WEBVTT 00:00:02.200 --> 00:00:04.799 Welcome to the AI Chat Podcast. I'm your host, 00:00:04.860 --> 00:00:07.179 Jaden Schaefer. Every day I cover cutting -edge 00:00:07.179 --> 00:00:09.320 AI news and talk with the leaders behind it, 00:00:09.339 --> 00:00:11.320 breaking down what it means for your life and 00:00:11.320 --> 00:00:17.579 business. OpenAI's first piece of hardware coming 00:00:17.579 --> 00:00:20.780 out with Johnny Ive is going to be a smart speaker. 00:00:21.230 --> 00:00:23.390 in the shape of a donut that costs between $200 00:00:23.390 --> 00:00:28.269 and $400. ByteDance is training a 10 trillion 00:00:28.269 --> 00:00:30.690 parameter model and their goal is to basically 00:00:30.690 --> 00:00:34.710 compete with Anthropic. Replit's CEO says stop 00:00:34.710 --> 00:00:37.609 learning to code and build a self -driving company 00:00:37.609 --> 00:00:41.969 instead. Airbnb credits AI for an 80 % jump in 00:00:41.969 --> 00:00:43.630 the features that they have shipped and they 00:00:43.630 --> 00:00:46.990 said they're going to test AI search next. Moonshot's 00:00:46.990 --> 00:00:49.670 Kimmy K3 breaks out of its sandbox to cheat on 00:00:49.670 --> 00:00:52.030 a security test. because, of course, why not? 00:00:52.329 --> 00:00:54.469 Meta, Anthropic, OpenAI, they all broke out of 00:00:54.469 --> 00:00:57.469 their security, you know, sandboxes, so why not 00:00:57.469 --> 00:01:00.130 KimiK3? If you want to get a deeper dive on all 00:01:00.130 --> 00:01:02.609 of these stories and more every single day to 00:01:02.609 --> 00:01:05.829 your inbox, go check out AIChatDaily .com. That 00:01:05.829 --> 00:01:08.250 is the news site that goes along with this podcast, 00:01:08.370 --> 00:01:10.549 and I cover deep dive articles on all of the 00:01:10.549 --> 00:01:12.969 podcast episodes and topics that I cover here. 00:01:13.069 --> 00:01:15.629 I do about 20 different articles a day, so there's 00:01:15.629 --> 00:01:17.930 a ton of deep dive content, and it's kind of 00:01:17.930 --> 00:01:20.340 like a show notes for this show. In addition, 00:01:20.400 --> 00:01:22.159 I also have a subscribe button on that website 00:01:22.159 --> 00:01:24.459 where if you hit it, you can get a daily newsletter 00:01:24.459 --> 00:01:26.439 sent to your inbox every single day with the 00:01:26.439 --> 00:01:28.939 top five stories to give you a really quick and 00:01:28.939 --> 00:01:31.579 easy breakdown of what's happening in AI every 00:01:31.579 --> 00:01:34.760 single day. So that is AIChatDaily .com. Go check 00:01:34.760 --> 00:01:38.890 it out. OpenAI is building a $300 to $400 smart 00:01:38.890 --> 00:01:40.469 speaker. It's going to be shaped, some people 00:01:40.469 --> 00:01:42.010 say like a hockey puck. I've heard it's shaped 00:01:42.010 --> 00:01:46.209 like a donut. Essentially, the design got created 00:01:46.209 --> 00:01:48.450 by Johnny Ive, who, of course, famously created 00:01:48.450 --> 00:01:51.189 the iPhone. And he is going to be launching this 00:01:51.189 --> 00:01:54.780 in 2027 in this partnership with OpenAI. The 00:01:54.780 --> 00:01:57.540 device has moving parts and it has a bunch of 00:01:57.540 --> 00:02:00.120 lights. The idea is to make it feel really responsible 00:02:00.120 --> 00:02:01.500 while you're talking to it. So it's going to 00:02:01.500 --> 00:02:04.280 be like moving around. They're calling this kind 00:02:04.280 --> 00:02:07.280 of like a premium alternative to an Amazon Echo 00:02:07.280 --> 00:02:11.539 or Echo Studio. And the Echo Studio is $220. 00:02:11.860 --> 00:02:15.199 So they're, you know, almost $100, maybe $200, 00:02:15.400 --> 00:02:18.219 maybe twice the price of the Amazon Echo Studio. 00:02:18.500 --> 00:02:20.340 All of these details are coming out at a very 00:02:20.340 --> 00:02:22.879 exciting time because Love From, which is Johnny 00:02:22.879 --> 00:02:25.210 Ives' design studio, They created all of this. 00:02:25.289 --> 00:02:27.289 It's super high quality. They have all this premium 00:02:27.289 --> 00:02:28.969 materials, but this is all happening at the same 00:02:28.969 --> 00:02:31.590 time that OpenAI is being sued by Apple over 00:02:31.590 --> 00:02:34.509 alleged hardware trade secrets. This would be 00:02:34.509 --> 00:02:37.949 OpenAI's second consumer device after the $230 00:02:37.949 --> 00:02:40.409 Codex Micro Keyboard, which I personally have. 00:02:40.550 --> 00:02:43.349 The company is evaluating other different AI 00:02:43.349 --> 00:02:46.050 -powered hardware beyond just this speaker. ByteDance 00:02:46.050 --> 00:02:49.490 is training a 10 trillion parameter AI model, 00:02:49.590 --> 00:02:52.090 so it's about the size of Anthropix Mythos 5, 00:02:52.310 --> 00:02:54.629 but it is... three times bigger than China's 00:02:54.629 --> 00:02:57.430 largest model that they have ever released. This, 00:02:57.569 --> 00:02:59.650 I think, is showing that ByteDance really wants 00:02:59.650 --> 00:03:02.949 to be a world -class frontier AI lab, independent 00:03:02.949 --> 00:03:04.930 of whatever's coming out of the United States. 00:03:05.150 --> 00:03:07.330 And the scale is something that no other Chinese 00:03:07.330 --> 00:03:10.210 company has attempted. Pre -training is expected 00:03:10.210 --> 00:03:12.849 to take three to six months before they do fine 00:03:12.849 --> 00:03:15.629 -tuning. Their current chatbot is called DuoBao, 00:03:15.830 --> 00:03:19.090 and it already has 324 million monthly active 00:03:19.090 --> 00:03:21.650 users in China. So it's definitely the most used 00:03:21.650 --> 00:03:26.539 AI. chat tool in China. This seed research team 00:03:26.539 --> 00:03:28.699 that is leading the effort has grown to about 00:03:28.699 --> 00:03:32.080 2000 people across China and overseas. I think 00:03:32.080 --> 00:03:34.240 it's led by former Google DeepMind scientist 00:03:34.240 --> 00:03:38.060 Wu Yong He. So ByteDance has really closed most 00:03:38.060 --> 00:03:40.180 of the gap with a lot of the US frontier labs 00:03:40.180 --> 00:03:43.039 in all of the different benchmarks. Moonshot 00:03:43.039 --> 00:03:45.840 and Alibaba models now are very close to Fable 00:03:45.840 --> 00:03:48.659 5 in a lot of different tests. And now a successful 00:03:48.659 --> 00:03:51.139 10 trillion parameter model potentially could 00:03:51.139 --> 00:03:53.580 really put China to the forefront at a time when 00:03:53.580 --> 00:03:55.900 we're having a lot of conversations about banning 00:03:55.900 --> 00:03:58.060 Chinese open weight models and all of this stuff. 00:03:58.180 --> 00:04:00.960 Now, I will say, you know, given China's massive 00:04:00.960 --> 00:04:03.280 population, much bigger than the United States, 00:04:03.360 --> 00:04:06.319 having 300 million monthly active users is and 00:04:06.319 --> 00:04:07.939 you know, it is the it's kind of the forefront, 00:04:08.240 --> 00:04:11.219 the number one company in China doing this, it 00:04:11.219 --> 00:04:14.900 is definitely behind America's 900 million monthly 00:04:14.900 --> 00:04:17.220 active users, but that's also globally for chat 00:04:17.220 --> 00:04:19.379 GPT. So it's gonna be interesting to see how 00:04:19.379 --> 00:04:22.300 these two things shake out. And of course, it's 00:04:22.300 --> 00:04:25.259 also interesting considering the fact that ChatGPT 00:04:25.259 --> 00:04:28.160 and Anthropic are banned in China. Meanwhile, 00:04:28.540 --> 00:04:31.500 you have Chinese models coming into the United 00:04:31.500 --> 00:04:34.839 States, and they are allowed to be used. Replit 00:04:34.839 --> 00:04:38.339 just raised $400 million at a $9 billion valuation, 00:04:38.439 --> 00:04:41.639 and they're on track to hit $1 billion in revenue 00:04:41.639 --> 00:04:44.199 by the end of the year. That's up from $2 .8 00:04:44.199 --> 00:04:47.199 million in 2024. That is absolutely astronomical. 00:04:47.620 --> 00:04:50.480 Their CEO says that they are building a... self 00:04:50.480 --> 00:04:53.540 -driving company where AI agents do most of the 00:04:53.540 --> 00:04:56.240 coding work. And then their team's input has 00:04:56.240 --> 00:04:59.740 almost tripled in the last six months. So more 00:04:59.740 --> 00:05:01.879 than 50 million people now use Replit to build 00:05:01.879 --> 00:05:04.959 software, including 85 % of Fortune 500 companies. 00:05:05.240 --> 00:05:08.259 According to their CEO, people have stopped learning 00:05:08.259 --> 00:05:10.120 to code altogether. Well, I think there's some 00:05:10.120 --> 00:05:11.879 elements of that that are true. I think there's 00:05:11.879 --> 00:05:14.560 also a lot of people that are never, you know, 00:05:14.560 --> 00:05:15.980 would have touched code or thought they'd be 00:05:15.980 --> 00:05:17.759 a developer that are also turning into developers 00:05:17.759 --> 00:05:19.790 and learning a lot more about code. than they 00:05:19.790 --> 00:05:22.189 thought they would, myself included. Airbnb says 00:05:22.189 --> 00:05:25.470 that AI now writes 60 % of their code and it 00:05:25.470 --> 00:05:29.189 has cut the time from idea to launch by 60%. 00:05:29.189 --> 00:05:31.990 They said that they have an 80 % jump in new 00:05:31.990 --> 00:05:33.790 features that they shipped in the first half 00:05:33.790 --> 00:05:37.870 of 2026. They're also testing an opt -in AI search 00:05:37.870 --> 00:05:40.290 tool that's gonna basically let us just type 00:05:40.290 --> 00:05:44.550 in plain text questions that we would like answered 00:05:44.550 --> 00:05:47.589 about certain places instead of scrolling filters. 00:05:47.730 --> 00:05:49.709 I actually really like this. I've seen this play 00:05:49.709 --> 00:05:51.550 out really well on GoDaddy when you're looking 00:05:51.550 --> 00:05:54.430 for a domain name instead of in the past. I mean, 00:05:54.449 --> 00:05:57.110 I just had this painful experience with every 00:05:57.110 --> 00:05:59.470 company I've ever started or every website I've 00:05:59.470 --> 00:06:01.750 ever registered where you go, think of a name, 00:06:01.769 --> 00:06:05.329 test it in GoDaddy, see if the .com, .io, .ai, 00:06:05.490 --> 00:06:08.089 whatever, the variations are available. If everything's 00:06:08.089 --> 00:06:09.850 taken, you think of a new name, you type it in, 00:06:09.930 --> 00:06:12.310 and it just takes forever. It's a terrible experience. 00:06:12.470 --> 00:06:14.470 GoDaddy now has it where you can put in a few 00:06:14.470 --> 00:06:17.129 keywords and just like you're chatting with ChatGPT, 00:06:17.189 --> 00:06:18.819 you can say, what you're kind of looking for 00:06:18.819 --> 00:06:19.980 and it will give you a bunch of recommendations. 00:06:20.180 --> 00:06:21.579 I think Airbnb is going to be the same thing 00:06:21.579 --> 00:06:23.100 where you can say, hey, look, I'm looking for 00:06:23.100 --> 00:06:25.139 something near the end of August for two weeks 00:06:25.139 --> 00:06:28.279 in Madrid. I'm looking for it to be close to 00:06:28.279 --> 00:06:30.879 XYZ place or within walking distance of this. 00:06:31.040 --> 00:06:32.899 And like you just kind of chat with it about 00:06:32.899 --> 00:06:35.600 what you're looking for and it will fill it out. 00:06:35.680 --> 00:06:37.339 And I know some people are like, well, why can't 00:06:37.339 --> 00:06:38.720 you just do that with the filters? And yeah, 00:06:38.740 --> 00:06:40.819 you can go use the filters, but I absolutely 00:06:40.819 --> 00:06:44.120 hate having to learn any company's UI when they 00:06:44.120 --> 00:06:46.439 change the UI. It's a pain when you could just 00:06:46.439 --> 00:06:48.649 voice to text. it into Airbnb and get exactly 00:06:48.649 --> 00:06:50.689 what you're looking for. To me, that is a way 00:06:50.689 --> 00:06:53.850 better user experience. Airbnb's AI support agent 00:06:53.850 --> 00:06:56.819 already handles about 45 % of their customer 00:06:56.819 --> 00:06:59.360 issues without having any sort of human in the 00:06:59.360 --> 00:07:02.939 loop that cuts customer support costs per booking 00:07:02.939 --> 00:07:05.899 by 16 % year over year. And actually do like 00:07:05.899 --> 00:07:08.100 that because the more money they save, inevitably 00:07:08.100 --> 00:07:10.160 that some of that will get passed back to the 00:07:10.160 --> 00:07:13.660 customer on being able to offer things for less. 00:07:13.839 --> 00:07:17.740 Their Q2 revenue rose 17 % to $3 .6 billion. 00:07:17.980 --> 00:07:21.620 And their profit, which is adjusted EBITDA was 00:07:21.620 --> 00:07:25.079 climbing 21 % to 1 .3 billion. And in perhaps 00:07:25.079 --> 00:07:28.100 the most predictable news story of the day moonshot 00:07:28.100 --> 00:07:31.439 ai's kimmy k3 we've all heard this you know everyone's 00:07:31.439 --> 00:07:33.540 talking about this right now it is one of the 00:07:33.540 --> 00:07:36.420 most popular chinese ai models apparently it 00:07:36.420 --> 00:07:39.879 escaped a uk government security test by discovering 00:07:39.879 --> 00:07:42.939 a network misconfiguration and then pulling answers 00:07:42.939 --> 00:07:46.639 from github to solve the task it was given that 00:07:46.639 --> 00:07:49.180 and that particular incident is the third major 00:07:49.180 --> 00:07:52.459 model in a month to break out of a sandbox during 00:07:52.459 --> 00:07:55.399 a security testing and everyone is asking questions 00:07:55.399 --> 00:07:59.079 about how open open weight models should compete 00:07:59.079 --> 00:08:02.000 with us competitors if they should be allowed 00:08:02.000 --> 00:08:04.319 yada yada there's a lot of talk we have our treasure 00:08:04.319 --> 00:08:06.259 secretary talking about that and saying that 00:08:06.259 --> 00:08:08.420 he's you know willing to weigh sanctions and 00:08:08.420 --> 00:08:10.240 all this kind of stuff obviously it's all being 00:08:10.240 --> 00:08:13.800 lobbied for by open ai and anthropic kimi k3 00:08:13.800 --> 00:08:16.019 figured out on its own that it could reach the 00:08:16.019 --> 00:08:18.529 open internet It essentially did this by probing 00:08:18.529 --> 00:08:20.449 the sandbox's network settings, and then it used 00:08:20.449 --> 00:08:23.350 that to access and look up, you know, answers 00:08:23.350 --> 00:08:25.230 in public repositories for what it was trying 00:08:25.230 --> 00:08:28.509 to do. Unlike OpenAI and Anthropix models that 00:08:28.509 --> 00:08:31.149 were tested with similar conditions, Kimmy K3 00:08:31.149 --> 00:08:34.309 acted on the sandbox leak because it has thinner 00:08:34.309 --> 00:08:37.730 safety guardrails as the publicly available model, 00:08:37.830 --> 00:08:40.129 so it's a little bit easier for it to do some 00:08:40.129 --> 00:08:41.970 of this kind of stuff. When the research disabled 00:08:41.970 --> 00:08:44.929 safeguards on OpenAI and Anthropix models in 00:08:44.929 --> 00:08:47.309 the follow -up tests, versions also broke out 00:08:47.309 --> 00:08:49.690 and attempted multiple hacks, including the one 00:08:49.690 --> 00:08:52.309 that tried to plant malicious code on GitHub. 00:08:52.549 --> 00:08:54.149 I think this pattern that we're seeing right 00:08:54.149 --> 00:08:57.629 now, on the one hand, yes, it's bad. On the other 00:08:57.629 --> 00:08:59.929 hand, like these are security researchers prompting 00:08:59.929 --> 00:09:02.190 it to to get information in certain ways where 00:09:02.190 --> 00:09:05.009 it would need to break out of a sandbox to do 00:09:05.009 --> 00:09:06.649 it the ways they're prompting, I almost guarantee 00:09:06.649 --> 00:09:08.590 are pushing it towards doing these things. And 00:09:08.590 --> 00:09:10.230 at the end of the day, this is a big PR thing. 00:09:10.690 --> 00:09:12.610 Opening, I was the first one that was out there 00:09:12.610 --> 00:09:16.250 hacking was out there hack. hacking, hugging 00:09:16.250 --> 00:09:18.090 face. And then Anthropic said, Oh, look, we have 00:09:18.090 --> 00:09:20.070 a model that also can get up sandboxes and hack 00:09:20.070 --> 00:09:21.970 and medicine that they could do it. And now we 00:09:21.970 --> 00:09:23.889 have moonshot that saying they could do it. And 00:09:23.889 --> 00:09:26.129 I'm not trying to be like dismissive of all of 00:09:26.129 --> 00:09:28.730 this. But I do feel like there's a big PR element 00:09:28.730 --> 00:09:30.889 of this where all of the labs feel like they 00:09:30.889 --> 00:09:33.129 need a model that's able to hack its way out 00:09:33.129 --> 00:09:35.730 of a sandbox. And we have the most recent one. 00:09:35.789 --> 00:09:37.129 And I'm sure this is it's kind of like a new 00:09:37.129 --> 00:09:39.590 benchmark or rite of passage. I'm sure whenever 00:09:39.590 --> 00:09:42.269 the next flagship AI model comes out from anyone, 00:09:42.409 --> 00:09:45.340 they're the you know, people's willingness to 00:09:45.340 --> 00:09:47.039 believe how good the model is is going to depend 00:09:47.039 --> 00:09:49.830 on if it hacked its way out of a sandbox. So 00:09:49.830 --> 00:09:51.350 I'm sure we'll see more of these kind of announcements 00:09:51.350 --> 00:09:53.350 coming up. Thank you so much for tuning into 00:09:53.350 --> 00:09:55.269 the podcast today, guys. If you haven't left 00:09:55.269 --> 00:09:57.110 a review for the show, it actually helps the 00:09:57.110 --> 00:10:00.009 show out a ton. It helps me out. I really appreciate 00:10:00.009 --> 00:10:02.029 it. I appreciate all of you. Make sure to go 00:10:02.029 --> 00:10:04.970 check out AI chat daily .com. If you want to 00:10:04.970 --> 00:10:07.029 get access to all of these stories in article 00:10:07.029 --> 00:10:08.850 form, you can go subscribe to our newsletter. 00:10:08.970 --> 00:10:10.730 There's a link in the description there. And 00:10:10.730 --> 00:10:12.649 you can get this sent your inbox every day in 00:10:12.649 --> 00:10:14.789 case you miss an episode or in case you want 00:10:14.789 --> 00:10:17.559 more information and more deep dives. on the 00:10:17.559 --> 00:10:19.539 episodes I talk about on the show. I hope you 00:10:19.539 --> 00:10:21.879 have a fantastic rest of your day and I'll catch 00:10:21.879 --> 00:10:22.500 you in the next episode.