WEBVTT 00:00:00.000 --> 00:00:04.800 Welcome to another episode of Tech News. Siri, we know you're watching and you're welcome. 00:00:04.800 --> 00:00:08.160 Because according to new research from Proof News, tech giants including Apple, 00:00:08.160 --> 00:00:13.520 Anthropic, and NVIDIA have been training their AI models using transcripts from over 170,000 00:00:13.520 --> 00:00:18.400 YouTube videos. This trove included work of big internet stars like MrBeast, PewDiePie, 00:00:18.400 --> 00:00:23.200 Jacksepticeye, and Marques Brownlee, but also clips from major television shows like Jimmy 00:00:23.200 --> 00:00:27.440 Kimmel Live, Last Week's Night, and The Late Show with Stephen Colbert. These transcripts 00:00:27.440 --> 00:00:33.520 weren't taken directly by these companies, but were collected by a third-party non-profit called 00:00:33.520 --> 00:00:39.760 Eleuther AI. That's a weird combination of letters. As part of The Pile, an open-source 00:00:39.760 --> 00:00:44.960 dataset intended to help academics and small developers to train their own AI models and 00:00:44.960 --> 00:00:50.080 not, as it might sound, a giant haemorrhoid. The material appears to have been scraped entirely 00:00:50.080 --> 00:00:54.560 without the original creator's permission. Obviously, these transcripts are generated from 00:00:54.640 --> 00:01:00.240 copyrighted material made by human artists with bills and mortgages, but as pointed out by Marques 00:01:00.240 --> 00:01:05.120 Brownlee, he uses a paid transcription service rather than generated subtitles, meaning those 00:01:05.120 --> 00:01:10.320 transcripts are kind of copyrighted squared. In addition to educational videos from Khan Academy, 00:01:10.320 --> 00:01:14.960 MIT, and Harvard, the database also contains materials from conspiracy theorists and flat 00:01:14.960 --> 00:01:19.600 earth videos, so don't be shocked when you ask directions and Siri gives you a lecture about 00:01:19.600 --> 00:01:25.840 the danger of chemtrails. Through the world's largest record companies, Universal Warner and 00:01:25.840 --> 00:01:32.000 Sony Music have launched a $2.6 billion lawsuit against Verizon, alleging that the telecom company 00:01:32.000 --> 00:01:37.040 has profited from ignoring widespread piracy. According to the suit, Verizon should be held liable 00:01:37.040 --> 00:01:41.600 for its decision to continue to provide internet services to thousands of flagrant pirates. It's 00:01:41.600 --> 00:01:46.640 not clear how likely this lawsuit is to succeed, given that it closely resembles a previous lawsuit 00:01:46.640 --> 00:01:51.920 by the same plaintiffs against Cox Communications. That suit found that Cox was guilty of failing to 00:01:51.920 --> 00:01:56.960 stop piracy, but wound up overturned on appeal because Cox didn't actually make any money from 00:01:56.960 --> 00:02:01.920 the infringement. See, while it's obviously true that Verizon makes more money if it doesn't force 00:02:01.920 --> 00:02:05.520 its paying customers to walk the plank the moment they download BitTorrent, it actually makes the 00:02:05.520 --> 00:02:09.920 same amount of money whether or not you steal birdemic shock and terror, or you pay for it like 00:02:09.920 --> 00:02:14.480 a good boy. Meta has launched a small pilot program that will give researchers access to 00:02:14.480 --> 00:02:19.280 Instagram's own data so they can study the app's effect on teenagers and young adults. 00:02:19.280 --> 00:02:23.200 Meta is now accepting proposals for studies and will accept up to seven submissions, 00:02:23.200 --> 00:02:28.480 one for each deadly sin. However, researchers still need to obtain consent from study participants 00:02:28.480 --> 00:02:32.880 and their parents. It's still unclear how extensive this information access will be, 00:02:32.880 --> 00:02:37.760 but it's a notable step given Meta's past hostility towards researchers. Prior research has 00:02:37.760 --> 00:02:42.960 shown a distinct correlation between heavy social media use and mental health issues like anxiety 00:02:42.960 --> 00:02:47.440 and depression, especially in teenagers. What's less clear however is whether the kids are sad 00:02:47.440 --> 00:02:51.360 because they're on Instagram all the time or if they're on Instagram all the time because they're 00:02:51.360 --> 00:02:55.840 sad. Regardless, parents and legislators are concerned and it's driving more and more scrutiny 00:02:55.840 --> 00:03:00.800 from regulators. TikTok lost its recent legal challenge to the EU's decision to classify it 00:03:00.800 --> 00:03:05.520 as a very large online platform on account of the size of its parent company and because 00:03:05.520 --> 00:03:09.680 it's the social media equivalent of free-basing. If you made it through our main stories, 00:03:09.680 --> 00:03:14.000 but wish they could have been slightly quicker? Congratulations, it's the quick bits, they're 00:03:14.000 --> 00:03:21.440 here. Noctua's new NHD 15 G2 cooler is leaving some users rattle, specifically their coolers. 00:03:21.440 --> 00:03:25.600 They're rattling. After a report from hardware busters and customer complaints, Noctua says 00:03:25.600 --> 00:03:29.920 they're investigating the issue. Noctua believes the problem could be temporarily fixed by using 00:03:29.920 --> 00:03:34.000 tape or foam until they find a permanent solution, but they're also offering full refunds. 00:03:34.720 --> 00:03:40.160 As much as we love to hock Tua for Noctua, it seems that even the beige and brown giant 00:03:40.160 --> 00:03:45.200 can avoid QA issues. Thankfully, Noctua has a good reputation for customer support. So as 00:03:45.200 --> 00:03:49.440 they say in the company's native Austria, you got a problem with Noctua, you talk to her. 00:03:51.920 --> 00:03:56.560 Google has announced Project Oscar, an open source platform allowing developers to create AI 00:03:56.560 --> 00:04:01.520 agents to help manage their open source projects. For example, the agent can summarize and highlight 00:04:01.520 --> 00:04:07.600 relevant information in issue reports from users. Rather than have the AI code, which is an awful 00:04:07.600 --> 00:04:12.320 idea that has only ever gone wrong, the hope seems to be that the agents can help with the 00:04:12.320 --> 00:04:17.280 disruptions in toil. I guess I don't understand why someone would want to use this. I'm pretty sure 00:04:17.280 --> 00:04:21.680 the only reason to maintain an open source project is to suffer thanklessly while making no money. 00:04:23.200 --> 00:04:27.840 Apple is making a Squidwardian attempt to prevent a future where everything is chrome 00:04:27.840 --> 00:04:33.040 with a new Hitchcockian ad for Safari. Apple continues to beat the dead horse saying they 00:04:33.040 --> 00:04:37.600 care about privacy and will keep pummeling the deceased stallion saying they don't care about 00:04:37.600 --> 00:04:42.720 your privacy, they just want your data with them in their sandbox. Also, it's not like they've 00:04:42.720 --> 00:04:46.960 stopped accepting Google's multi-billion dollar bribes to make Google the default search engine 00:04:46.960 --> 00:04:51.040 in Safari. At least not until regulators force them to stop. Also, Apple, I don't know whose 00:04:51.040 --> 00:04:56.080 idea this was, but maybe you should stop feeding your AI. Videos about how birds aren't real. 00:04:56.240 --> 00:05:06.160 Research and recovery company RMS Titanic, Inc. is prepared to send remote submersibles down 12,000 00:05:06.160 --> 00:05:11.120 feet to take 3D scans of the wreck of the Titanic. This is the first mission to the wreck since the 00:05:11.120 --> 00:05:15.840 implosion of the Titan submersible last year, but this expedition is intended to preserve 00:05:15.840 --> 00:05:23.120 knowledge for study rather than turn thinking-breathing humans into soup. In similar news, 00:05:23.120 --> 00:05:28.480 drone manufacturer DJI has released high-quality footage of Mount Everest taking with the DJI 00:05:28.480 --> 00:05:32.800 Mavic 3 Pro. Now you can travel from base camp to summit and you don't even have to step over 00:05:32.800 --> 00:05:38.480 the frozen bodies of dead billionaires to get there. The FTC has said he health can no longer scan 00:05:38.480 --> 00:05:43.920 your little man with their app, Calmera. It aims to be an AI pecker protector that could detect 00:05:43.920 --> 00:05:49.920 10 STIs that scourge your courgette with 94% accuracy. That's a 1 in 20 chance they can screw 00:05:49.920 --> 00:05:55.920 up identifying a bug in your love slug and that's just the STIs you can see. Turns out the app can 00:05:55.920 --> 00:06:01.040 only accurately detect four diseases between your knees and could be fooled by inanimate 00:06:01.040 --> 00:06:06.240 nobjects like phallic vases and pastries. I mean that's what I think every time I look at a croissant. 00:06:06.240 --> 00:06:11.040 Looks like you should block the AI schlock. Get a doctor gock at your...penis. 00:06:12.080 --> 00:06:15.520 And doctors recommend you get a tri-weekly injection of techno, so make sure you come back 00:06:15.520 --> 00:06:18.720 on Friday. Personally, I've got something I need you to take a look at.