride towards 00:00:26.879 --> 00:00:28.820 general intelligence out there in the messy real 00:00:28.820 --> 00:00:31.480 world. Welcome to the Deep Dive. Today, we're 00:00:31.480 --> 00:00:34.479 looking at the snapshot of AI's, frankly, incredibly 00:00:34.479 --> 00:00:37.679 fast maturation. We're pulling out the key insights 00:00:37.679 --> 00:00:39.359 from the source material you've shared with us. 00:00:39.479 --> 00:00:41.020 Yeah, and these sources, they really show that 00:00:41.020 --> 00:00:44.280 AI isn't just improving the challenges, the risks, 00:00:44.359 --> 00:00:46.460 they're scaling up dramatically too. So we've 00:00:46.460 --> 00:00:48.399 broken this down into three main parts for you. 00:00:48.479 --> 00:00:50.500 First up, the tech breakthroughs. We're talking 00:00:50.500 --> 00:00:52.820 about the shift to genuinely general purpose 00:00:52.820 --> 00:00:58.100 robots. Think laundry and logic. Then we really 00:00:58.100 --> 00:00:59.880 have to get into the tougher stuff, the high 00:00:59.880 --> 00:01:03.420 stakes world of security and, well, legal battles. 00:01:03.460 --> 00:01:06.239 We're talking actual bio threats emerging and 00:01:06.239 --> 00:01:10.680 these huge multibillion dollar lawsuits over 00:01:10.680 --> 00:01:13.739 AI training data. And finally, we'll dig into 00:01:13.739 --> 00:01:15.540 this really interesting report on what. being 00:01:15.540 --> 00:01:18.500 called the developer paradox it's fascinating 00:01:18.500 --> 00:01:20.939 data showing how developers are now super dependent 00:01:20.939 --> 00:01:24.180 on ai tools tools they openly say they don't 00:01:24.180 --> 00:01:26.939 entirely trust right it's like this hidden fragility 00:01:26.939 --> 00:01:29.079 inside a tool everyone's using for productivity 00:01:29.079 --> 00:01:31.700 exactly okay let's get into it starting in the 00:01:31.700 --> 00:01:33.540 lab all right let's ground this in the physical 00:01:33.540 --> 00:01:36.579 world first google deep mind they just showed 00:01:36.579 --> 00:01:40.140 off gemini robotics 1 .5 and i have to say this 00:01:40.140 --> 00:01:42.180 one feels like it could be the chat gpt moment 00:01:42.180 --> 00:01:44.560 but you know for machines that have to actually 00:01:44.620 --> 00:01:47.799 do things outside a clean lab. Exactly. That's 00:01:47.799 --> 00:01:49.959 the perfect way to put it. The absolute key here 00:01:49.959 --> 00:01:53.299 is handling, quote, real world messiness. We're 00:01:53.299 --> 00:01:55.439 finally moving beyond those super controlled 00:01:55.439 --> 00:01:57.920 single task setups. Yeah. And the sources gave 00:01:57.920 --> 00:02:00.420 some really concrete examples, complex ones that 00:02:00.420 --> 00:02:02.640 show some deep planning going on. So what were 00:02:02.640 --> 00:02:05.219 those examples? Tell us about them. Well, OK, 00:02:05.340 --> 00:02:07.900 folding laundry. Sure, we've seen robots fold 00:02:07.900 --> 00:02:10.419 laundry, but this wasn't just folding. It was 00:02:10.419 --> 00:02:13.240 classifying items, sorting them into different 00:02:13.240 --> 00:02:16.840 baskets based on color type. That takes planning. 00:02:17.229 --> 00:02:19.569 Multi -step thinking. Right. Not just a simple 00:02:19.569 --> 00:02:21.669 repetitive motion. And it's not just using its 00:02:21.669 --> 00:02:24.310 own internal logic either. Yeah. The really crucial 00:02:24.310 --> 00:02:27.770 bit, I think, is its ability to tap into Google 00:02:27.770 --> 00:02:31.030 search like MidTask. MidTask. How did that work? 00:02:31.250 --> 00:02:33.629 So they demoed it sorting recycling. But get 00:02:33.629 --> 00:02:36.719 this. It was sorting based on the specific. often 00:02:36.719 --> 00:02:40.000 really complicated recycling rules for that particular 00:02:40.000 --> 00:02:42.780 city. Rules that pulled directly from the web 00:02:42.780 --> 00:02:45.020 just minutes before starting the task. Okay, 00:02:45.099 --> 00:02:47.180 that's different. That's autonomous adaptation. 00:02:47.360 --> 00:02:50.139 It's using outside information on the fly. Precisely. 00:02:50.300 --> 00:02:52.659 Robots aren't isolated islands anymore. They 00:02:52.659 --> 00:02:54.800 can learn and adapt from the world's information 00:02:54.800 --> 00:02:57.229 in real time. And what's really interesting is 00:02:57.229 --> 00:03:00.009 the tech upgrade that makes this possible. Because 00:03:00.009 --> 00:03:02.689 the older Gemini robotics models, they could 00:03:02.689 --> 00:03:05.650 basically do one thing, one time. Change the 00:03:05.650 --> 00:03:07.789 lighting, move the object a bit, and boom, you 00:03:07.789 --> 00:03:09.930 often had to retrain the whole system. Yeah, 00:03:09.949 --> 00:03:12.469 that was the bottleneck. Now we've shifted to 00:03:12.469 --> 00:03:16.310 multi -step planning, and this is key, reusable 00:03:16.310 --> 00:03:19.689 motion logic. Reusable motion logic. Think of 00:03:19.689 --> 00:03:23.150 it like having basic Lego blocks of motor skills, 00:03:23.349 --> 00:03:27.080 fundamental movements. And the AI can now quickly 00:03:27.080 --> 00:03:30.280 stack and reconfigure these blocks for totally 00:03:30.280 --> 00:03:32.939 new goals. That's what gives it that real world 00:03:32.939 --> 00:03:34.979 flexibility. Okay. That makes sense. And this 00:03:34.979 --> 00:03:36.659 is where it potentially gets really transformative 00:03:36.659 --> 00:03:39.280 for robotics, right? This idea of motion transfer. 00:03:39.460 --> 00:03:41.699 Absolutely critical. Motion transfer. Yeah. It's 00:03:41.699 --> 00:03:44.280 the ability to take skills learned on one specific 00:03:44.280 --> 00:03:47.819 robot, say a simple industrial arm, learning 00:03:47.819 --> 00:03:51.210 to flip a pancake. And instantly apply that complex 00:03:51.210 --> 00:03:53.349 skill, that knowledge of pancake flipping, to 00:03:53.349 --> 00:03:55.169 a completely different robot. Like a humanoid 00:03:55.169 --> 00:03:56.870 bot. Totally different body, different joints. 00:03:57.110 --> 00:03:59.569 So you skip potentially thousands of hours of 00:03:59.569 --> 00:04:02.530 retraining for the new robot form. You just transfer 00:04:02.530 --> 00:04:05.210 the concept of the skill. Exactly. The concept, 00:04:05.289 --> 00:04:08.250 the logic. Whoa. Just imagine scaling that up. 00:04:08.349 --> 00:04:10.830 The future proofs training efforts massively. 00:04:11.490 --> 00:04:14.710 You teach one machine a complex task, like operating 00:04:14.710 --> 00:04:18.259 a drill press. And suddenly, every robot. regardless 00:04:18.259 --> 00:04:20.420 of its shape or size, can potentially do it. 00:04:20.519 --> 00:04:22.819 That could drastically cut down the time to mass 00:04:22.819 --> 00:04:25.519 adoption. That's a huge deal. So if the tech 00:04:25.519 --> 00:04:27.379 is getting this good, what's the big remaining 00:04:27.379 --> 00:04:30.019 hurdle? What does Google themselves still admit 00:04:30.019 --> 00:04:33.220 is preventing, say, mass consumer adoption right 00:04:33.220 --> 00:04:35.860 now? It still comes down to the sheer unpredictability 00:04:35.860 --> 00:04:38.040 of the real world. Yeah. Handling all the messy, 00:04:38.139 --> 00:04:40.600 unexpected edge cases. Yeah. That remains the 00:04:40.600 --> 00:04:42.160 biggest challenge. That makes total sense. The 00:04:42.160 --> 00:04:44.480 real world is, well, it's messy. Okay, let's 00:04:44.480 --> 00:04:46.600 pivot now, pretty hard, to the risks and the 00:04:46.600 --> 00:04:48.740 market side of things. Because it feels like 00:04:48.740 --> 00:04:50.939 the security and legal landscape is moving even 00:04:50.939 --> 00:04:53.480 faster than the core technology. It's almost 00:04:53.480 --> 00:04:56.019 dizzying, isn't it? So on the feature side, you 00:04:56.019 --> 00:05:00.779 saw OpenAI drop chat GPT pulse. This thing basically 00:05:00.779 --> 00:05:04.100 uses AI to cook up personalized briefings for 00:05:04.100 --> 00:05:06.480 you while you sleep. Right. Delivers them as 00:05:06.480 --> 00:05:08.060 these little interactive cards when you wake 00:05:08.060 --> 00:05:10.339 up. It's definitely pushing AI towards being 00:05:10.339 --> 00:05:12.680 like a personal chief of staff. And Meta jumped 00:05:12.680 --> 00:05:15.060 in with something called Vibes, which sounds 00:05:15.060 --> 00:05:18.589 like. basically a TikTok feed, but just for AI 00:05:18.589 --> 00:05:20.970 generated videos. Pretty much. And it encourages 00:05:20.970 --> 00:05:23.750 users to remix and share this stuff, you know, 00:05:23.750 --> 00:05:26.149 trying to capture that viral creative energy 00:05:26.149 --> 00:05:28.850 for AI content. So while the big guys are adding 00:05:28.850 --> 00:05:31.250 features, the sources also mention Elon Musk's 00:05:31.250 --> 00:05:33.550 XAI making a kind of interesting market move. 00:05:33.730 --> 00:05:36.069 Yeah. Undercutting the competition. offering 00:05:36.069 --> 00:05:38.009 their Grok model to the U .S. government for 00:05:38.009 --> 00:05:40.829 just 42 cents. Now, okay, it's framed partly 00:05:40.829 --> 00:05:43.329 as a joke, but it's also a very clear signal, 00:05:43.410 --> 00:05:45.689 right? A challenge to the pricing and dominance 00:05:45.689 --> 00:05:48.410 of OpenAI and Tropic, the established players. 00:05:49.600 --> 00:05:51.579 But while these market games are playing out, 00:05:51.620 --> 00:05:53.540 the actual security risks in the infrastructure 00:05:53.540 --> 00:05:56.800 are really popping up. Salesforce, for instance, 00:05:57.100 --> 00:06:00.120 just had a patch, a critical AI bug. This one 00:06:00.120 --> 00:06:02.279 involves something called prompt injection. Right. 00:06:02.399 --> 00:06:04.980 Prompt injection. So for anyone listening who 00:06:04.980 --> 00:06:07.620 isn't deep in the security weeds, that's basically 00:06:07.620 --> 00:06:11.680 tricking the AI. A hacker feeds it text, not 00:06:11.680 --> 00:06:13.660 as a normal question, but as a hidden command. 00:06:14.730 --> 00:06:17.129 To make the AI do something it shouldn't, like 00:06:17.129 --> 00:06:20.730 bypass its own safety rules or access data. And 00:06:20.730 --> 00:06:22.750 in the Salesforce case, it let attackers potentially 00:06:22.750 --> 00:06:25.250 steal valuable customer relationship management, 00:06:25.470 --> 00:06:28.470 CRM, data. It really highlights how traditional 00:06:28.470 --> 00:06:31.069 security like firewalls can fail when the AI 00:06:31.069 --> 00:06:34.290 itself is the weak point. The model can be inherently 00:06:34.290 --> 00:06:37.009 leaky. It's a massive risk for companies. But 00:06:37.009 --> 00:06:39.870 maybe the most... unsettling thing in these sources 00:06:39.870 --> 00:06:42.170 was the bio -threat angle. Yeah, that stood out. 00:06:42.250 --> 00:06:44.670 A Stanford lab apparently used AI to successfully 00:06:44.670 --> 00:06:48.110 design new working viruses. And crucially, some 00:06:48.110 --> 00:06:50.189 of these AI -designed viruses were reportedly 00:06:50.189 --> 00:06:53.170 stronger, more virulent than the natural ones 00:06:53.170 --> 00:06:55.310 they were based on. The report framed this as 00:06:55.310 --> 00:06:57.970 an immediate critical threat, and it specifically 00:06:57.970 --> 00:07:00.769 mentioned that U .S. response systems just aren't 00:07:00.769 --> 00:07:03.449 prepared for AI created bio threats, the speed, 00:07:03.550 --> 00:07:06.269 the accessibility. It completely changes the 00:07:06.269 --> 00:07:08.970 game for pathogen creation. It absolutely demands 00:07:08.970 --> 00:07:12.769 urgent global attention and regulation that, 00:07:12.889 --> 00:07:15.980 frankly. doesn't seem to exist yet. Now, shifting 00:07:15.980 --> 00:07:18.279 gears slightly to the legal side, the financial 00:07:18.279 --> 00:07:20.720 costs are becoming astronomical. No kidding. 00:07:20.920 --> 00:07:24.160 Anthropic, a major AI player, got hit with this 00:07:24.160 --> 00:07:27.860 potential $1 .5 billion wake -up call from authors 00:07:27.860 --> 00:07:30.420 suing over copyright. A billion and a half dollars. 00:07:30.680 --> 00:07:33.360 Yeah. That figure just underscores the colossal 00:07:33.360 --> 00:07:35.920 legal liability tied up in training these large 00:07:35.920 --> 00:07:38.240 language models on potentially copyrighted material. 00:07:38.560 --> 00:07:41.000 That number, $1 .5 billion, it signals we are 00:07:41.000 --> 00:07:43.139 definitely past the move fast and break thing. 00:07:43.240 --> 00:07:46.000 phase for AI development. Companies now have 00:07:46.000 --> 00:07:48.019 to factor in potentially massive legal settlements 00:07:48.019 --> 00:07:51.180 just as a cost of doing business, or they need 00:07:51.180 --> 00:07:53.399 to fundamentally rethink how they source training 00:07:53.399 --> 00:07:57.019 data. Yet, despite all this risk, the money keeps 00:07:57.019 --> 00:08:00.220 flowing into certain areas, particularly secure 00:08:00.220 --> 00:08:04.879 enterprise AI, Cohere. which focuses specifically 00:08:04.879 --> 00:08:07.779 on that secure niche. Right. They just raised 00:08:07.779 --> 00:08:10.699 another $100 million, hitting a $7 billion valuation. 00:08:11.399 --> 00:08:13.600 They're clearly betting that businesses will 00:08:13.600 --> 00:08:16.319 pay a premium for AI solutions that address these 00:08:16.319 --> 00:08:18.759 security vulnerabilities and legal landmines. 00:08:18.860 --> 00:08:21.180 Makes sense. So thinking about those two major 00:08:21.180 --> 00:08:23.339 risks, we just discussed the critical software 00:08:23.339 --> 00:08:26.360 bug at Salesforce along with data theft and the 00:08:26.360 --> 00:08:28.660 engineered viruses coming out of Stanford. Yeah. 00:08:28.699 --> 00:08:30.379 Which one holds the greater immediate threat 00:08:30.379 --> 00:08:32.620 potential? Based on the sources, the possibility 00:08:32.620 --> 00:08:36.039 of AI -borne bio -threats demands the most urgent 00:08:36.039 --> 00:08:39.940 global attention right now. Okay, before we dive 00:08:39.940 --> 00:08:41.559 into that developer trust paradox, let's just 00:08:41.559 --> 00:08:43.519 take a quick pause. Placeholder for sponsor message. 00:08:43.960 --> 00:08:46.480 Welcome back to the Deep Dive. So we really need 00:08:46.480 --> 00:08:48.200 to unpack the data coming out of Google Cloud's 00:08:48.200 --> 00:08:50.559 latest DORA report, because it paints this picture 00:08:50.559 --> 00:08:53.220 of what we're calling the core paradox. AI is 00:08:53.220 --> 00:08:55.519 becoming absolutely essential business infrastructure. 00:08:55.820 --> 00:08:58.500 Yet a significant chunk of the users who rely 00:08:58.500 --> 00:09:01.360 on it daily admit they don't fully trust what 00:09:01.360 --> 00:09:04.340 it produces. The numbers really do seem contradictory 00:09:04.340 --> 00:09:09.129 at first glance. Usage is... Well, almost universal 00:09:09.129 --> 00:09:12.169 now in development. 90 % of developers are using 00:09:12.169 --> 00:09:16.009 AI co -pilots regularly. 90%. Yeah. And often 00:09:16.009 --> 00:09:19.070 spending around two hours per day working alongside 00:09:19.070 --> 00:09:22.330 these AI assistants. It's deeply embedded in 00:09:22.330 --> 00:09:24.970 the workflow. Foundational. But then you hit 00:09:24.970 --> 00:09:29.269 the trust gap, and it's pretty stark. 30%, nearly 00:09:29.269 --> 00:09:32.289 one in three of those same developers, said they 00:09:32.289 --> 00:09:35.730 trust the AI's output only a little, or worse. 00:09:36.190 --> 00:09:38.690 Not at all. Think about that. A third of the 00:09:38.690 --> 00:09:40.970 workforce relies heavily on software they fundamentally 00:09:40.970 --> 00:09:43.470 distrust. It's incredibly relatable, though, 00:09:43.490 --> 00:09:45.549 isn't it? I mean, we all kind of experience this. 00:09:45.590 --> 00:09:47.490 Oh, absolutely. I still wrestle with prompt drift 00:09:47.490 --> 00:09:50.250 myself regularly. And, you know, for listeners 00:09:50.250 --> 00:09:52.190 maybe not living in the code daily, prompt drift, 00:09:52.350 --> 00:09:54.830 it's that really frustrating thing where during 00:09:54.830 --> 00:09:57.509 a long chat with an AI, it starts to forget the 00:09:57.509 --> 00:09:59.750 original instructions. It loses the context, 00:09:59.970 --> 00:10:02.110 the rules you set up at the very beginning. And 00:10:02.110 --> 00:10:05.169 that lack of reliable consistency, that's a genuine 00:10:05.169 --> 00:10:07.490 vulnerability when you're relying on it for professional 00:10:07.490 --> 00:10:10.710 work. So given that level of distrust, that drift. 00:10:11.500 --> 00:10:14.279 Why hasn't it killed adoption? The practical 00:10:14.279 --> 00:10:17.019 upsides must just be overwhelming that feeling. 00:10:17.240 --> 00:10:20.059 They really are. The data shows the benefits 00:10:20.059 --> 00:10:22.799 clearly win out over the fear of, you know, the 00:10:22.799 --> 00:10:25.639 AI making stuff up or getting things wrong. 80 00:10:25.639 --> 00:10:28.440 % of developers reported clear productivity gains. 00:10:28.980 --> 00:10:32.519 Huge number. 80%. Wow. And 59 % said it actually 00:10:32.519 --> 00:10:34.940 improved the quality of their code. So the overall 00:10:34.940 --> 00:10:37.659 feeling seems to be a very pragmatic, okay, I 00:10:37.659 --> 00:10:40.080 don't fully trust this thing, but there's absolutely 00:10:40.080 --> 00:10:41.740 no way I'm going back to working without it. 00:10:41.879 --> 00:10:44.580 That really is the definition of essential infrastructure, 00:10:44.860 --> 00:10:47.440 isn't it? It's somewhat fragile, maybe unreliable 00:10:47.440 --> 00:10:49.820 at times, but it's become irreplaceable. The 00:10:49.820 --> 00:10:52.700 speed boost is worth the extra effort of constantly 00:10:52.700 --> 00:10:55.779 checking its work. Exactly. And because this... 00:10:57.129 --> 00:10:59.429 slightly fragile tool is now critical infrastructure, 00:10:59.769 --> 00:11:02.330 you see companies like Google responding. They're 00:11:02.330 --> 00:11:04.929 pushing to standardize how it's used. They released 00:11:04.929 --> 00:11:07.610 this thing called the DORA AI capabilities model. 00:11:07.830 --> 00:11:09.909 Right. The DORA model. What does that actually 00:11:09.909 --> 00:11:12.830 do? It's basically a framework. It lays out seven 00:11:12.830 --> 00:11:16.169 best practices for development teams using AI 00:11:16.169 --> 00:11:18.850 effectively. Things like how to test AI outputs 00:11:18.850 --> 00:11:21.490 responsibly, integrate security checks, ensure 00:11:21.490 --> 00:11:24.250 consistent deployment processes, stuff like that. 00:11:24.330 --> 00:11:26.850 So it's a move towards governance. Putting guard 00:11:26.850 --> 00:11:30.029 rails in place. Precisely. It signals that companies 00:11:30.029 --> 00:11:31.909 get it now. They realize they can't just leave 00:11:31.909 --> 00:11:33.690 it up to individual developers to figure out 00:11:33.690 --> 00:11:35.669 how to manage all these risks and inconsistencies 00:11:35.669 --> 00:11:38.330 on their own. They need common standards. OK, 00:11:38.409 --> 00:11:41.029 so thinking about that push for standards, why 00:11:41.029 --> 00:11:43.690 is standardizing AI use? With frameworks like 00:11:43.690 --> 00:11:46.029 DORA, why is that so immediately critical for 00:11:46.029 --> 00:11:48.169 big organizations right now? Because essential 00:11:48.169 --> 00:11:51.110 AI infrastructure requires unified best practices 00:11:51.110 --> 00:11:53.690 to manage risk and output consistency. OK, so 00:11:53.690 --> 00:11:55.850 let's pull back for a moment and just recap the 00:11:55.850 --> 00:11:58.990 big ideas from this deep dive. It feels like 00:11:58.990 --> 00:12:02.370 we landed on three key takeaways that really 00:12:02.370 --> 00:12:05.769 define this complex, fast -moving moment in AI's 00:12:05.769 --> 00:12:07.649 development. Yeah, I think so. First, on the 00:12:07.649 --> 00:12:09.909 robotics front, Gemini showed us this really 00:12:09.909 --> 00:12:12.289 sophisticated multi -step planning. Right. Using 00:12:12.289 --> 00:12:14.950 live web data. That's a huge leap, right? Towards 00:12:14.950 --> 00:12:17.610 robots being genuinely general -purpose tools 00:12:17.610 --> 00:12:19.990 actually integrated into the real world. Right. 00:12:20.070 --> 00:12:23.669 Laundry and logic achieved. Second, the stakes 00:12:23.669 --> 00:12:26.009 around security and law are just skyrocketing. 00:12:26.009 --> 00:12:28.190 We've got the emergence of... potentially AI 00:12:28.190 --> 00:12:31.789 design bio threats on one hand and these absolutely 00:12:31.789 --> 00:12:35.269 massive multibillion dollar legal fights over 00:12:35.269 --> 00:12:37.450 copyrighted training data on the other. The risks 00:12:37.450 --> 00:12:39.889 are immense now. And finally, that developer 00:12:39.889 --> 00:12:42.970 paradox. AI adoption isn't really being driven 00:12:42.970 --> 00:12:45.450 by blind faith or trust. It's driven by hard 00:12:45.450 --> 00:12:48.960 results. productivity games developers are using 00:12:48.960 --> 00:12:50.840 these tools because they make them faster more 00:12:50.840 --> 00:12:52.960 effective even if they have to constantly double 00:12:52.960 --> 00:12:55.720 check the output so the tech is shifting incredibly 00:12:55.720 --> 00:12:59.080 fast from being this novel experimental thing 00:12:59.080 --> 00:13:02.320 to being critical yeah but still kind of fragile 00:13:02.320 --> 00:13:05.559 infrastructure It's a really fascinating and 00:13:05.559 --> 00:13:08.779 maybe precarious balance point we're at. So building 00:13:08.779 --> 00:13:10.940 on that, that advanced planning we saw with Gemini, 00:13:11.019 --> 00:13:13.139 you know, the robot checking the weather, deciding 00:13:13.139 --> 00:13:15.919 autonomously to pack an umbrella and connecting 00:13:15.919 --> 00:13:18.720 that capability with the escalating threats we 00:13:18.720 --> 00:13:20.600 discussed, like bio threats and the huge legal 00:13:20.600 --> 00:13:23.679 risks. We wanted to leave you, the listener, 00:13:23.879 --> 00:13:26.440 with this thought to chew on. Considering how 00:13:26.440 --> 00:13:28.740 complex and autonomous these AI systems are becoming, 00:13:28.899 --> 00:13:31.340 what do you think will be the first genuinely 00:13:31.340 --> 00:13:34.580 helpful, but also ethically challenging autonomous 00:13:34.580 --> 00:13:37.519 decision AI makes for humanity in the next year 00:13:37.519 --> 00:13:40.340 or so. Yeah, something complex, something vital, 00:13:40.379 --> 00:13:41.860 maybe something that feels just a little bit 00:13:41.860 --> 00:13:44.580 scary involving real world consequences. Was 00:13:44.580 --> 00:13:47.259 that first big autonomous ethically tricky choice 00:13:47.259 --> 00:13:50.000 going to be? Something to ponder. Thank you for 00:13:50.000 --> 00:13:51.840 joining us for this deep dive into your sources 00:13:51.840 --> 00:13:54.559 today. As always, we encourage you to keep exploring 00:13:54.559 --> 00:13:55.980 these incredibly important topics.