WEBVTT 00:00:00.000 --> 00:00:02.759 Okay, let's unpack this. Our sources this week, 00:00:02.799 --> 00:00:05.059 they're pointing to something, well, really unsettling 00:00:05.059 --> 00:00:07.620 about how we perceive things. It seems like our 00:00:07.620 --> 00:00:10.640 own ears, you know, the tool we trust most for 00:00:10.640 --> 00:00:13.500 truth, they're starting to fail us. Yeah, it 00:00:13.500 --> 00:00:15.339 sounds pretty dramatic, but the data is backing 00:00:15.339 --> 00:00:18.640 it up. Exactly. We saw this statistic, really 00:00:18.640 --> 00:00:22.739 stunning, that 58 % of these advanced clone AI 00:00:22.739 --> 00:00:25.160 voices, people think they're real, that line's 00:00:25.160 --> 00:00:29.399 just... Welcome to this deep dive. And hey, if 00:00:29.399 --> 00:00:31.399 you feel like reality is sort of blending together 00:00:31.399 --> 00:00:34.619 faster than you can process, well, you're definitely 00:00:34.619 --> 00:00:37.100 not alone. That speed, that's what we're here 00:00:37.100 --> 00:00:39.859 to help manage for you. So today, we're taking 00:00:39.859 --> 00:00:41.619 all the source material you've got and boiling 00:00:41.619 --> 00:00:45.259 it down to three key things. First, we're going 00:00:45.259 --> 00:00:46.899 to look at just how bad we are now at telling 00:00:46.899 --> 00:00:48.780 real voices from synthetic ones. It's kind of 00:00:48.780 --> 00:00:51.420 shocking. Then we'll jump into the big tech battleground 00:00:51.420 --> 00:00:54.240 for AI, mapping out everything from like... The 00:00:54.240 --> 00:00:56.460 immediate tools companies are using to the really 00:00:56.460 --> 00:00:59.119 futuristic stuff like Neuralink. Right. And finally, 00:00:59.140 --> 00:01:01.939 we tackle this huge paradox. Why are we using 00:01:01.939 --> 00:01:04.480 AI tools twice as much but trusting them less 00:01:04.480 --> 00:01:06.819 and less for things like news? Our mission here, 00:01:06.959 --> 00:01:09.620 cut through all that noise and just pull out 00:01:09.620 --> 00:01:11.480 the stuff that really matters for you right now. 00:01:11.640 --> 00:01:13.959 Okay. Let's start with that first problem, the 00:01:13.959 --> 00:01:16.500 voice deception study. It feels very immediate. 00:01:16.579 --> 00:01:20.629 So researchers, they took 80 voice samples. Half 00:01:20.629 --> 00:01:23.370 real, half AI generated. And they just asked 00:01:23.370 --> 00:01:26.230 people, which is which? And the results. Yeah. 00:01:26.310 --> 00:01:28.010 Yeah, they really lay out the confusion. They 00:01:28.010 --> 00:01:31.530 found, OK, the basic kind of robotic AI voices, 00:01:31.569 --> 00:01:34.430 people could usually spot those as fake. That's 00:01:34.430 --> 00:01:36.010 sort of the baseline we expect. Yes, the obviously 00:01:36.010 --> 00:01:39.230 fake ones. Exactly. Yeah. But the second the 00:01:39.230 --> 00:01:42.090 AI used voice cloning, you know, tools like 11 00:01:42.090 --> 00:01:45.069 Labs, they're cheap. Anyone can get them. the 00:01:45.069 --> 00:01:46.569 ability to tell the difference just completely 00:01:46.569 --> 00:01:49.010 fell apart. And that's where that 58 % number 00:01:49.010 --> 00:01:51.849 comes in, right? Cloned AI voices mistaken for 00:01:51.849 --> 00:01:53.870 human over half the time. But the thing that 00:01:53.870 --> 00:01:56.530 really got me was the control group data. Actual, 00:01:56.530 --> 00:02:00.250 real human voices. Only correctly identified 00:02:00.250 --> 00:02:02.969 62 % of the time. Just think about that for a 00:02:02.969 --> 00:02:06.629 second. We're only 4 % better. Just 4 % at identifying 00:02:06.629 --> 00:02:08.930 a real person compared to a really good AI clone. 00:02:09.150 --> 00:02:11.330 Wow. Your brain just isn't a reliable filter 00:02:11.330 --> 00:02:14.129 for this anymore. I mean, functionally, if you 00:02:14.129 --> 00:02:16.490 can't trust what you're hearing, that changes 00:02:16.490 --> 00:02:19.590 reality, doesn't it? Beat. It is terrifying, 00:02:19.789 --> 00:02:22.370 that capability. But we also have to look at 00:02:22.370 --> 00:02:24.849 the other side, the positive uses. This tech 00:02:24.849 --> 00:02:27.310 isn't just malicious. That tension is really 00:02:27.310 --> 00:02:29.449 important here. Oh, absolutely. You've got huge 00:02:29.449 --> 00:02:32.169 accessibility benefits. Think about people who 00:02:32.169 --> 00:02:34.539 are mute. Being able to recreate their original 00:02:34.539 --> 00:02:38.000 voice or users speaking fluently in other languages, 00:02:38.099 --> 00:02:40.620 but, you know, using their own voice or even 00:02:40.620 --> 00:02:44.120 helping students, say, with ADHD, using tailored 00:02:44.120 --> 00:02:47.159 audio for learning. Massive value there. Yes, 00:02:47.439 --> 00:02:50.740 that utility is crucial, but the fact that this 00:02:50.740 --> 00:02:53.469 power exists. basically guarantees someone's 00:02:53.469 --> 00:02:55.009 going to misuse it. And we're already seeing 00:02:55.009 --> 00:02:57.289 that happen. Like the phone scams. Exactly. Phone 00:02:57.289 --> 00:02:59.729 scams. Actively using AI voice cloning right 00:02:59.729 --> 00:03:02.210 now to sound like family members. They're targeting 00:03:02.210 --> 00:03:04.689 the elderly specifically. This isn't some future 00:03:04.689 --> 00:03:06.909 threat. It's happening today. You might have 00:03:06.909 --> 00:03:08.590 even encountered it without consciously realizing 00:03:08.590 --> 00:03:11.889 it was a fake. So if our ears are failing us 00:03:11.889 --> 00:03:15.270 like this, what does that actually mean for essential 00:03:15.270 --> 00:03:18.310 public communication going forward? It means 00:03:18.310 --> 00:03:21.780 trust needs new foundations. New verification 00:03:21.780 --> 00:03:24.419 methods because the biological filter, well, 00:03:24.539 --> 00:03:26.680 it's broken. Okay, let's pivot then from this 00:03:26.680 --> 00:03:30.740 audio detection problem to the broader. explosion, 00:03:30.919 --> 00:03:33.680 really, in AI tech and how companies are placing 00:03:33.680 --> 00:03:35.939 their bets. Yeah. And what's fascinating is just 00:03:35.939 --> 00:03:38.439 the sheer range of these bets. It goes from very 00:03:38.439 --> 00:03:40.840 practical corporate tools launching right now 00:03:40.840 --> 00:03:44.539 all the way to the really far out future stuff. 00:03:44.759 --> 00:03:46.300 We can almost call it the spectrum of the AI 00:03:46.300 --> 00:03:48.520 investment. Okay. So on the corporate end, you 00:03:48.520 --> 00:03:50.900 had Amazon QuickSuite and Google Gemini Enterprise 00:03:50.900 --> 00:03:53.719 launching their AI agent suites almost simultaneously, 00:03:53.800 --> 00:03:55.979 like boom, boom. This is all about getting that 00:03:55.979 --> 00:03:57.719 immediate functionality into businesses. And 00:03:57.719 --> 00:03:59.819 the pricing tells a story too, doesn't it? Amazon's 00:03:59.819 --> 00:04:02.379 got that range, $20 to $40 a month. Google's 00:04:02.379 --> 00:04:06.379 sitting flat at $30. That $10 difference. It's 00:04:06.379 --> 00:04:08.560 interesting. It suggests maybe Amazon's saying, 00:04:08.699 --> 00:04:10.599 hey, we have a lighter, cheaper option, while 00:04:10.599 --> 00:04:12.159 Google's like, nope, this is the high -level 00:04:12.159 --> 00:04:14.139 package, shows the market's fragmenting already. 00:04:14.300 --> 00:04:15.659 Yeah, absolutely. And while that's happening 00:04:15.659 --> 00:04:17.899 in the Office Suite world, the visual side is 00:04:17.899 --> 00:04:20.759 just, it's accelerating like crazy. We saw mentions 00:04:20.759 --> 00:04:24.310 of OpenAI Sora 2 videos. Well, yeah, yeah. The 00:04:24.310 --> 00:04:27.430 realism is getting shocking to the point where 00:04:27.430 --> 00:04:29.709 even the pros, the people whose job it is to 00:04:29.709 --> 00:04:32.790 debunk fakes, they're starting to struggle. It's 00:04:32.790 --> 00:04:34.610 getting really hard to tell what's real anymore. 00:04:35.449 --> 00:04:37.569 It really is. The realism is just outpacing us. 00:04:37.949 --> 00:04:40.250 Honestly, I still wrestle with prompt drift myself 00:04:40.250 --> 00:04:43.009 when I'm trying out new models. It's hard to 00:04:43.009 --> 00:04:44.810 get exactly what you want sometimes, but the 00:04:44.810 --> 00:04:46.790 quality of what comes out, even if it's slightly 00:04:46.790 --> 00:04:49.209 off, it's terrifyingly good. And then you push 00:04:49.209 --> 00:04:52.250 even further out to the extremes. Neuralink, 00:04:52.410 --> 00:04:55.050 Elon Musk's company, right? Advancing the brain 00:04:55.050 --> 00:04:57.370 -computer interface stuff. They're claiming users 00:04:57.370 --> 00:04:59.269 will soon be able to command devices just with 00:04:59.269 --> 00:05:01.930 their thoughts. Thoughts and gestures? Whoa. 00:05:02.379 --> 00:05:04.420 I mean, just imagine scaling that complexity. 00:05:04.600 --> 00:05:07.420 A billion people sending queries with their minds 00:05:07.420 --> 00:05:10.000 all at once. That's a completely different kind 00:05:10.000 --> 00:05:12.680 of interaction. Two -sec silence. And we also 00:05:12.680 --> 00:05:14.420 need to touch on the policy side, the friction 00:05:14.420 --> 00:05:17.620 there. Right. The sources mentioned Meta hiring 00:05:17.620 --> 00:05:20.339 a controversial new AI fairness advisor, Robbie 00:05:20.339 --> 00:05:24.199 Starbuck. Apparently, there are reports citing 00:05:24.199 --> 00:05:26.759 him spreading disinformation on some really sensitive 00:05:26.759 --> 00:05:30.399 topics like vaccines and shootings. Yeah. And 00:05:30.399 --> 00:05:31.980 just to be clear for everyone listening, we're 00:05:31.980 --> 00:05:34.060 not endorsing any viewpoints here. We're just 00:05:34.060 --> 00:05:36.339 reporting what the source material highlighted 00:05:36.339 --> 00:05:39.300 about this conflict happening at the policy level. 00:05:39.399 --> 00:05:42.100 It shows how quickly these ideological fights 00:05:42.100 --> 00:05:44.459 are embedding themselves in AI governance. Exactly. 00:05:44.560 --> 00:05:46.060 It's becoming part of the structure. And for 00:05:46.060 --> 00:05:48.540 users just trying to make sense of all this chaos, 00:05:48.759 --> 00:05:51.860 there was that mention of a free prompt engineering 00:05:51.860 --> 00:05:55.829 repo. A resource. Yeah, good point. And for anyone 00:05:55.829 --> 00:05:57.930 listening who's maybe new to that term prompt 00:05:57.930 --> 00:05:59.829 engineering, it's basically just learning how 00:05:59.829 --> 00:06:02.529 to talk to an AI effectively, how to write your 00:06:02.529 --> 00:06:04.850 instructions, your prompts, so you get the result 00:06:04.850 --> 00:06:07.449 you actually need. It's becoming like the essential 00:06:07.449 --> 00:06:10.689 skill in this space. So with all these developments, 00:06:11.009 --> 00:06:13.970 the corporate race, the mind blowing tech, the 00:06:13.970 --> 00:06:17.329 policy fights, how should someone even approach 00:06:17.329 --> 00:06:19.410 all this? It feels like a lot coming at once. 00:06:19.589 --> 00:06:21.959 I think you approach it thoughtfully. balance 00:06:21.959 --> 00:06:25.379 that excitement about the potential with a healthy 00:06:25.379 --> 00:06:27.839 dose of critical skepticism. That skepticism. 00:06:27.980 --> 00:06:29.519 Yeah, that feels like the perfect way to get 00:06:29.519 --> 00:06:32.040 into our final segment. This huge paradox we 00:06:32.040 --> 00:06:35.019 mentioned between how much we use AI and how 00:06:35.019 --> 00:06:37.139 little we trust it. This came from a Reuters 00:06:37.139 --> 00:06:39.220 Institute survey, right? Six countries. Yeah. 00:06:39.279 --> 00:06:40.939 And the headline finding is really dramatic. 00:06:41.579 --> 00:06:44.139 Globally, AI use is doubling. People are jumping 00:06:44.139 --> 00:06:46.560 on these tools. But at the same time, trust in 00:06:46.560 --> 00:06:49.329 AI specifically for news. It's dropping hard. 00:06:49.529 --> 00:06:52.230 So we're using it more, trusting it less. How 00:06:52.230 --> 00:06:54.110 does that break down? Where are people using 00:06:54.110 --> 00:06:56.410 it? Mostly for utility. Yeah. For the foundational 00:06:56.410 --> 00:06:59.889 stuff. So about 24 % use AI Weekly for information 00:06:59.889 --> 00:07:02.550 seeking, you know, research, asking quick questions. 00:07:02.730 --> 00:07:04.850 21 % are using it to actually generate stuff, 00:07:04.970 --> 00:07:07.550 text, code, images. But here's the really subtle 00:07:07.550 --> 00:07:10.470 important part. A lot of this adoption isn't 00:07:10.470 --> 00:07:12.689 people consciously deciding, I'm going to use 00:07:12.689 --> 00:07:16.949 an AI now. 54%, more than half, are encountering 00:07:16.949 --> 00:07:19.490 AI through search engines. Ah, so like the AI 00:07:19.490 --> 00:07:21.910 summary is embedded in Google results. Exactly. 00:07:21.949 --> 00:07:24.490 You might be getting an AI -generated summary 00:07:24.490 --> 00:07:27.310 without even specifically asking an AI tool for 00:07:27.310 --> 00:07:30.170 it. It's just woven into the background infrastructure 00:07:30.170 --> 00:07:32.569 now. But then the second you shift from that 00:07:32.569 --> 00:07:36.529 basic utility to like... actual belief or trust, 00:07:36.810 --> 00:07:39.209 the whole picture flips. It completely flips. 00:07:39.350 --> 00:07:42.009 Only 12 % of people surveyed felt okay reading 00:07:42.009 --> 00:07:45.269 news articles written entirely by AI. Just 12%. 00:07:45.269 --> 00:07:48.769 Wow. And the vast majority, 62%, they were emphatic. 00:07:48.949 --> 00:07:51.389 They want humans writing and filtering the news 00:07:51.389 --> 00:07:54.889 they read. So, okay, we trust AI for the... The 00:07:54.889 --> 00:07:57.629 functional tasks, the heavy lifting, maybe. That's 00:07:57.629 --> 00:07:59.009 a great way to put it. It's like people trust 00:07:59.009 --> 00:08:01.329 AI to find the data, maybe stack the Lego blocks 00:08:01.329 --> 00:08:03.610 of information for them. But they absolutely 00:08:03.610 --> 00:08:07.189 demand a human to actually build the final structure, 00:08:07.310 --> 00:08:09.990 to filter it, explain it, make sense of it. It 00:08:09.990 --> 00:08:12.149 really confirms that the human role in journalism, 00:08:12.269 --> 00:08:15.230 in communication, it's still vital, critically 00:08:15.230 --> 00:08:16.990 important. Which makes sense, doesn't it? Especially 00:08:16.990 --> 00:08:19.189 when you hear about things noted elsewhere in 00:08:19.189 --> 00:08:22.910 the sources. Mark Zuckerberg and Sam Altman apparently 00:08:22.910 --> 00:08:25.829 warning about a potential AI bubble. If the creators 00:08:25.829 --> 00:08:28.629 are skeptical about the market, then why would 00:08:28.629 --> 00:08:30.709 the public fully trust the output yet? Exactly. 00:08:30.990 --> 00:08:33.289 And we also saw that quick mention of the legal 00:08:33.289 --> 00:08:36.269 landscape shifting to a judge lifting that rule 00:08:36.269 --> 00:08:38.789 requiring open AI to keep all GPT logs forever. 00:08:38.909 --> 00:08:41.250 Yeah. So the rules of the game are changing just 00:08:41.250 --> 00:08:43.730 as fast as the tech itself. It adds to the uncertainty. 00:08:43.750 --> 00:08:46.389 So given all that data, the high usage, the low 00:08:46.389 --> 00:08:49.500 trust, the need for human filtering. Where does 00:08:49.500 --> 00:08:52.820 the real value of, say, human journalism actually 00:08:52.820 --> 00:08:56.179 lie now in this world saturated with AI? Human 00:08:56.179 --> 00:08:59.379 credibility is paramount, especially for filtering 00:08:59.379 --> 00:09:01.840 complex information and interpreting events. 00:09:02.039 --> 00:09:04.200 That's the core value. So wrapping this all up, 00:09:04.279 --> 00:09:06.279 what does this really mean for you listening 00:09:06.279 --> 00:09:09.000 right now? Well, it means we're basically speeding 00:09:09.000 --> 00:09:11.870 towards a future that's both. incredibly useful 00:09:11.870 --> 00:09:15.509 and potentially incredibly confusing. Your ability 00:09:15.509 --> 00:09:18.710 to tell a real voice from a fake one, it's functionally 00:09:18.710 --> 00:09:21.919 gone. thanks to these cheap, easy -to -use cloning 00:09:21.919 --> 00:09:24.740 tools. And that corporate battle for AI agents, 00:09:24.840 --> 00:09:26.980 it's heating up. These sophisticated tools are 00:09:26.980 --> 00:09:29.159 becoming unavoidable in your daily work life, 00:09:29.279 --> 00:09:31.639 whether you actively chose them or not. And we're 00:09:31.639 --> 00:09:33.639 stuck in this really strange paradox, aren't 00:09:33.639 --> 00:09:36.940 we? We rely on AI for speed, for function, to 00:09:36.940 --> 00:09:40.200 get things done, but we absolutely demand human 00:09:40.200 --> 00:09:42.960 experts to give us the context, the filtering. 00:09:43.559 --> 00:09:46.419 The truth. That blending of reality where our 00:09:46.419 --> 00:09:49.299 own basic human senses are becoming unreliable. 00:09:49.559 --> 00:09:52.039 That feels like the critical thing to grasp. 00:09:52.399 --> 00:09:54.200 Yeah. Remember that first statistic we talked 00:09:54.200 --> 00:09:56.299 about, that tiny 4 % difference in correctly 00:09:56.299 --> 00:09:59.720 identifying real versus fake voices. If our ears 00:09:59.720 --> 00:10:01.100 just aren't good enough anymore, the question 00:10:01.100 --> 00:10:03.179 for the legal system maybe isn't, can we prove 00:10:03.179 --> 00:10:05.360 this recording is fake? Maybe the question becomes, 00:10:05.559 --> 00:10:09.120 can we ever truly trust audio or video evidence 00:10:09.120 --> 00:10:13.340 again? How fast can our legal systems adapt to 00:10:13.340 --> 00:10:16.580 a world where perfectly cloned voices, hyper 00:10:16.580 --> 00:10:20.340 -realistic fake videos are just standard, maybe 00:10:20.340 --> 00:10:22.240 even unprovable? That's definitely something 00:10:22.240 --> 00:10:23.840 to think about. That's the thought we'll leave 00:10:23.840 --> 00:10:25.639 you with today. Thank you for joining us for 00:10:25.639 --> 00:10:27.700 this deep dive. We really encourage you to keep 00:10:27.700 --> 00:10:29.759 exploring these issues. They're not going away.