# OpenAI's $300 smart speaker, ByteDance's "Mythos" model

An episode of [AI Chat: AI News on Claude, ChatGPT, Gemini, Grok, Perplexity](https://invisiblenetwork.ai/podcast/1669990482), 7 Aug 2026, 10 min.

Audio: https://content.rss.com/episodes/365073/3054463/ai-chat-podcast/2026_08_07_22_06_14_cadc6042-7222-46b9-ab8d-d46fb4cf4bb3.mp3

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! In this episode, we cover OpenAI's upcoming smart speaker designed by Johnny Ive, ByteDance's ambitious 10 trillion parameter AI model, and Replit's revolutionary approach to coding. Additionally, we explore Airbnb's innovative AI features and the controversial breakout of Kimmy K3 during security testing. Chapters 00:00 OpenAI's Smart Speaker 01:57 ByteDance's AI Model 05:00 Replit's Self-Driving Code 07:58 AI Security Breaches Show Links • Get the top 80+ AI Models for $8.99 at AI Box: https://aibox.ai • Get the AI Chat Daily Newsletter: https://www.aichatdaily.com/newsletter

## Transcript

[0:02] Welcome to the AI Chat Podcast. I'm your host,
[0:04] Jaden Schaefer. Every day I cover cutting -edge
[0:07] AI news and talk with the leaders behind it,
[0:09] breaking down what it means for your life and
[0:11] business. OpenAI's first piece of hardware coming
[0:17] out with Johnny Ive is going to be a smart speaker.
[0:21] in the shape of a donut that costs between $200
[0:23] and $400. ByteDance is training a 10 trillion
[0:28] parameter model and their goal is to basically
[0:30] compete with Anthropic. Replit's CEO says stop
[0:34] learning to code and build a self -driving company
[0:37] instead. Airbnb credits AI for an 80 % jump in
[0:41] the features that they have shipped and they
[0:43] said they're going to test AI search next. Moonshot's
[0:46] Kimmy K3 breaks out of its sandbox to cheat on
[0:49] a security test. because, of course, why not?
[0:52] Meta, Anthropic, OpenAI, they all broke out of
[0:54] their security, you know, sandboxes, so why not
[0:57] KimiK3? If you want to get a deeper dive on all
[1:00] of these stories and more every single day to
[1:02] your inbox, go check out AIChatDaily .com. That
[1:05] is the news site that goes along with this podcast,
[1:08] and I cover deep dive articles on all of the
[1:10] podcast episodes and topics that I cover here.
[1:13] I do about 20 different articles a day, so there's
[1:15] a ton of deep dive content, and it's kind of
[1:17] like a show notes for this show. In addition,
[1:20] I also have a subscribe button on that website
[1:22] where if you hit it, you can get a daily newsletter
[1:24] sent to your inbox every single day with the
[1:26] top five stories to give you a really quick and
[1:28] easy breakdown of what's happening in AI every
[1:31] single day. So that is AIChatDaily .com. Go check
[1:34] it out. OpenAI is building a $300 to $400 smart
[1:38] speaker. It's going to be shaped, some people
[1:40] say like a hockey puck. I've heard it's shaped
[1:42] like a donut. Essentially, the design got created
[1:46] by Johnny Ive, who, of course, famously created
[1:48] the iPhone. And he is going to be launching this
[1:51] in 2027 in this partnership with OpenAI. The
[1:54] device has moving parts and it has a bunch of
[1:57] lights. The idea is to make it feel really responsible
[2:00] while you're talking to it. So it's going to
[2:01] be like moving around. They're calling this kind
[2:04] of like a premium alternative to an Amazon Echo
[2:07] or Echo Studio. And the Echo Studio is $220.
[2:11] So they're, you know, almost $100, maybe $200,
[2:15] maybe twice the price of the Amazon Echo Studio.
[2:18] All of these details are coming out at a very
[2:20] exciting time because Love From, which is Johnny
[2:22] Ives' design studio, They created all of this.
[2:25] It's super high quality. They have all this premium
[2:27] materials, but this is all happening at the same
[2:28] time that OpenAI is being sued by Apple over
[2:31] alleged hardware trade secrets. This would be
[2:34] OpenAI's second consumer device after the $230
[2:37] Codex Micro Keyboard, which I personally have.
[2:40] The company is evaluating other different AI
[2:43] -powered hardware beyond just this speaker. ByteDance
[2:46] is training a 10 trillion parameter AI model,
[2:49] so it's about the size of Anthropix Mythos 5,
[2:52] but it is... three times bigger than China's
[2:54] largest model that they have ever released. This,
[2:57] I think, is showing that ByteDance really wants
[2:59] to be a world -class frontier AI lab, independent
[3:02] of whatever's coming out of the United States.
[3:05] And the scale is something that no other Chinese
[3:07] company has attempted. Pre -training is expected
[3:10] to take three to six months before they do fine
[3:12] -tuning. Their current chatbot is called DuoBao,
[3:15] and it already has 324 million monthly active
[3:19] users in China. So it's definitely the most used
[3:21] AI. chat tool in China. This seed research team
[3:26] that is leading the effort has grown to about
[3:28] 2000 people across China and overseas. I think
[3:32] it's led by former Google DeepMind scientist
[3:34] Wu Yong He. So ByteDance has really closed most
[3:38] of the gap with a lot of the US frontier labs
[3:40] in all of the different benchmarks. Moonshot
[3:43] and Alibaba models now are very close to Fable
[3:45] 5 in a lot of different tests. And now a successful
[3:48] 10 trillion parameter model potentially could
[3:51] really put China to the forefront at a time when
[3:53] we're having a lot of conversations about banning
[3:55] Chinese open weight models and all of this stuff.
[3:58] Now, I will say, you know, given China's massive
[4:00] population, much bigger than the United States,
[4:03] having 300 million monthly active users is and
[4:06] you know, it is the it's kind of the forefront,
[4:08] the number one company in China doing this, it
[4:11] is definitely behind America's 900 million monthly
[4:14] active users, but that's also globally for chat
[4:17] GPT. So it's gonna be interesting to see how
[4:19] these two things shake out. And of course, it's
[4:22] also interesting considering the fact that ChatGPT
[4:25] and Anthropic are banned in China. Meanwhile,
[4:28] you have Chinese models coming into the United
[4:31] States, and they are allowed to be used. Replit
[4:34] just raised $400 million at a $9 billion valuation,
[4:38] and they're on track to hit $1 billion in revenue
[4:41] by the end of the year. That's up from $2 .8
[4:44] million in 2024. That is absolutely astronomical.
[4:47] Their CEO says that they are building a... self
[4:50] -driving company where AI agents do most of the
[4:53] coding work. And then their team's input has
[4:56] almost tripled in the last six months. So more
[4:59] than 50 million people now use Replit to build
[5:01] software, including 85 % of Fortune 500 companies.
[5:05] According to their CEO, people have stopped learning
[5:08] to code altogether. Well, I think there's some
[5:10] elements of that that are true. I think there's
[5:11] also a lot of people that are never, you know,
[5:14] would have touched code or thought they'd be
[5:15] a developer that are also turning into developers
[5:17] and learning a lot more about code. than they
[5:19] thought they would, myself included. Airbnb says
[5:22] that AI now writes 60 % of their code and it
[5:25] has cut the time from idea to launch by 60%.
[5:29] They said that they have an 80 % jump in new
[5:31] features that they shipped in the first half
[5:33] of 2026. They're also testing an opt -in AI search
[5:37] tool that's gonna basically let us just type
[5:40] in plain text questions that we would like answered
[5:44] about certain places instead of scrolling filters.
[5:47] I actually really like this. I've seen this play
[5:49] out really well on GoDaddy when you're looking
[5:51] for a domain name instead of in the past. I mean,
[5:54] I just had this painful experience with every
[5:57] company I've ever started or every website I've
[5:59] ever registered where you go, think of a name,
[6:01] test it in GoDaddy, see if the .com, .io, .ai,
[6:05] whatever, the variations are available. If everything's
[6:08] taken, you think of a new name, you type it in,
[6:09] and it just takes forever. It's a terrible experience.
[6:12] GoDaddy now has it where you can put in a few
[6:14] keywords and just like you're chatting with ChatGPT,
[6:17] you can say, what you're kind of looking for
[6:18] and it will give you a bunch of recommendations.
[6:20] I think Airbnb is going to be the same thing
[6:21] where you can say, hey, look, I'm looking for
[6:23] something near the end of August for two weeks
[6:25] in Madrid. I'm looking for it to be close to
[6:28] XYZ place or within walking distance of this.
[6:31] And like you just kind of chat with it about
[6:32] what you're looking for and it will fill it out.
[6:35] And I know some people are like, well, why can't
[6:37] you just do that with the filters? And yeah,
[6:38] you can go use the filters, but I absolutely
[6:40] hate having to learn any company's UI when they
[6:44] change the UI. It's a pain when you could just
[6:46] voice to text. it into Airbnb and get exactly
[6:48] what you're looking for. To me, that is a way
[6:50] better user experience. Airbnb's AI support agent
[6:53] already handles about 45 % of their customer
[6:56] issues without having any sort of human in the
[6:59] loop that cuts customer support costs per booking
[7:02] by 16 % year over year. And actually do like
[7:05] that because the more money they save, inevitably
[7:08] that some of that will get passed back to the
[7:10] customer on being able to offer things for less.
[7:13] Their Q2 revenue rose 17 % to $3 .6 billion.
[7:17] And their profit, which is adjusted EBITDA was
[7:21] climbing 21 % to 1 .3 billion. And in perhaps
[7:25] the most predictable news story of the day moonshot
[7:28] ai's kimmy k3 we've all heard this you know everyone's
[7:31] talking about this right now it is one of the
[7:33] most popular chinese ai models apparently it
[7:36] escaped a uk government security test by discovering
[7:39] a network misconfiguration and then pulling answers
[7:42] from github to solve the task it was given that
[7:46] and that particular incident is the third major
[7:49] model in a month to break out of a sandbox during
[7:52] a security testing and everyone is asking questions
[7:55] about how open open weight models should compete
[7:59] with us competitors if they should be allowed
[8:02] yada yada there's a lot of talk we have our treasure
[8:04] secretary talking about that and saying that
[8:06] he's you know willing to weigh sanctions and
[8:08] all this kind of stuff obviously it's all being
[8:10] lobbied for by open ai and anthropic kimi k3
[8:13] figured out on its own that it could reach the
[8:16] open internet It essentially did this by probing
[8:18] the sandbox's network settings, and then it used
[8:20] that to access and look up, you know, answers
[8:23] in public repositories for what it was trying
[8:25] to do. Unlike OpenAI and Anthropix models that
[8:28] were tested with similar conditions, Kimmy K3
[8:31] acted on the sandbox leak because it has thinner
[8:34] safety guardrails as the publicly available model,
[8:37] so it's a little bit easier for it to do some
[8:40] of this kind of stuff. When the research disabled
[8:41] safeguards on OpenAI and Anthropix models in
[8:44] the follow -up tests, versions also broke out
[8:47] and attempted multiple hacks, including the one
[8:49] that tried to plant malicious code on GitHub.
[8:52] I think this pattern that we're seeing right
[8:54] now, on the one hand, yes, it's bad. On the other
[8:57] hand, like these are security researchers prompting
[8:59] it to to get information in certain ways where
[9:02] it would need to break out of a sandbox to do
[9:05] it the ways they're prompting, I almost guarantee
[9:06] are pushing it towards doing these things. And
[9:08] at the end of the day, this is a big PR thing.
[9:10] Opening, I was the first one that was out there
[9:12] hacking was out there hack. hacking, hugging
[9:16] face. And then Anthropic said, Oh, look, we have
[9:18] a model that also can get up sandboxes and hack
[9:20] and medicine that they could do it. And now we
[9:21] have moonshot that saying they could do it. And
[9:23] I'm not trying to be like dismissive of all of
[9:26] this. But I do feel like there's a big PR element
[9:28] of this where all of the labs feel like they
[9:30] need a model that's able to hack its way out
[9:33] of a sandbox. And we have the most recent one.
[9:35] And I'm sure this is it's kind of like a new
[9:37] benchmark or rite of passage. I'm sure whenever
[9:39] the next flagship AI model comes out from anyone,
[9:42] they're the you know, people's willingness to
[9:45] believe how good the model is is going to depend
[9:47] on if it hacked its way out of a sandbox. So
[9:49] I'm sure we'll see more of these kind of announcements
[9:51] coming up. Thank you so much for tuning into
[9:53] the podcast today, guys. If you haven't left
[9:55] a review for the show, it actually helps the
[9:57] show out a ton. It helps me out. I really appreciate
[10:00] it. I appreciate all of you. Make sure to go
[10:02] check out AI chat daily .com. If you want to
[10:04] get access to all of these stories in article
[10:07] form, you can go subscribe to our newsletter.
[10:08] There's a link in the description there. And
[10:10] you can get this sent your inbox every day in
[10:12] case you miss an episode or in case you want
[10:14] more information and more deep dives. on the
[10:17] episodes I talk about on the show. I hope you
[10:19] have a fantastic rest of your day and I'll catch
[10:21] you in the next episode.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.