WEBVTT - AI is Smart At Math But Stupid With Love 0:00:04.480 --> 0:00:12.360 Welcome to tech Stuff, a production from iHeartRadio. Hey there, 0:00:12.400 --> 0:00:15.920 and welcome to tech Stuff. I'm your host Jonathan Stricklund. 0:00:15.960 --> 0:00:19.439 I'm an executive producer with iHeart Podcasts and how the 0:00:19.480 --> 0:00:22.040 tech are you. It's time for the tech news for 0:00:22.079 --> 0:00:26.560 the week ending July twenty sixth, twenty twenty four and 0:00:26.720 --> 0:00:31.120 let's start off with a follow up on the CrowdStrike story. So, 0:00:31.160 --> 0:00:35.880 if somehow you missed that event, cybersecurity company CrowdStrike pushed 0:00:35.880 --> 0:00:39.640 out and update the customers on Windows based platforms that 0:00:39.760 --> 0:00:43.919 caused machines to go into a reboot loop. And this 0:00:44.000 --> 0:00:48.159 little oopsey caused massive outages all around the world and 0:00:48.240 --> 0:00:53.239 some companies like major airlines are still dealing with the fallout. 0:00:53.560 --> 0:00:58.680 This week, CrowdStrike posted a quote preliminary post incident review 0:00:58.880 --> 0:01:03.880 or pir are end quote giving a high level view 0:01:04.080 --> 0:01:08.520 of what happened while promising a more thorough investigation that 0:01:08.640 --> 0:01:11.720 will be shared down the line. So the post says 0:01:11.760 --> 0:01:16.440 that last Friday, quote CrowdStrike released a content configuration update 0:01:16.480 --> 0:01:20.240 for the Windows Sensor to gather telemetry on possible novel 0:01:20.280 --> 0:01:23.360 threat techniques. These updates are a regular part of the 0:01:23.440 --> 0:01:28.280 dynamic protection mechanisms of the Falcon platform. The problematic rapid 0:01:28.360 --> 0:01:33.800 response content configuration update resulted in a Windows system crash 0:01:34.080 --> 0:01:36.520 end quote. So the post goes on to say that 0:01:36.560 --> 0:01:40.360 this rapid response content feature is meant to quote respond 0:01:40.400 --> 0:01:44.440 to the changing threat landscape at operational speed end quote, 0:01:44.600 --> 0:01:47.640 and by necessity that means CrowdStrike has to be very 0:01:47.720 --> 0:01:51.440 quick to respond to emerging threats in order to protect customers. 0:01:51.680 --> 0:01:55.200 But this particular update had an undetected error in it, 0:01:55.480 --> 0:01:59.520 and in addition, the content validator you can think of 0:01:59.520 --> 0:02:02.280 it as kind of like a bouncer standing outside a nightclub. 0:02:02.520 --> 0:02:05.080 The validator is a system that's meant to look for 0:02:05.200 --> 0:02:09.440 mistakes before allowing updates to go through, but the validator 0:02:09.480 --> 0:02:12.840 itself had a bug that essentially meant it was looking 0:02:12.880 --> 0:02:15.360 the other way. So the post goes on to stress 0:02:15.400 --> 0:02:19.160 that CrowdStrike will be instituting new processes to validate code 0:02:19.200 --> 0:02:21.880 and to test it thoroughly for errors before pushing it 0:02:21.919 --> 0:02:26.000 out to customers in the future, because boy howdy, this 0:02:26.080 --> 0:02:32.560 outage was incredibly disruptive and yeah, uh, the CrowdStrike partners 0:02:32.639 --> 0:02:36.040 are mightily miffed about the outage, so not just the 0:02:36.200 --> 0:02:39.600 customers who were directly affected, but people who then had 0:02:39.680 --> 0:02:45.120 to step in and try to mitigate and repair systems 0:02:45.200 --> 0:02:48.840 that were affected by this. Countless people had to work 0:02:49.040 --> 0:02:53.040 extremely long hours in an effort to get things working again. 0:02:53.400 --> 0:02:57.119 But don't worry, CrowdStrike cares. That's why the company sent 0:02:57.160 --> 0:03:00.480 out a ten dollars voucher to some teammates and partners 0:03:00.520 --> 0:03:03.120 as a gesture of gratitude and sympathy. Now, as you 0:03:03.160 --> 0:03:06.440 might imagine, not everyone took this very well. Some folks 0:03:06.520 --> 0:03:10.120 interpreting crowdstrikes move as being insult added to injury that 0:03:10.200 --> 0:03:13.720 somehow a ten dollars gift vouchure for coffee and donuts 0:03:13.840 --> 0:03:17.440 would make up for having to provide seemingly endless support 0:03:17.520 --> 0:03:20.959 to countless customers who were affected by this outage. Some 0:03:21.080 --> 0:03:24.200 people maybe saw the gesture as being well intentioned and 0:03:24.240 --> 0:03:28.079 they didn't take umbrage to it, but unfortunately a lot 0:03:28.120 --> 0:03:30.720 of them also had trouble using the vouchure in the 0:03:30.760 --> 0:03:33.880 first place. The vouchuer was for uber eats, and uber 0:03:33.919 --> 0:03:37.280 Eats actually flagged it as fraud because so many people 0:03:37.360 --> 0:03:41.440 were trying to use the code. Interestingly, I saw something 0:03:41.520 --> 0:03:45.520 similar with ride haling services because someone I know was 0:03:45.680 --> 0:03:49.440 trying to make use of a promotional code, essentially to 0:03:49.920 --> 0:03:52.600 use a right hailing service so that they could go 0:03:52.680 --> 0:03:55.680 and help their company out during the outage. The company 0:03:55.760 --> 0:03:58.920 was saying, hey, if you need to get a ride, 0:03:59.400 --> 0:04:01.640 here's a code you can use that we really need 0:04:01.680 --> 0:04:04.400 your help. But then when they tried to use the code, 0:04:04.760 --> 0:04:07.400 they found that the code was flagged as being invalid, 0:04:07.760 --> 0:04:11.160 that too many people were using it. And this was 0:04:11.200 --> 0:04:14.200 not a small, unknown company that was extending this right 0:04:14.280 --> 0:04:17.560 hailing offer in return for more help. It was a major, 0:04:18.360 --> 0:04:21.400 major employer in the city of Atlanta. Anyway, it sounds 0:04:21.440 --> 0:04:24.080 to me like an effort to do something small but 0:04:24.120 --> 0:04:29.040 a little nice has backfired in crowdstrikes proverbial face. Okay, 0:04:29.080 --> 0:04:31.560 now let's move on to get into a whole bunch 0:04:31.560 --> 0:04:35.440 of AI news. This week, Meta unveiled the Lama three 0:04:35.600 --> 0:04:40.000 AI model and showed off generative AI capabilities, including support 0:04:40.040 --> 0:04:43.359 for eight languages and the ability to write code, with 0:04:43.520 --> 0:04:47.400 promises of multi modal capabilities in the future. So how 0:04:47.400 --> 0:04:51.240 does Lama three stack up against say, open AI's GPT 0:04:51.400 --> 0:04:54.640 model or other AI models for that matter. That's actually 0:04:54.640 --> 0:04:59.760 hard to say because it's very tricky to measure AI performance. 0:05:00.000 --> 0:05:03.400 There's not a whole lot of agreement over what constitutes 0:05:03.640 --> 0:05:06.440 a proper benchmark for AI capabilities. That are a lot 0:05:06.440 --> 0:05:09.440 of benchmarks that are being used, but they don't necessarily 0:05:09.480 --> 0:05:12.440 say that one model is superior to others. It might 0:05:12.480 --> 0:05:15.760 be that one model is better at specific sets of 0:05:15.880 --> 0:05:19.359 tasks than others, but doesn't perform as well in other regards. 0:05:19.720 --> 0:05:23.760 Meta's model did score smack dab between open ais GPT 0:05:23.960 --> 0:05:27.159 four oh and Claude three point five on a math 0:05:27.279 --> 0:05:31.640 benchmark for word problems. Meta clearly hopes that developers will 0:05:31.680 --> 0:05:35.799 rely on LAMA three more than on competing models running forward, 0:05:35.920 --> 0:05:38.400 And it sounds to me like the various AI models 0:05:38.440 --> 0:05:42.320 out there right now perform at fairly similar levels, Like 0:05:42.360 --> 0:05:45.320 there is some variation, but from the big ones out there, 0:05:45.360 --> 0:05:47.640 they tend to be kind of neck and neck. So 0:05:47.800 --> 0:05:49.760 I think for a lot of developers it'll mostly come 0:05:49.800 --> 0:05:53.640 down to which model works best from a financial standpoint, 0:05:53.880 --> 0:05:56.479 not necessarily just a technical one. But you know which 0:05:56.520 --> 0:06:00.640 one's going to be the cheapest to use. Brianna Herschbach 0:06:00.880 --> 0:06:04.359 of the Star Tribune has a piece titled X's AI 0:06:04.480 --> 0:06:09.320 chat bot groc spreads misinformation about Minnesota's ballots. Does the 0:06:09.440 --> 0:06:13.000 tech giant care well? I suspect the answer to the 0:06:13.040 --> 0:06:16.560 headline is that I'm sure there are some folks at 0:06:16.760 --> 0:06:19.719 X who very much care. But whether that's the sentiment 0:06:19.839 --> 0:06:23.200 held by the people in charge, I can't say, though 0:06:23.240 --> 0:06:27.080 I have my suspicions anyway. According to the article, Groc, 0:06:27.440 --> 0:06:31.080 the AI chat bot from X falsely claimed that it 0:06:31.120 --> 0:06:35.640 was too late for presidential ballots to change in nine states, 0:06:35.680 --> 0:06:40.200 including in Minnesota, that the candidates were already locked and loaded, 0:06:40.680 --> 0:06:42.839 and that anyone who had failed to get on the 0:06:42.880 --> 0:06:46.360 ballot already is out of luck. But quote, hey, there's 0:06:46.400 --> 0:06:49.800 always twenty twenty eight right end quote. Because of course 0:06:49.920 --> 0:06:54.520 Grock is sassy. And this is relevant obviously because here 0:06:54.520 --> 0:06:58.080 in the United States, current President Joe Biden has dropped 0:06:58.120 --> 0:07:03.360 out of the election rate and Kamala Harris, the vice president, 0:07:03.600 --> 0:07:08.960 is currently in line to get the Democratic nomination for 0:07:09.040 --> 0:07:13.920 the Democratic Party. So clearly, if we were at a 0:07:14.000 --> 0:07:17.800 point where it's too late to put people on the 0:07:17.840 --> 0:07:20.640 ballot to change the presidential ballots, then that would be 0:07:20.800 --> 0:07:26.840 a huge problem for the Democrats, because they're presumptive candidate 0:07:26.880 --> 0:07:29.560 would not be allowed to be added However, that's not 0:07:29.640 --> 0:07:33.680 the case. No state has passed the deadline for candidates 0:07:33.720 --> 0:07:37.120 to be added to the ballot, not in Minnesota, not 0:07:37.520 --> 0:07:41.160 any other state. We haven't reached that point. But now 0:07:41.280 --> 0:07:45.000 misinformation is spreading online about this because GROX said, Hey, 0:07:45.000 --> 0:07:49.320 it's too late, so tough luck, and it's just not true. 0:07:50.080 --> 0:07:52.760 I think most of us have a handle on how 0:07:52.800 --> 0:07:56.920 social platforms can facilitate the distribution and misinformation. But this 0:07:56.960 --> 0:08:00.920 is a case where X isn't just facility hating the 0:08:01.040 --> 0:08:05.880 spread of misinformation, it is generating the actual misinformation, Like 0:08:06.160 --> 0:08:09.440 the platform itself, or at least the AI chatbot that's 0:08:09.520 --> 0:08:13.560 created for that platform, is doing it. Folks at X 0:08:13.560 --> 0:08:16.120 have pointed out that GROC comes with a disclaimer that 0:08:16.280 --> 0:08:19.880 urges users to independently verify the information that's coming out 0:08:19.920 --> 0:08:21.920 of the chatbot. I don't see this as a get 0:08:21.960 --> 0:08:25.000 off jail free card. Actually, I see this as a 0:08:25.160 --> 0:08:30.760 condemnation of AI, specifically generative AI, because if you know 0:08:31.360 --> 0:08:35.319 that your tool is prone to or at least capable 0:08:35.440 --> 0:08:40.440 of generating falsehoods, maybe don't distribute the tool to customers. Like, 0:08:40.520 --> 0:08:44.200 what is the point of having an AI chatbot if 0:08:44.240 --> 0:08:48.760 the user can't trust what the chatbot says, why have 0:08:48.840 --> 0:08:52.680 a chatbot at all unless the chatbot's reliable. If you 0:08:52.840 --> 0:08:55.840 have to look up everything the chatbot tells you anyway, 0:08:56.040 --> 0:08:59.120 you might as well skip having the chatbot, just go 0:08:59.200 --> 0:09:04.320 straight to researching the question, right, because otherwise what are 0:09:04.360 --> 0:09:08.240 you doing? It seems in line, however, with Musk's approach 0:09:08.800 --> 0:09:11.679 in general. It reminds me of how Tesla has both 0:09:11.760 --> 0:09:16.720 autopilot and full self driving products and Tesla cars. Both 0:09:16.760 --> 0:09:19.080 of those, I would argue, give a false sense of 0:09:19.080 --> 0:09:22.079 what they're capable of based upon their names. Like I've 0:09:22.080 --> 0:09:25.160 often said, full self driving, which by the way, isn't 0:09:25.200 --> 0:09:28.920 full self driving. That's my opinion, But I feel like 0:09:28.960 --> 0:09:31.120 it is in line with that, And I'm sure this 0:09:31.240 --> 0:09:33.079 is just the tip of the iceberg as far as 0:09:33.120 --> 0:09:36.040 how AI is playing a part in the dissemination of 0:09:36.640 --> 0:09:41.520 information and misinformation during the election cycle. Okay, I've got 0:09:41.520 --> 0:09:43.920 a bunch more stories to go through before we start 0:09:44.120 --> 0:09:56.920 into those, let's take a quick break. We're back. So, 0:09:57.280 --> 0:10:00.160 in a rare event in US politics this week, week, 0:10:00.280 --> 0:10:03.960 the Senate unanimously passed a federal bill that would let 0:10:04.080 --> 0:10:08.079 victims of non consensual deep fakes of a sexually explicit 0:10:08.160 --> 0:10:13.400 nature sue the people who create and traffic in such material. 0:10:13.880 --> 0:10:17.560 This can include images, video, audio, et cetera. So deep 0:10:17.600 --> 0:10:21.079 fakes have been a problem for a while now, and 0:10:21.200 --> 0:10:25.600 it's a problem that disproportionately victimizes women. So this bill 0:10:25.760 --> 0:10:29.480 would make it possible for victims of deep fakes sexually 0:10:29.480 --> 0:10:33.240 explicit deep fakes to seek civil penalties against those who 0:10:33.280 --> 0:10:38.000 are making or sharing or receiving those deep fakes. The 0:10:38.080 --> 0:10:40.800 bill will still need to pass the House of Representatives 0:10:40.800 --> 0:10:43.240 before it can move on to the President's desk to 0:10:43.280 --> 0:10:45.800 get signed into law, and as of right now, that 0:10:45.840 --> 0:10:49.520 bill is in committee at the House of Representatives. Alexandra 0:10:49.679 --> 0:10:54.640 Occasio Cortes AOC. She's championing this and has made some 0:10:54.880 --> 0:10:58.760 very passionate arguments as to why the bill is needed 0:10:59.120 --> 0:11:03.559 because as of right now, there are very few avenues 0:11:03.600 --> 0:11:07.079 that victims of deep fis can pursue to get justice 0:11:07.559 --> 0:11:11.480 or any kind of action on their behalf. And that 0:11:11.640 --> 0:11:14.800 is pretty horrifying, because, I mean, this is like a 0:11:14.880 --> 0:11:20.000 consent issue, and deep fakes can cause not just incredible 0:11:20.040 --> 0:11:23.600 emotional and psychological trauma, which they can. That alone is 0:11:23.800 --> 0:11:27.640 enough to require some sort of of measure against them, 0:11:27.880 --> 0:11:30.960 but they can also impact a person's ability to make 0:11:31.000 --> 0:11:33.320 a living through no fault of their own, Like the 0:11:33.360 --> 0:11:39.040 person literally hasn't done anything, but an AI generated image 0:11:39.120 --> 0:11:44.160 or video has given them a terrible reputation and there 0:11:44.240 --> 0:11:48.400 has to be a system in place to seek justice 0:11:48.440 --> 0:11:50.280 for this. So I feel this is a very good 0:11:50.360 --> 0:11:54.960 step toward addressing a growing and disturbing problem. Moving on 0:11:55.040 --> 0:11:59.640 to other issues with AI. Sag AFTRA that's a union 0:11:59.679 --> 0:12:06.480 that presents typically actors and other performers in film, television, 0:12:06.760 --> 0:12:12.920 and video games. Well, the video game group of performers 0:12:13.200 --> 0:12:17.240 are now going on strike, so this includes voice performers 0:12:17.360 --> 0:12:20.760 as well as motion capture performers, and the reason they're 0:12:20.800 --> 0:12:24.199 going on strike is over the fact that the union 0:12:24.240 --> 0:12:27.840 has been unable to reach a satisfying agreement during contract 0:12:27.880 --> 0:12:30.880 negotiations with various video game companies. And these are like 0:12:30.920 --> 0:12:35.480 the big video game companies, so the contract agreement span 0:12:35.679 --> 0:12:38.720 a lot of territory. It's not just one thing, but 0:12:38.960 --> 0:12:42.600 one important element within all of that is over the 0:12:42.720 --> 0:12:48.000 role of AI, and specifically AI's ability to replicate voices 0:12:48.200 --> 0:12:51.320 and likenesses. And that makes a lot of sense to me. That. 0:12:51.400 --> 0:12:53.920 This would be a big sticking point because if I 0:12:54.040 --> 0:12:57.920 were hired for a gig because of my voice or 0:12:57.960 --> 0:13:01.760 my likeness, I don't want AI to copy me, because 0:13:01.800 --> 0:13:05.240 then that copy could be used in perpetuity without my 0:13:05.480 --> 0:13:09.520 participation or compensation. I would be taking one gig and 0:13:09.679 --> 0:13:13.640 essentially putting myself out of a job forever because that 0:13:13.760 --> 0:13:17.080 copy could get passed around, and who knows what stuff 0:13:17.120 --> 0:13:21.080 my AI mimic might be incorporated into. I could find 0:13:21.120 --> 0:13:23.840 a copy of myself in projects that I never would 0:13:23.880 --> 0:13:27.200 have agreed to work on. That's not really cool, right, 0:13:27.480 --> 0:13:31.040 Like if I were hired to voice something that I 0:13:31.080 --> 0:13:32.560 really believed in, I was like, Oh, this is a 0:13:32.559 --> 0:13:36.720 really cool story, and I believe in the developer, and 0:13:36.760 --> 0:13:38.319 I believe in the story they're trying to tell. I 0:13:38.360 --> 0:13:40.160 would love to be part of this, by the way, 0:13:40.160 --> 0:13:42.040 I would love to be part of stuff like that. 0:13:42.040 --> 0:13:44.760 That would be a lot of fun. Then, yeah, that's great, 0:13:44.880 --> 0:13:47.840 But what if they then used my voice to generate 0:13:47.920 --> 0:13:50.240 a copy of it and put it into something that 0:13:50.400 --> 0:13:55.120 I would find morally objectionable or it goes against my 0:13:55.160 --> 0:14:00.680 own personal beliefs. That's a real problem. Apparently, these AI 0:14:00.960 --> 0:14:04.120 points in the agreements have been contentious, with video game 0:14:04.160 --> 0:14:06.800 companies reluctant to promise that they're not going to use 0:14:06.840 --> 0:14:11.640 AI tools that could lead to the detriment of union members. Interestingly, 0:14:11.880 --> 0:14:14.960 striking members are still allowed to participate in appearances at 0:14:15.000 --> 0:14:18.240 San Diego Comic Con, which started earlier this week and 0:14:18.360 --> 0:14:20.520 goes through the weekend, and they can do so without 0:14:20.600 --> 0:14:24.920 penalty because usually a striking sag after member isn't supposed 0:14:24.960 --> 0:14:28.160 to promote any work that is covered by the strike, 0:14:28.400 --> 0:14:31.000 but in this case, the proximity of the strike to 0:14:31.520 --> 0:14:35.240 the Comic Con event called for an exception, so folks 0:14:35.280 --> 0:14:37.920 who were scheduled to appear can still do so. They 0:14:37.960 --> 0:14:40.680 can still talk about the projects they were hired to do, 0:14:40.960 --> 0:14:43.600 and that's really good news for fans. Potentially, it's also 0:14:43.680 --> 0:14:46.240 good news for union members who may be able to 0:14:46.280 --> 0:14:50.360 take the opportunity to speak about their concerns regarding AI 0:14:50.560 --> 0:14:53.720 and other aspects of the dispute. At the top of 0:14:53.760 --> 0:14:57.240 this episode, I talked about how Meta's Lama III performed 0:14:57.240 --> 0:15:01.960 well in certain benchmarks against other AS models well. Google's 0:15:02.120 --> 0:15:06.120 Deep Mind department similarly unveiled a couple of AI systems, 0:15:06.160 --> 0:15:10.120 one called Alpha Proof and another called alpha geometry two 0:15:10.560 --> 0:15:14.920 that performed really well on a very tough math test. 0:15:15.200 --> 0:15:18.160 According to Google, the AI models were able to achieve 0:15:18.200 --> 0:15:22.000 the equivalent of a silver metal performance by solving problems 0:15:22.080 --> 0:15:26.520 that were part of the most recent International Mathematical olympiad BENJ. 0:15:26.760 --> 0:15:29.280 Edwards of Ours Technica has a great piece about this. 0:15:29.480 --> 0:15:34.120 It's titled Google claims Math breakthrough with proof solving AI models. 0:15:34.360 --> 0:15:36.360 Benje also does a really good job of pointing out 0:15:36.440 --> 0:15:41.960 that measuring AI capabilities is tricky. It is not always straightforward. 0:15:41.960 --> 0:15:46.240 In fact, it rarely is straightforward. The achievement does mark 0:15:46.280 --> 0:15:50.240 advancements in AI's ability to parse and then solve complex 0:15:50.360 --> 0:15:53.840 mathematical problems, and that is pretty cool. But ben J. 0:15:53.920 --> 0:15:59.160 Edwards also quotes Sir Timothy Gowers about this. Sir Timothy 0:15:59.200 --> 0:16:02.640 Gowers took to AX and wrote, quote the program needed 0:16:02.880 --> 0:16:05.920 a lot longer than the human competitors, some of the 0:16:05.960 --> 0:16:10.640 problems over sixty hours end quote, and that if human 0:16:10.680 --> 0:16:14.000 competitors had been given the same luxury of time, they 0:16:14.040 --> 0:16:17.760 probably would have scored higher in this test as well. 0:16:18.000 --> 0:16:20.840 So what Gowers is saying is this is not an 0:16:20.880 --> 0:16:24.560 apples to apples comparison. You can't say, oh, this AI 0:16:24.720 --> 0:16:28.680 program is x times smarter than the average person when 0:16:28.720 --> 0:16:31.240 it comes to math, because you have to take into 0:16:31.320 --> 0:16:34.680 account all the different parameters and say, well, if the 0:16:34.880 --> 0:16:37.040 people who are really good at math were given the 0:16:37.040 --> 0:16:39.800 same amount of time to solve a problem, they might 0:16:39.840 --> 0:16:42.440 have done it as well or better than the AI 0:16:42.560 --> 0:16:45.840 model did. Gowers also pointed out that humans had to 0:16:45.840 --> 0:16:49.000 convert these problems into a language that the AI models 0:16:49.000 --> 0:16:51.560 could understand before they could tackle the problem in the 0:16:51.560 --> 0:16:55.160 first place. So again, the AI models weren't presented with 0:16:55.240 --> 0:16:59.640 the exact same problem, or the problem wasn't framed the 0:16:59.680 --> 0:17:03.080 exact same way for the AI versus the human competitors 0:17:03.120 --> 0:17:07.000 who actually participated in the Olympiad, and that if the 0:17:07.119 --> 0:17:10.640 AI had had to parse the problems the way that 0:17:10.800 --> 0:17:15.400 humans did, maybe the AI wouldn't have performed as well. 0:17:15.520 --> 0:17:21.160 So yeah, it's complicated hopping over to open ai because yeah, 0:17:21.200 --> 0:17:24.280 we're not done with AI yet. Open ai announced this 0:17:24.359 --> 0:17:28.400 week that a prototype version of its search GPT product 0:17:28.840 --> 0:17:32.639 is now ready to go. This is open AI's artificial 0:17:32.680 --> 0:17:36.840 intelligence powered search tool. It is a clear shot across 0:17:37.000 --> 0:17:40.120 Google's bow, and according to open ai, the tool will 0:17:40.160 --> 0:17:44.240 provide quote fast and timely answers with clear and relevant 0:17:44.280 --> 0:17:47.280 sources end quote. I have signed up to join the 0:17:47.280 --> 0:17:50.080 wait list to try this out, but as of this recording, 0:17:50.119 --> 0:17:52.800 I have not yet personally been able to use this tool. 0:17:53.080 --> 0:17:55.880 I do have concerns similar to what I was saying earlier. 0:17:56.600 --> 0:17:59.080 I have these concerns with pretty much all AI enabled 0:17:59.119 --> 0:18:01.760 tools that are like And one of those concerns is 0:18:01.760 --> 0:18:05.280 that I worry AI enhanced search will mean fewer people 0:18:05.320 --> 0:18:08.679 will actually navigate to what the real sources of the 0:18:08.760 --> 0:18:12.400 information were. They'll just be satisfied with the AI summarized 0:18:12.760 --> 0:18:17.520 or synthesized results and that's it, Which means the pages 0:18:17.600 --> 0:18:20.199 that the AI is drawing from the source pages are 0:18:20.200 --> 0:18:23.480 going to get less traffic, and that means less revenue, 0:18:23.760 --> 0:18:27.760 which ultimately means the businesses that actually operate these pages 0:18:27.800 --> 0:18:30.880 could find themselves strapped for cash. And then what happens, 0:18:31.119 --> 0:18:34.080 you know, when the underlying infrastructure that the AI is 0:18:34.119 --> 0:18:38.680 dependent upon goes rotten. What happens then? Plus there's still 0:18:38.720 --> 0:18:42.840 the worry of mistakes, you know, like confabulations slash hallucinations 0:18:42.880 --> 0:18:46.320 with Generative AI. Google's version of this same sort of 0:18:46.400 --> 0:18:50.720 tool famously suggested that people include non toxic glue in 0:18:50.800 --> 0:18:54.120 their pizza ingredients in order to prevent cheese from sliding 0:18:54.160 --> 0:18:58.239 off their pizza. Clearly that is not good advice. So 0:18:58.520 --> 0:19:02.520 how reliable will search GPTB and how will folks know 0:19:03.000 --> 0:19:05.440 if the answers they get are actually good? I mean, 0:19:05.480 --> 0:19:07.320 I guess one way you could check is you could 0:19:07.359 --> 0:19:10.439 click through to the sources. But then again, why do 0:19:10.480 --> 0:19:13.760 you even need AI search at that point? Like, if 0:19:13.760 --> 0:19:16.159 you already have to double check to make sure the 0:19:16.200 --> 0:19:19.800 answer that AI gave you is relevant and accurate, what 0:19:19.960 --> 0:19:22.639 good is the AI? If you have to double check, 0:19:22.680 --> 0:19:24.919 then you might as well just get rid of the 0:19:24.960 --> 0:19:27.720 AI and go straight to the source in the first place. 0:19:28.040 --> 0:19:30.320 It's just like I was saying before, AI is just 0:19:30.400 --> 0:19:35.000 making things, giving an extra step. It's like turning to 0:19:35.040 --> 0:19:37.120 a random person next to you and saying, like, how 0:19:37.160 --> 0:19:39.480 do you repair a car? Like you have no idea 0:19:39.760 --> 0:19:44.240 they have any experience or knowledge in that realm, and 0:19:44.280 --> 0:19:47.200 if they give you advice, you don't know if it's trustworthy. 0:19:47.240 --> 0:19:50.840 You have to look it up. Well whatever open AI 0:19:50.920 --> 0:19:54.119 made this announcement, and after that, alphabet shares fell around 0:19:54.160 --> 0:19:58.320 three percent. So I guess like these questions are ones 0:19:58.359 --> 0:20:02.640 that don't necessarily run around the minds of investors. I'm 0:20:02.720 --> 0:20:08.080 just gonna progressively go bonkers over here as the world 0:20:08.240 --> 0:20:11.920 continues to embrace AI and I keep asking, but wait, 0:20:12.280 --> 0:20:16.600 I have questions. I also have more news. But before 0:20:16.640 --> 0:20:28.320 we get to that, let's take another quick break. So 0:20:28.520 --> 0:20:31.680 before the break, I was talking about Google and speaking 0:20:31.720 --> 0:20:35.080 of alphabet slash Google. One of the other companies in 0:20:35.200 --> 0:20:39.040 that group of companies under alphabet is YouTube, and it 0:20:39.080 --> 0:20:43.040 could be facing some pretty tough issues in Russia. So 0:20:43.600 --> 0:20:46.840 the Russian government has indicated that YouTube could see that 0:20:47.080 --> 0:20:51.199 the agents will throttle YouTube traffic by forty percent on 0:20:51.240 --> 0:20:54.479 desktop computers this week, So slowing down YouTube traffic by 0:20:54.520 --> 0:20:57.679 forty percent, that could increase to a seventy percent slowed 0:20:57.720 --> 0:21:01.200 down by next week. And you might ask why why 0:21:01.280 --> 0:21:06.360 is the Russian government throttling traffic from YouTube in Russia? Well, 0:21:06.400 --> 0:21:09.960 it's because YouTube previously placed a block on some channels 0:21:10.040 --> 0:21:14.240 that were carrying Russian state media. Russian state media has 0:21:15.040 --> 0:21:17.520 a bit of a reputation for having, let us say, 0:21:17.600 --> 0:21:22.359 a biased view of the news, and that it often 0:21:22.440 --> 0:21:25.359 is just really seen as a mouthpiece for the Russian 0:21:25.400 --> 0:21:29.640 government itself, and thus is really a way of distributing propaganda. 0:21:29.880 --> 0:21:34.760 So presumably what has happened is YouTube determined that some 0:21:35.040 --> 0:21:39.520 Russian state media YouTube channels were in violation of YouTube's 0:21:39.520 --> 0:21:43.280 policies and so block them, And the Russian government's response 0:21:43.400 --> 0:21:47.200 is We're going to make all traffic to YouTube super 0:21:47.280 --> 0:21:51.640 duper slow, so that Russian users will get really frustrated 0:21:51.680 --> 0:21:56.200 about this unless you reinstate those channels. So essentially this 0:21:56.240 --> 0:21:59.240 is a threat. It's a standoff between YouTube and the 0:21:59.320 --> 0:22:03.160 Russian government. And I did not see any responses from 0:22:03.280 --> 0:22:06.960 YouTube about this as of the time I'm recording this episode. 0:22:07.000 --> 0:22:08.879 That may be different by the time you hear it, 0:22:09.280 --> 0:22:13.280 but yeah, right now, it is like a standoff, which 0:22:13.320 --> 0:22:16.280 is a big old yikes from me. Now, if you 0:22:16.400 --> 0:22:19.480 don't use Google, but you are used to seeing search 0:22:19.480 --> 0:22:23.880 results include stuff from Reddit threads, that could be changing. 0:22:24.320 --> 0:22:27.960 So Reddit has a content policy that forbid sites from 0:22:28.040 --> 0:22:32.480 crawling Reddit forums without first agreeing to follow Reddit's rules, 0:22:32.720 --> 0:22:35.080 and those rules cover a lot of ground. One of 0:22:35.119 --> 0:22:38.560 the big rules that Reddit has is that companies that 0:22:38.640 --> 0:22:42.240 want to crawl Reddit for the purposes of gathering data 0:22:42.280 --> 0:22:45.760 to train AI models have to pay for that privilege 0:22:45.880 --> 0:22:48.119 that you've got to fork over some cash if you're 0:22:48.119 --> 0:22:50.879 going to use Reddit to train up your AI. This 0:22:50.920 --> 0:22:53.240 is something that a lot of Reddit users are actually 0:22:53.240 --> 0:22:56.520 salty about because you know, a lot of people don't 0:22:56.560 --> 0:22:59.199 want their stuff to be used to train AI in 0:22:59.240 --> 0:23:01.359 the first place. They don't have any say in the matter, 0:23:01.640 --> 0:23:04.800 like Reddit gets to make that call. They don't, And 0:23:05.080 --> 0:23:07.600 some users are thinking that's not really fair that they 0:23:07.600 --> 0:23:10.560 should have more control over the content they create. Reddit 0:23:10.600 --> 0:23:13.239 says no, once you post it here, it's ours, like 0:23:13.280 --> 0:23:16.280 it was yours then you hit post and now it 0:23:16.280 --> 0:23:19.760 belongs to us. And users also are kind of salty 0:23:19.880 --> 0:23:23.840 because the compensation it's going to Reddit doesn't go to 0:23:23.880 --> 0:23:26.600 the people who are generating the content in the first place. Right, Like, 0:23:26.640 --> 0:23:30.240 these AI models are using Reddit content to train themselves, 0:23:30.400 --> 0:23:33.880 but the people who are making the content, they don't 0:23:33.880 --> 0:23:36.320 get any of the money. The money goes to Reddit, 0:23:36.720 --> 0:23:38.720 So it's going to the platform, but not the people 0:23:38.720 --> 0:23:41.520 who are creating the actual content that's training AI. So yeah, 0:23:41.560 --> 0:23:43.719 there are a couple of reasons why Reddit users are 0:23:43.760 --> 0:23:47.080 not happy about this. But anyway, this means that any 0:23:47.119 --> 0:23:51.679 company that has not agreed to these these rules that 0:23:51.720 --> 0:23:55.640 Reddit has posted, they are not allowed to crawl the site. 0:23:55.960 --> 0:24:00.199 Without crawling the site, you can't index it, and specific 0:24:00.240 --> 0:24:03.359 threads will not pop up in search results. If that 0:24:03.600 --> 0:24:07.520 is the case. Apparently, search engines not named Google have 0:24:08.080 --> 0:24:12.399 largely not yet forged these kinds of agreements with Reddit, 0:24:12.960 --> 0:24:16.320 So the result is that if you search for topics 0:24:16.320 --> 0:24:20.720 that are covered on Reddit forums or Reddit threads, you 0:24:20.760 --> 0:24:24.600 will get fewer results on non Google search engines. Some 0:24:24.640 --> 0:24:28.600 of them will return some Reddit results, but others won't. 0:24:29.000 --> 0:24:32.240 For a full breakdown on this issue, including some really 0:24:32.280 --> 0:24:35.520 great insight into the search engine landscape in general, i 0:24:35.600 --> 0:24:41.040 recommend reading Sharon Harding's excellent piece in Ours Technica's article 0:24:41.119 --> 0:24:45.800 titled non Google search engines blocked from showing recent Reddit results. 0:24:46.359 --> 0:24:50.840 Shila Chiang CNBC reports that Apple is no longer among 0:24:50.920 --> 0:24:55.440 the top five smartphone brands in China. Rather, Chinese companies 0:24:55.600 --> 0:24:58.720 domestic companies are taking up the top five spots, and 0:24:58.800 --> 0:25:02.040 Apple has fallen to hold fourteen percent of the market 0:25:02.119 --> 0:25:04.639 share in the country. The Chinese market has been a 0:25:04.680 --> 0:25:08.480