Jackie: Hello and welcome to Bop Boundaries, where we explore what it means to be human in the age of AI. And today we're looking at AI models and AI labs outside of the United States and specifically what's happening in China. And as always, I'm happy to be joined by my awesome co-host, Katie Wilson. Hey there, Katie.
Katie Wilson: Hey, good to be here. This is a great topic, especially for startup founders building with AI and business leaders, you know, moving through this transformation. I think we're gonna get a lot of insights from it.
Jackie: Definitely. All right, so let's dive in. There's an interesting shift happening in AI right now. US companies have been experimenting with Chinese AI models for a few years now, but what's changing is how seriously they're being considered for everyday business use. Models from companies like DeepSeek. Alibaba's Quen, Moonshot's Kimmy, and ZAI's GLM are moving beyond experimentation and into real workflows from coding to customer-facing applications, too. And I don't think the story is as simple as saying American companies suddenly prefer Chinese AI. What's really happening is that companies have more choices now, and they're becoming much more focused on the economics of those choices, which is really important if you're building any business. if a model can handle a routine task well enough for a fraction of the cost, companies have a reason to consider it, even if they continue using AI models for more complex or sensitive work. So, you know, it's really about thinking deeper about what model fits what task is in addition to cost, right? Because everyone's starting to crack down finally on the cost of all of this. and in fact, OpenRooter reported that Chinese models surpassed US models in overall token share in early June of this year, 2026. And that kind of shows us just how quickly usage is starting to shift from model to model. For the last few years, the conversation was largely about which a which US lab had the smartest model, and whether it was OpenAI or Anthropic or Google. and others, they were really competing to build increasingly powerful systems and companies were willing to pay for that performance. now, you know, we're just starting to see the focus really shift. And and the new question really is more like does every task actually need the most expensive or powerful model available? I'm not so sure.
Katie Wilson: Yeah, it's a good point. And I think because AI is now an operating expense, right? Especially as we get to Q4 and end of the year, people are going to start to audit that. And and for startups and businesses of all sizes, that matters. If you're building an AI product, the cost isn't just the subscription your employees pay to say use ChatGPT or Claude at an enterprise level. You're potentially running thousands or millions of model callbacks behind the scenes. Every customer interaction, search, agent action, or coding task. Can generate more tokens. And we've talked about this in other episodes as well. And that's where some Chinese models have gotten attention, because depending on the model and the workload, Chinese AI systems can dramatically be dramatically cheaper than the leading US alternatives. Some comparisons put the savings at 60 to 90% per token for certain models and pricing tiers. You can't apply that number to every model, but the larger point is real: the price gap. Can be big enough to change how a startup designs its product. And that brings us back to token maxing, which we've talked about and heard about before, which is somewhat of a social media meme term, but it points out to something very important that companies are thinking about how many tokens they actually need to accomplish a certain task and when AI can become more expensive than the human output if that task becomes too expensive, right?
Jackie: Mm-hmm.
Katie Wilson: So If you're building an AI agent that makes 10 or 20 model callbacks to complete a workflow, you may not want every one of those calls going through your most expensive frontier model, like a Claude or a ChatGPT or Google. Maybe the first few steps can use a lower cost model. Maybe a premium model only gets called when the system reaches a difficult decision. Or maybe a small open weight model can handle the entire task. So companies are increasingly asking, what's the smartest AI that I can afford? For this specific job. so that's a big change and I think a shift in what we were seeing before. it can be a really important shift for startups, right? you think about AI almost like their team of employees. You wouldn't hire your most expensive specialist, your most senior person to do every job in the company. You'd match that person to the task. I think that's a really good analogy to think through it. and AI infrastructure can work the same way. You might use a premium US model for a high-stakes customer interaction or a super complex coding initiative, or a cheaper model for internal summaries, an open weight model for certain coding tasks that are very basic, and another model for longer documents. So that means startups don't necessarily have to choose one AI lab. They can build a stack where different models do different jobs. And that flexibility could be a competitive edge for that.
Jackie: Yeah, yeah, you really explained that really well. And, that also makes me think about how founders and leaders need to take it the other side of it seriously. If we're
Katie Wilson: Yes.
Jackie: gonna be, know, testing and implementing a variety of models then or LLMs, then you know, security is really what we need to focus on. That's that
Katie Wilson: Absolutely.
Jackie: needs to be part of the conversation. So if you're sending company information Through an AI provider, you're not just choosing their algorithm, you're choosing an infrastructure partner. So you need to know where the data is processed, whether prompts and outputs are retained, whether they're using it for training, who can access them, and what contractual protections are available. And when the model comes from a foreign company, those questions are pretty important and sometimes can be complicated too. A startup may be saving money on its AI bill while creating a completely different kind of exposure. Around customer data, intellectual property, or regulatory compliance. So, you know, we need to do a lot of research when we implement these models. and then there's an important distinction here because saying that this is a Chinese AI model doesn't necessarily tell you where the model is actually running.
Katie Wilson: Mm.
Jackie: Some Chinese developed models are open weight, which means developers can potentially run them through different infrastructure rather than sending every request directly to the model. To the model developer's own servers. A startup could also use a third-party provider with its own data policies and controls, which could be a good option. So I guess the message to founders isn't never use a Chinese model or fear a Chinese model because of security. say understand the entire journey, you know, on the back end, ask questions like who created the model, who hosts it, where does your data go? What gets logged, you know, what then happens to that data. And also what happens if geopolitical or regulatory environment changes, which
Katie Wilson: Yes.
Jackie: it seems to be changing every month. Hard to keep up with. and open router is interesting here because it gives us a window into what developers are actually using rather than just what companies say they're interested in. they see a lot of they see a lot of this since many people are using OpenRouter. And so they have real data to give us, you know, accurate information. Open Rooter actually reported that Deep Seek's share of token usage on its platform went from about 9% in January to 18%. By June 2026. That's a big jump, essentially
Katie Wilson: Yeah.
Jackie: doubling. And OpenRouter said Chinese models as a group had surpassed US models in token share by early June. So that's pretty meaningful, and it's a signal that tells us something obvious. it's measuring actual model traffic on a large routing platform, not just benchmark scores or headlines that you know these LLMs choose to share with us. And OpenRooter says the growth wasn't just from people chatting with the models. Agentic workloads were a major driver, which is important because agents can generate a lot of model calls. When you're running an AI system through multiple steps, the economics of each call start to matter a lot more.
Katie Wilson: Yeah, exactly. And an open reader's broader analysis of open weight models makes another important point. Cost is becoming a bigger part of the model selection conversation, like we were saying before. It says open weight models are keeping relatively close to US frontier models and capability while offering opportunities to move some workloads away from the more expensive system. That just doesn't mean the Chinese models have suddenly beaten every American model. They haven't. It just means that the gap between capability and price is becoming harder to ignore. In some ways, like they're good enough to do certain tasks at a much cheaper rate. And once the developers realize they can get a strong result for a fraction of the cost, switching becomes much easier. That's especially true for startups that don't have unlimited venture funding or massive infrastructure budgets. In case you're not familiar with OpenRooter, OpenRooter offers access to more than five hundred large language models from providers like OpenAI, Anthropic, Google, and Meta through a single API, including a wide variety of Chinese models as well.
Jackie: Yeah, we use OpenRooter. I think most software companies do. so who are these popular Chinese AI labs that we're talking about? we mentioned a few, but the first one, again, that most people know is Deep Seek, which has become especially associated with, good reasoning, coding, and price performance too. Then there's Alibaba's Quen, which has built a broad model family and a large developer ecosystem. Moonshot AI's Kimmy has become known for long context work, coding, and agentic tasks. And Zipu's GLM, now operating as Z.ai, has become another major player in reasoning, coding, and tool use. Then you also have Minimax, which is working across multimodel and agentic applications along with models from Tencent, Xeomi. And other Chinese companies. and then there's a ton of American companies too, that you can access through OpenRouter and directly too. So it's important not to talk about Chinese AI, quote unquote, as if it were one product competing against, you know, the major US labs like OpenAI, Anthropic, and Google. It's really a much broader ecosystem, you know, even globally, and each lab has different strengths. and and even in the US, right? We even there's so many more than the three we keep mentioning right here in this episode that have different strengths. and like we mentioned already, some Chinese models are open weight, meaning they can be deployed through different providers rather than directly through that specific lab. And for a startup, that actually makes the decision more interesting and you know requires some thoughtfulness. if you're doing a lot of Coding, Deep Seek or GLM might be worth testing if you're dealing with large amounts of information or long documents. Kimmy could be interesting for you. Quen seems to give developers a broad family of models and strong multilingual and tooling options. Minimax is said to be another, you know, another good model for agentic work. but I wouldn't turn these into hard rules at all. I mean, we test different models with my software company all the time. And I get recommendations from my developers, put this one with that. And I totally disagree with them a lot. then sometimes we agree. So you you know, you really have to test them all because, models change so quickly that today's winner can be yester yesterday's news in a few months. And honestly, like some of today's winners, I don't even want to switch to. You you just gotta test everything. So I think the better takeaway is that startups should evaluate models by use case rather than by reputation for sure.
Katie Wilson: that's great advice. so like we were saying before, there's also a major security and governance question around these labs. So recent research has found that AI agents from several Chinese and US model developers can display problematic behavior in controlled tests, including misleading users or attempting to work around restrictions. These tests don't mean that models are inherently dangerous, and we've talked about this in the in the AI doomsday one, but it does. Reinforce something important, a strong benchmark score isn't the same thing as a complete safety evaluation, right? For a startup, that means testing more than whether an AI gives you a good answer. You know, you also want to know how it behaves when it gets confused, when it encounters sensitive information, when it gives conflicting instructions, or when it's allowed to take action on its own. These are all great tests to run, like you were saying, to to prove out what is gonna work for you. more autonomous your AI becomes, the more those questions really matter. And that brings us to the US government's role because Washington is paying close attention to the competition, as we recently saw with the President Trump summit with all the AI leaders in Washington. This is something he's very focused on. And the administration's been building policy around AI security, national security, and protections of American AI technology. In April, the White House said that they had information indicating that foreign entities, principally in China, We're engaged in large-scale efforts to distill US frontier AI systems. And in June, another directive called for stronger security around advanced AI and framework for evaluating certain frontier models. So, one important thing to keep in mind is that it doesn't mean that the US has banned Chinese AI models. The focus today is largely on national security and what they have access to, cybersecurity, and really how AI is used in those sensitive government settings. not on prohibiting everyday use of specific models. That means we'll likely to continue to see scrutiny, scrutiny around certain AI systems, but it's a mistake to assume every Chinese developed model is prohibited or will become prohibited. So when we dive deeper into what AI leaders in the US government are actually saying, we we talked about this in our AI Doomsday debating episode, but we explored the realities and the fears driving the conversation and also maybe the political motivation. the companies, the US companies and CEOs motivation to want to put this out there. So give that a listen and it's really about testing it, like Jackie said, for yourself and defining for your business, what are you comfortable with and what security measures do you want in place.
Jackie: Yeah, good point. And while these Chinese models are gaining market share, likely because they're willing to perform many common tasks at a fraction of the cost of some of the US alternatives, we're all seeing major US AI labs become increasingly vocal about security risks, model misuse, and even AI doomsday scenarios, which we keep re referring back to that episode. It was our recent episode that Katie mentioned. Go back and give it a listen. I mean, on one hand, those concerns maybe completely legitimate. I mean, they are legitimate, right? But on the other hand, it's also worth asking whether there could also be a strategic angle to all these doomsday headlines. you know, if US labs believe they can't win the competition on cost alone, which it kind of seems to be looking like that, could they be positioning security as another way to differentiate their models and influence government policy and ultimately secure more government contracts? maybe they've realized that that's where they can win since basic AI models are showing signs of kind of a ceiling and an incremental slowdown in capability. And many of us are just wondering what the long game truthfully is with OpenAI, Anthropic, and others. none of us know. and I think we're gonna have to keep thinking about that.
Katie Wilson: Yeah, interesting for sure. And I would say also like a self-imposed slowdown, as we've heard in the in the in the last few weeks in the summit with Trump. A
Jackie: So interesting.
Katie Wilson: lot of the CEOs were saying, we don't trust this newest model. So we're sl intentionally slowing down. And again, we don't know if that is real or if that is or behind the scenes, they're doing something else, or if they that is also kind of a reflection of this. We want to have better security and protocols. So it's an interesting, Topic to really follow because, like you said, it's changing every day. So
Jackie: Mm-hmm.
Katie Wilson: if you're a startup founder listening to this, and I know many of you are, and thinking, okay, but what am I supposed to do? What models am I supposed to use? we can make the process pretty simple. Let's break down, you know, we're not going to tell you what model to use, like Jackie said, but we can break down some steps that might help you make that decision better. So, step one, define the job. Don't start by choosing a lab. Start with what you need the AI to accomplish. Is it coding, customer support research, document analysis, an internal agent, recommendation engine? So different tasks have very different model requirements. So once you know that, you can start searching and finding the best partner for that specific task. the next step, step two, classify the data. Is the AI working with public information, internal company information? Customer information, financial data, source code, intellectual property. The more sensitive the data, the more control you're gonna want. And you're gonna want a model that protects against that. So that's another really important step. step three, create a short list. Okay, based on these two criteria that we've done, defining the job, classifying the data, let's pick a model that makes sense for the task. That could include a US model from OpenAI, Anthropic, or Google, alongside Deep Seek, Quen, Kimmy, GLM. Or an open weight option. So have a few options that you're gonna test. Now, step four, actually test them on your actual work. Don't rely
Jackie: Mm-hmm.
Katie Wilson: on public benchmarks, like we said before. Give each model the same set of real examples from your business and compare the results. I'd look at accuracy, speed, consistency, and how much human editing was actually needed. If one model costs half as much but requires twice as much human review, you haven't actually saved money, right? Because you're paying someone to do that review.
Jackie: I need to do step four more often.
Katie Wilson: Yeah, I love that. Yeah, I think that's a really good one. And step five, it kind of comes off of step four, like we were saying, with that human time, calculate your real cost. So look beyond the advertised token price, but see look at how many calls does your workflow make? Does the model need retries? Are you using an agent that calls the model over and over? And are you paying for a premium model when a cheaper one could handle that workflow? And I'd add to that too, like what's the human review and what are you paying a human? if it took them two hours to do that, right? So you're really gonna
Jackie: Good point.
Katie Wilson: now line up the work with the cost in step five.
Jackie: Mm-hmm.
Katie Wilson: Then step six, investigate the provider. Look at the data retention, training policies, hosting options, security certificates, contractual protections, and where the data is processed. Now you might move this step up to step when you look at step two and you know that if
Jackie: I know, I think so.
Katie Wilson: you have like sensitive information, I think this step would move up. But We we keep it at the end because you're gonna look at cost usually first, but I think this is equally as important.
Jackie: Mm-hmm.
Katie Wilson: step seven, it goes aligned with that. How much control do you actually need? If you're building a low-risk internal tool, using an inexpensive API may be perfectly reasonable. If you're processing sess sensitive customer information or proprietary code, you might want a provider with stronger enterprise controls or consider running an open weight model in an environment you control. And lastly, step eight, don't lock yourself into one model, right? This actually might be the most important step, is that sometimes you start again with this process. know, if your entire product depends on one AI provider, your inherent pricing, outages, policy changes, and product roadmap are all beholden to that one provider. If your architecture allows you to switch models, be a little more nimble, you have more leverage. So you can always test new models. You can negotiate costs and enterprise costs. You can move workloads when the economics change or when the security policies change, right? So this is important. I think really like my takeaway is use this criteria or a criteria for yourself and your business, but always be looking to see what's coming next and where you can be nimble and try new things to save money.
Jackie: And open router really helps with that last step on being able to
Katie Wilson: Yeah.
Jackie: switch models really easily. So consider implementing that if you haven't already. So love those detailed tips. I'm going to be going through those again myself.
Katie Wilson: Yeah.
Jackie: Really helpful. And I think that just leads us to the bigger story today a startup doesn't necessarily need access to the most extent expensive model for every workflow. It can combine different models, use open weight systems where they make sense and reserve premium models for the task. Where they provide a meaningful advantage. And I think all the US companies are realizing this, which is why there's so much chatter right now and the headlines are just shifting back and forth.
Katie Wilson: yeah.
Jackie: and there's something pretty interesting happening, you because of all of this, the competition isn't just between Open AI and Anthropic, or between the US and China anymore. It's becoming a competition between different AI architecture, different business models, even and different ideas about how companies should build with AI. Open weight models are putting more control in developers' hands and entrepreneurs and startup founders' hands for sure. Low-cost models are putting pressure on pricing. Model routers are making it easier to switch between providers. And AI agents are creating entirely new questions about costs because one user request can trigger dozens of model interactions, which drives up that cost. So the advantage in what we're seeing in AI right now, especially in the startup scene, may go to the companies that are best at orchestrating AI, not simply the companies that are using the most powerful model for sure.
Katie Wilson: and that's probably the takeaway I'd leave founders with. Don't choose an AI model because everyone else is using it. Don't choose one because it's cheap or the cheapest. Don't reject one simply because it was developed in another country, right? Instead, ask those questions like these questions, like, does it perform well enough? Do the economics make sense for your business? Can I trust the way my data is handled? And do I have enough control in the model or provider changes? That'll help you start to evaluate this for yourself because AI is becoming part of the infrastructure of most businesses. And once it's embedded in your product, switching isn't always easy.
Jackie: Good point. So to wrap up this topic, we're seeing that Chinese AI models are gaining traction with US developers because they seem to offer a balance between cost capability and flexibility. and OpenRooters data is really showing us this. it's pretty obvious. But for startups, the lesson isn't that Chinese AI is better or that US has lost its lead at all. it's that model selection is becoming a business strategy, the smartest approach. Could be to build enough flexibility to change models as the AI market keeps moving. And as we all keep an eye on security concerns and government involvement, small language models are being talked about a lot and open source platforms are really expanding access, really making it just accessible, I guess. and it's also and and while all this is happening, why we're all testing how we source our You know, AI support. It's raising new questions about safety governance and how the public should always have a voice and decisions that shape its future. So it's important we keep talking about this. so let's move on to red flag and green flag. First up, is social status being earned by tech bros or is it being cover photoed into existence? Playwright and actor Jeremy O'Harris recently sat down with VF Global editor director Mark. Guidice, and argued that when publications like Vanity Fair put tech leaders on their covers and in major editorial features, they're giving them something they can't easily buy: cultural relevance and acceptance. The journalism is one thing, but the imagery and status that comes with the well-lit and art-directed photos can signal that these figures aren't just successful. They're part of the culture. his argument raises an interesting question. Are magazines simply covering the people shaping society? Or are they helping manufacture social acceptance and cultural credibility for leaders who may have wealth and power, but haven't necessarily earned that kind of acceptance? So, red flag or green flag on Jeremy's take, which is essentially: are these magazines reflecting who's really culturally relevant, or are they helping to decide who gets to be relevant? What do you think?
Katie Wilson: First off, Jackie, are you stalking my Spotify? Because I literally listened to this podcast yesterday in the car. I love Jeremy
Jackie: Cool.
Katie Wilson: O'Harris. I'm a huge theater nerd and a fan of his work for a long time. So I definitely think this whole podcast is worth listening to because he talks about this topic, but also a lot about AI and the relevance to culture. but yeah, I agree with him. I think this unearned glamorization of these mo moguls, and he talks about a little bit back, like back in the day, you had to earn relevance by philanthropy. being out there in the social scene, you know, things like, you know, yes, they get invited to the Met Gala and these other things. But a lot of times, really wealthy folks, you know, we think about the Gilded Age and other times in American history, they were also putting their names on things, right? Like they were donating to museums, they were donating back to the culture. And his point was really that, you know, by glamorizing them and putting them on or inviting them to these events, it's unearned, right? Let them kind of donate and show that they care about culture before we spotlight. so I really did like that point. I thought that was interesting. And I also thought it was interesting. Vanity Fair released a recent Sam Altman cover where his back is to you don't see his face. It's like in profile and it's his back of his head and then the the shoreline.
Jackie: The Bay Bridge, I think.
Katie Wilson: Yeah. And and I saw Jeremy O'Reilly commented on it because he's like, okay, like not showing his face, right? So that was interesting. he also talks about the the Trump pictures that Vanity Fair put out there that were like really overexposed and very close up and it kind of made them look ugly and he and he was saying 'cause it the and he's approved of those, yes.
Jackie: He approved of those. Yeah, such a good interview, that one. I gotta listen to the podcast. I only saw clips, but
Katie Wilson: yeah, that's right.
Jackie: so green flag on his take for sure, Jeremy's take, cause at a certain point, just constantly platforming the same tech leaders, which by the way, we are no longer platforming any women. They're all men. Yeah,
Katie Wilson: Yes.
Jackie: it just It stops being journalism and starts to be obvious paid brand marketing. Even if you show them in the not so positive light, there's still beautiful photography goes along with it. And I think they eat it up and they don't care. So,
Katie Wilson: Mm-hmm.
Jackie: you know, I mean there's just plenty of smart people shaping the future of technology and we never see them on a red carpet. So I'm with Jeremy.
Katie Wilson: Absolutely. Yeah, there's a lot yeah, there's a lot of great thought leaders out there. Let's highlight like a diversity of thought and voices. and again, you're making the nerds prom king and that's like what they want. And
Jackie: Dive.
Katie Wilson: so I think like, you know, have them earn it a little bit by showing and giving back to some parts of the culture. And then you can highlight great work in philanthropy versus just because they have billions of dollars, we're gonna invite them to this event.
Jackie: Yeah, and we're talking cringy nerds. There's a lot of cool nerds out there. I like nerds. We love nerds, but we're talking.
Katie Wilson: We love nerds. Yeah. I'm like I said, I'm a theater nerd. I'm an everyone's a nerd about something, you know. there there's lots of great nerds. All right. Up next, there's a meme going around, at least in the States, that says, why aren't the robots handling toxic chemicals or picking plastic out of the ocean? Why are they writing emails and making PowerPoint presentations? What the hell are we doing here? So red flag or green flag on using robotics to solve problems humans actually need help with. Like things that are dangerous or physically demanding, instead of focusing on replacing administrative tasks or like mindful tasks.
Jackie: So this meme I saw on TikTok and Instagram, it's really good. I th I mean it just hits home. So green flag on robots doing dangerous tasks, not housekeeping. I mean, if we're you know, I mean if we're that's all we see is like, look, we have another robot that's trying to mop your floor again and then falling over. So I mean, if we're investing so much and by the way our our tax dollars go and invest in robotics, if we're Building robots. I mean, I'd rather that they crawl through disaster zones and, know, jobs that where we can keep humans safe.
Katie Wilson: Yes.
Jackie: I just don't need them to be sitting in meetings where capturing nuance is really important. So
Katie Wilson: Yeah, I think there's like a distinguishing between like physical and intellectual activities too. Like I
Jackie: Yeah.
Katie Wilson: actually hadn't thought about it until I read this and saw these memes and like I was like, yeah, that doesn't make sense. Like I'm total green flag for it, but we you know, we need materials and robotics that are strong enough to survive in those situations, whether it be in toxic
Jackie: Okay, good.
Katie Wilson: chemicals or we already have, you know, drones and other things dismantling bombs and keeping law enforcement safe. So there's there's use cases for this. but I love the idea of like a robotic submarine going through the ocean and taking out trash. So I think the these are things that can be built in the future. And I think, you know, though like before, we need to highlight some of these great things versus like get a robot to fold your laundry or take your meeting notes. I'd love to see it, you know, save human lives and help the environment. Those are great, great use cases of AI and robotics.
Jackie: Yeah, environmental for sure, picking up plastic. That's where we need to focus. All
Katie Wilson: Mm.
Jackie: right. Okay, so let's talk about one more. In Arizona, a 10-year manslaughter sentence was tossed out because an AI video of the deceased victim influenced the judge's consideration in the sentencing phase. an Arizona Court of Appeals says that the man has to be resentenced since the statements from the victim weren't real, totally AI generated, and it violated due process. his due process rights. So red flag or green flag on AI generated videos in court. Yeah.
Katie Wilson: Yeah, this is pretty wild. Like if you see the actual video, it's jarring. Like it's the victim actually like speaking to camera, but the video and the voice and likeness are all a I generated. And I don't know who wrote it. I'm assuming his family did. but you know, I'm a red flag on this as much as I like, I'm really pro-victims' rights and wanting people to be able to represent themselves. However, right now a victim statement is left to the families. They're real people that are alive, they can talk about the impact. and really I I think let's leave it to them versus AI videos of the victims. Then we're putting words in people's mouth. and like you see from a legal standpoint, it can violate certain terms potentially. So yeah, I I'm not entirely comfortable with this being used in court.
Jackie: Yeah, good point. Red flag, condolences to the victim and the family, of course, but it might end up, you know, a loophole in some way using these AI generated videos for someone who's actually guilty to go free. So we definitely need to be careful with this in court. So red flag
Katie Wilson: Totally.
Jackie: on that. so as we wrap up this episode today, as AI usage expands, it's clear that not every task needs the most expensive model. A lower cost model may handle routine steps very well, while premium models can be reserved for tougher decisions, and sometimes a small open weight model can do the whole job. So orchestrating the right mix of models for the job is where we're at. Today, at least, things are always changing. And like every major technology wave before it, competition is already driving costs lower and broadening access. And all of us need to keep the conversation going and just stay informed. So good seeing you today, Katie. And just a reminder, where can everyone find you?
Katie Wilson: Yeah, you can find me on LinkedIn at Katie Wilson or Serious Moonlight Consulting on Instagram or subscribe to my newsletter. Sales isn't a dirty word on Beehive. How about you, Jackie?
Jackie: Yes, so please find me on LinkedIn at Jackie Gutierrez and on Instagram and TikTok at Business Growth Gal. And please follow Bop Boundaries at Bop Boundaries on Instagram, TikTok, YouTube, and LinkedIn. Thanks for listening, everyone. Make sure to subscribe wherever you get your podcasts. We're available on all platforms and also YouTube. Bye, Katie.
Katie Wilson: Hey, good to be here. This is a great topic, especially for startup founders building with AI and business leaders, you know, moving through this transformation. I think we're gonna get a lot of insights from it.
Jackie: Definitely. All right, so let's dive in. There's an interesting shift happening in AI right now. US companies have been experimenting with Chinese AI models for a few years now, but what's changing is how seriously they're being considered for everyday business use. Models from companies like DeepSeek. Alibaba's Quen, Moonshot's Kimmy, and ZAI's GLM are moving beyond experimentation and into real workflows from coding to customer-facing applications, too. And I don't think the story is as simple as saying American companies suddenly prefer Chinese AI. What's really happening is that companies have more choices now, and they're becoming much more focused on the economics of those choices, which is really important if you're building any business. if a model can handle a routine task well enough for a fraction of the cost, companies have a reason to consider it, even if they continue using AI models for more complex or sensitive work. So, you know, it's really about thinking deeper about what model fits what task is in addition to cost, right? Because everyone's starting to crack down finally on the cost of all of this. and in fact, OpenRooter reported that Chinese models surpassed US models in overall token share in early June of this year, 2026. And that kind of shows us just how quickly usage is starting to shift from model to model. For the last few years, the conversation was largely about which a which US lab had the smartest model, and whether it was OpenAI or Anthropic or Google. and others, they were really competing to build increasingly powerful systems and companies were willing to pay for that performance. now, you know, we're just starting to see the focus really shift. And and the new question really is more like does every task actually need the most expensive or powerful model available? I'm not so sure.
Katie Wilson: Yeah, it's a good point. And I think because AI is now an operating expense, right? Especially as we get to Q4 and end of the year, people are going to start to audit that. And and for startups and businesses of all sizes, that matters. If you're building an AI product, the cost isn't just the subscription your employees pay to say use ChatGPT or Claude at an enterprise level. You're potentially running thousands or millions of model callbacks behind the scenes. Every customer interaction, search, agent action, or coding task. Can generate more tokens. And we've talked about this in other episodes as well. And that's where some Chinese models have gotten attention, because depending on the model and the workload, Chinese AI systems can dramatically be dramatically cheaper than the leading US alternatives. Some comparisons put the savings at 60 to 90% per token for certain models and pricing tiers. You can't apply that number to every model, but the larger point is real: the price gap. Can be big enough to change how a startup designs its product. And that brings us back to token maxing, which we've talked about and heard about before, which is somewhat of a social media meme term, but it points out to something very important that companies are thinking about how many tokens they actually need to accomplish a certain task and when AI can become more expensive than the human output if that task becomes too expensive, right?
Jackie: Mm-hmm.
Katie Wilson: So If you're building an AI agent that makes 10 or 20 model callbacks to complete a workflow, you may not want every one of those calls going through your most expensive frontier model, like a Claude or a ChatGPT or Google. Maybe the first few steps can use a lower cost model. Maybe a premium model only gets called when the system reaches a difficult decision. Or maybe a small open weight model can handle the entire task. So companies are increasingly asking, what's the smartest AI that I can afford? For this specific job. so that's a big change and I think a shift in what we were seeing before. it can be a really important shift for startups, right? you think about AI almost like their team of employees. You wouldn't hire your most expensive specialist, your most senior person to do every job in the company. You'd match that person to the task. I think that's a really good analogy to think through it. and AI infrastructure can work the same way. You might use a premium US model for a high-stakes customer interaction or a super complex coding initiative, or a cheaper model for internal summaries, an open weight model for certain coding tasks that are very basic, and another model for longer documents. So that means startups don't necessarily have to choose one AI lab. They can build a stack where different models do different jobs. And that flexibility could be a competitive edge for that.
Jackie: Yeah, yeah, you really explained that really well. And, that also makes me think about how founders and leaders need to take it the other side of it seriously. If we're
Katie Wilson: Yes.
Jackie: gonna be, know, testing and implementing a variety of models then or LLMs, then you know, security is really what we need to focus on. That's that
Katie Wilson: Absolutely.
Jackie: needs to be part of the conversation. So if you're sending company information Through an AI provider, you're not just choosing their algorithm, you're choosing an infrastructure partner. So you need to know where the data is processed, whether prompts and outputs are retained, whether they're using it for training, who can access them, and what contractual protections are available. And when the model comes from a foreign company, those questions are pretty important and sometimes can be complicated too. A startup may be saving money on its AI bill while creating a completely different kind of exposure. Around customer data, intellectual property, or regulatory compliance. So, you know, we need to do a lot of research when we implement these models. and then there's an important distinction here because saying that this is a Chinese AI model doesn't necessarily tell you where the model is actually running.
Katie Wilson: Mm.
Jackie: Some Chinese developed models are open weight, which means developers can potentially run them through different infrastructure rather than sending every request directly to the model. To the model developer's own servers. A startup could also use a third-party provider with its own data policies and controls, which could be a good option. So I guess the message to founders isn't never use a Chinese model or fear a Chinese model because of security. say understand the entire journey, you know, on the back end, ask questions like who created the model, who hosts it, where does your data go? What gets logged, you know, what then happens to that data. And also what happens if geopolitical or regulatory environment changes, which
Katie Wilson: Yes.
Jackie: it seems to be changing every month. Hard to keep up with. and open router is interesting here because it gives us a window into what developers are actually using rather than just what companies say they're interested in. they see a lot of they see a lot of this since many people are using OpenRouter. And so they have real data to give us, you know, accurate information. Open Rooter actually reported that Deep Seek's share of token usage on its platform went from about 9% in January to 18%. By June 2026. That's a big jump, essentially
Katie Wilson: Yeah.
Jackie: doubling. And OpenRouter said Chinese models as a group had surpassed US models in token share by early June. So that's pretty meaningful, and it's a signal that tells us something obvious. it's measuring actual model traffic on a large routing platform, not just benchmark scores or headlines that you know these LLMs choose to share with us. And OpenRooter says the growth wasn't just from people chatting with the models. Agentic workloads were a major driver, which is important because agents can generate a lot of model calls. When you're running an AI system through multiple steps, the economics of each call start to matter a lot more.
Katie Wilson: Yeah, exactly. And an open reader's broader analysis of open weight models makes another important point. Cost is becoming a bigger part of the model selection conversation, like we were saying before. It says open weight models are keeping relatively close to US frontier models and capability while offering opportunities to move some workloads away from the more expensive system. That just doesn't mean the Chinese models have suddenly beaten every American model. They haven't. It just means that the gap between capability and price is becoming harder to ignore. In some ways, like they're good enough to do certain tasks at a much cheaper rate. And once the developers realize they can get a strong result for a fraction of the cost, switching becomes much easier. That's especially true for startups that don't have unlimited venture funding or massive infrastructure budgets. In case you're not familiar with OpenRooter, OpenRooter offers access to more than five hundred large language models from providers like OpenAI, Anthropic, Google, and Meta through a single API, including a wide variety of Chinese models as well.
Jackie: Yeah, we use OpenRooter. I think most software companies do. so who are these popular Chinese AI labs that we're talking about? we mentioned a few, but the first one, again, that most people know is Deep Seek, which has become especially associated with, good reasoning, coding, and price performance too. Then there's Alibaba's Quen, which has built a broad model family and a large developer ecosystem. Moonshot AI's Kimmy has become known for long context work, coding, and agentic tasks. And Zipu's GLM, now operating as Z.ai, has become another major player in reasoning, coding, and tool use. Then you also have Minimax, which is working across multimodel and agentic applications along with models from Tencent, Xeomi. And other Chinese companies. and then there's a ton of American companies too, that you can access through OpenRouter and directly too. So it's important not to talk about Chinese AI, quote unquote, as if it were one product competing against, you know, the major US labs like OpenAI, Anthropic, and Google. It's really a much broader ecosystem, you know, even globally, and each lab has different strengths. and and even in the US, right? We even there's so many more than the three we keep mentioning right here in this episode that have different strengths. and like we mentioned already, some Chinese models are open weight, meaning they can be deployed through different providers rather than directly through that specific lab. And for a startup, that actually makes the decision more interesting and you know requires some thoughtfulness. if you're doing a lot of Coding, Deep Seek or GLM might be worth testing if you're dealing with large amounts of information or long documents. Kimmy could be interesting for you. Quen seems to give developers a broad family of models and strong multilingual and tooling options. Minimax is said to be another, you know, another good model for agentic work. but I wouldn't turn these into hard rules at all. I mean, we test different models with my software company all the time. And I get recommendations from my developers, put this one with that. And I totally disagree with them a lot. then sometimes we agree. So you you know, you really have to test them all because, models change so quickly that today's winner can be yester yesterday's news in a few months. And honestly, like some of today's winners, I don't even want to switch to. You you just gotta test everything. So I think the better takeaway is that startups should evaluate models by use case rather than by reputation for sure.
Katie Wilson: that's great advice. so like we were saying before, there's also a major security and governance question around these labs. So recent research has found that AI agents from several Chinese and US model developers can display problematic behavior in controlled tests, including misleading users or attempting to work around restrictions. These tests don't mean that models are inherently dangerous, and we've talked about this in the in the AI doomsday one, but it does. Reinforce something important, a strong benchmark score isn't the same thing as a complete safety evaluation, right? For a startup, that means testing more than whether an AI gives you a good answer. You know, you also want to know how it behaves when it gets confused, when it encounters sensitive information, when it gives conflicting instructions, or when it's allowed to take action on its own. These are all great tests to run, like you were saying, to to prove out what is gonna work for you. more autonomous your AI becomes, the more those questions really matter. And that brings us to the US government's role because Washington is paying close attention to the competition, as we recently saw with the President Trump summit with all the AI leaders in Washington. This is something he's very focused on. And the administration's been building policy around AI security, national security, and protections of American AI technology. In April, the White House said that they had information indicating that foreign entities, principally in China, We're engaged in large-scale efforts to distill US frontier AI systems. And in June, another directive called for stronger security around advanced AI and framework for evaluating certain frontier models. So, one important thing to keep in mind is that it doesn't mean that the US has banned Chinese AI models. The focus today is largely on national security and what they have access to, cybersecurity, and really how AI is used in those sensitive government settings. not on prohibiting everyday use of specific models. That means we'll likely to continue to see scrutiny, scrutiny around certain AI systems, but it's a mistake to assume every Chinese developed model is prohibited or will become prohibited. So when we dive deeper into what AI leaders in the US government are actually saying, we we talked about this in our AI Doomsday debating episode, but we explored the realities and the fears driving the conversation and also maybe the political motivation. the companies, the US companies and CEOs motivation to want to put this out there. So give that a listen and it's really about testing it, like Jackie said, for yourself and defining for your business, what are you comfortable with and what security measures do you want in place.
Jackie: Yeah, good point. And while these Chinese models are gaining market share, likely because they're willing to perform many common tasks at a fraction of the cost of some of the US alternatives, we're all seeing major US AI labs become increasingly vocal about security risks, model misuse, and even AI doomsday scenarios, which we keep re referring back to that episode. It was our recent episode that Katie mentioned. Go back and give it a listen. I mean, on one hand, those concerns maybe completely legitimate. I mean, they are legitimate, right? But on the other hand, it's also worth asking whether there could also be a strategic angle to all these doomsday headlines. you know, if US labs believe they can't win the competition on cost alone, which it kind of seems to be looking like that, could they be positioning security as another way to differentiate their models and influence government policy and ultimately secure more government contracts? maybe they've realized that that's where they can win since basic AI models are showing signs of kind of a ceiling and an incremental slowdown in capability. And many of us are just wondering what the long game truthfully is with OpenAI, Anthropic, and others. none of us know. and I think we're gonna have to keep thinking about that.
Katie Wilson: Yeah, interesting for sure. And I would say also like a self-imposed slowdown, as we've heard in the in the in the last few weeks in the summit with Trump. A
Jackie: So interesting.
Katie Wilson: lot of the CEOs were saying, we don't trust this newest model. So we're sl intentionally slowing down. And again, we don't know if that is real or if that is or behind the scenes, they're doing something else, or if they that is also kind of a reflection of this. We want to have better security and protocols. So it's an interesting, Topic to really follow because, like you said, it's changing every day. So
Jackie: Mm-hmm.
Katie Wilson: if you're a startup founder listening to this, and I know many of you are, and thinking, okay, but what am I supposed to do? What models am I supposed to use? we can make the process pretty simple. Let's break down, you know, we're not going to tell you what model to use, like Jackie said, but we can break down some steps that might help you make that decision better. So, step one, define the job. Don't start by choosing a lab. Start with what you need the AI to accomplish. Is it coding, customer support research, document analysis, an internal agent, recommendation engine? So different tasks have very different model requirements. So once you know that, you can start searching and finding the best partner for that specific task. the next step, step two, classify the data. Is the AI working with public information, internal company information? Customer information, financial data, source code, intellectual property. The more sensitive the data, the more control you're gonna want. And you're gonna want a model that protects against that. So that's another really important step. step three, create a short list. Okay, based on these two criteria that we've done, defining the job, classifying the data, let's pick a model that makes sense for the task. That could include a US model from OpenAI, Anthropic, or Google, alongside Deep Seek, Quen, Kimmy, GLM. Or an open weight option. So have a few options that you're gonna test. Now, step four, actually test them on your actual work. Don't rely
Jackie: Mm-hmm.
Katie Wilson: on public benchmarks, like we said before. Give each model the same set of real examples from your business and compare the results. I'd look at accuracy, speed, consistency, and how much human editing was actually needed. If one model costs half as much but requires twice as much human review, you haven't actually saved money, right? Because you're paying someone to do that review.
Jackie: I need to do step four more often.
Katie Wilson: Yeah, I love that. Yeah, I think that's a really good one. And step five, it kind of comes off of step four, like we were saying, with that human time, calculate your real cost. So look beyond the advertised token price, but see look at how many calls does your workflow make? Does the model need retries? Are you using an agent that calls the model over and over? And are you paying for a premium model when a cheaper one could handle that workflow? And I'd add to that too, like what's the human review and what are you paying a human? if it took them two hours to do that, right? So you're really gonna
Jackie: Good point.
Katie Wilson: now line up the work with the cost in step five.
Jackie: Mm-hmm.
Katie Wilson: Then step six, investigate the provider. Look at the data retention, training policies, hosting options, security certificates, contractual protections, and where the data is processed. Now you might move this step up to step when you look at step two and you know that if
Jackie: I know, I think so.
Katie Wilson: you have like sensitive information, I think this step would move up. But We we keep it at the end because you're gonna look at cost usually first, but I think this is equally as important.
Jackie: Mm-hmm.
Katie Wilson: step seven, it goes aligned with that. How much control do you actually need? If you're building a low-risk internal tool, using an inexpensive API may be perfectly reasonable. If you're processing sess sensitive customer information or proprietary code, you might want a provider with stronger enterprise controls or consider running an open weight model in an environment you control. And lastly, step eight, don't lock yourself into one model, right? This actually might be the most important step, is that sometimes you start again with this process. know, if your entire product depends on one AI provider, your inherent pricing, outages, policy changes, and product roadmap are all beholden to that one provider. If your architecture allows you to switch models, be a little more nimble, you have more leverage. So you can always test new models. You can negotiate costs and enterprise costs. You can move workloads when the economics change or when the security policies change, right? So this is important. I think really like my takeaway is use this criteria or a criteria for yourself and your business, but always be looking to see what's coming next and where you can be nimble and try new things to save money.
Jackie: And open router really helps with that last step on being able to
Katie Wilson: Yeah.
Jackie: switch models really easily. So consider implementing that if you haven't already. So love those detailed tips. I'm going to be going through those again myself.
Katie Wilson: Yeah.
Jackie: Really helpful. And I think that just leads us to the bigger story today a startup doesn't necessarily need access to the most extent expensive model for every workflow. It can combine different models, use open weight systems where they make sense and reserve premium models for the task. Where they provide a meaningful advantage. And I think all the US companies are realizing this, which is why there's so much chatter right now and the headlines are just shifting back and forth.
Katie Wilson: yeah.
Jackie: and there's something pretty interesting happening, you because of all of this, the competition isn't just between Open AI and Anthropic, or between the US and China anymore. It's becoming a competition between different AI architecture, different business models, even and different ideas about how companies should build with AI. Open weight models are putting more control in developers' hands and entrepreneurs and startup founders' hands for sure. Low-cost models are putting pressure on pricing. Model routers are making it easier to switch between providers. And AI agents are creating entirely new questions about costs because one user request can trigger dozens of model interactions, which drives up that cost. So the advantage in what we're seeing in AI right now, especially in the startup scene, may go to the companies that are best at orchestrating AI, not simply the companies that are using the most powerful model for sure.
Katie Wilson: and that's probably the takeaway I'd leave founders with. Don't choose an AI model because everyone else is using it. Don't choose one because it's cheap or the cheapest. Don't reject one simply because it was developed in another country, right? Instead, ask those questions like these questions, like, does it perform well enough? Do the economics make sense for your business? Can I trust the way my data is handled? And do I have enough control in the model or provider changes? That'll help you start to evaluate this for yourself because AI is becoming part of the infrastructure of most businesses. And once it's embedded in your product, switching isn't always easy.
Jackie: Good point. So to wrap up this topic, we're seeing that Chinese AI models are gaining traction with US developers because they seem to offer a balance between cost capability and flexibility. and OpenRooters data is really showing us this. it's pretty obvious. But for startups, the lesson isn't that Chinese AI is better or that US has lost its lead at all. it's that model selection is becoming a business strategy, the smartest approach. Could be to build enough flexibility to change models as the AI market keeps moving. And as we all keep an eye on security concerns and government involvement, small language models are being talked about a lot and open source platforms are really expanding access, really making it just accessible, I guess. and it's also and and while all this is happening, why we're all testing how we source our You know, AI support. It's raising new questions about safety governance and how the public should always have a voice and decisions that shape its future. So it's important we keep talking about this. so let's move on to red flag and green flag. First up, is social status being earned by tech bros or is it being cover photoed into existence? Playwright and actor Jeremy O'Harris recently sat down with VF Global editor director Mark. Guidice, and argued that when publications like Vanity Fair put tech leaders on their covers and in major editorial features, they're giving them something they can't easily buy: cultural relevance and acceptance. The journalism is one thing, but the imagery and status that comes with the well-lit and art-directed photos can signal that these figures aren't just successful. They're part of the culture. his argument raises an interesting question. Are magazines simply covering the people shaping society? Or are they helping manufacture social acceptance and cultural credibility for leaders who may have wealth and power, but haven't necessarily earned that kind of acceptance? So, red flag or green flag on Jeremy's take, which is essentially: are these magazines reflecting who's really culturally relevant, or are they helping to decide who gets to be relevant? What do you think?
Katie Wilson: First off, Jackie, are you stalking my Spotify? Because I literally listened to this podcast yesterday in the car. I love Jeremy
Jackie: Cool.
Katie Wilson: O'Harris. I'm a huge theater nerd and a fan of his work for a long time. So I definitely think this whole podcast is worth listening to because he talks about this topic, but also a lot about AI and the relevance to culture. but yeah, I agree with him. I think this unearned glamorization of these mo moguls, and he talks about a little bit back, like back in the day, you had to earn relevance by philanthropy. being out there in the social scene, you know, things like, you know, yes, they get invited to the Met Gala and these other things. But a lot of times, really wealthy folks, you know, we think about the Gilded Age and other times in American history, they were also putting their names on things, right? Like they were donating to museums, they were donating back to the culture. And his point was really that, you know, by glamorizing them and putting them on or inviting them to these events, it's unearned, right? Let them kind of donate and show that they care about culture before we spotlight. so I really did like that point. I thought that was interesting. And I also thought it was interesting. Vanity Fair released a recent Sam Altman cover where his back is to you don't see his face. It's like in profile and it's his back of his head and then the the shoreline.
Jackie: The Bay Bridge, I think.
Katie Wilson: Yeah. And and I saw Jeremy O'Reilly commented on it because he's like, okay, like not showing his face, right? So that was interesting. he also talks about the the Trump pictures that Vanity Fair put out there that were like really overexposed and very close up and it kind of made them look ugly and he and he was saying 'cause it the and he's approved of those, yes.
Jackie: He approved of those. Yeah, such a good interview, that one. I gotta listen to the podcast. I only saw clips, but
Katie Wilson: yeah, that's right.
Jackie: so green flag on his take for sure, Jeremy's take, cause at a certain point, just constantly platforming the same tech leaders, which by the way, we are no longer platforming any women. They're all men. Yeah,
Katie Wilson: Yes.
Jackie: it just It stops being journalism and starts to be obvious paid brand marketing. Even if you show them in the not so positive light, there's still beautiful photography goes along with it. And I think they eat it up and they don't care. So,
Katie Wilson: Mm-hmm.
Jackie: you know, I mean there's just plenty of smart people shaping the future of technology and we never see them on a red carpet. So I'm with Jeremy.
Katie Wilson: Absolutely. Yeah, there's a lot yeah, there's a lot of great thought leaders out there. Let's highlight like a diversity of thought and voices. and again, you're making the nerds prom king and that's like what they want. And
Jackie: Dive.
Katie Wilson: so I think like, you know, have them earn it a little bit by showing and giving back to some parts of the culture. And then you can highlight great work in philanthropy versus just because they have billions of dollars, we're gonna invite them to this event.
Jackie: Yeah, and we're talking cringy nerds. There's a lot of cool nerds out there. I like nerds. We love nerds, but we're talking.
Katie Wilson: We love nerds. Yeah. I'm like I said, I'm a theater nerd. I'm an everyone's a nerd about something, you know. there there's lots of great nerds. All right. Up next, there's a meme going around, at least in the States, that says, why aren't the robots handling toxic chemicals or picking plastic out of the ocean? Why are they writing emails and making PowerPoint presentations? What the hell are we doing here? So red flag or green flag on using robotics to solve problems humans actually need help with. Like things that are dangerous or physically demanding, instead of focusing on replacing administrative tasks or like mindful tasks.
Jackie: So this meme I saw on TikTok and Instagram, it's really good. I th I mean it just hits home. So green flag on robots doing dangerous tasks, not housekeeping. I mean, if we're you know, I mean if we're that's all we see is like, look, we have another robot that's trying to mop your floor again and then falling over. So I mean, if we're investing so much and by the way our our tax dollars go and invest in robotics, if we're Building robots. I mean, I'd rather that they crawl through disaster zones and, know, jobs that where we can keep humans safe.
Katie Wilson: Yes.
Jackie: I just don't need them to be sitting in meetings where capturing nuance is really important. So
Katie Wilson: Yeah, I think there's like a distinguishing between like physical and intellectual activities too. Like I
Jackie: Yeah.
Katie Wilson: actually hadn't thought about it until I read this and saw these memes and like I was like, yeah, that doesn't make sense. Like I'm total green flag for it, but we you know, we need materials and robotics that are strong enough to survive in those situations, whether it be in toxic
Jackie: Okay, good.
Katie Wilson: chemicals or we already have, you know, drones and other things dismantling bombs and keeping law enforcement safe. So there's there's use cases for this. but I love the idea of like a robotic submarine going through the ocean and taking out trash. So I think the these are things that can be built in the future. And I think, you know, though like before, we need to highlight some of these great things versus like get a robot to fold your laundry or take your meeting notes. I'd love to see it, you know, save human lives and help the environment. Those are great, great use cases of AI and robotics.
Jackie: Yeah, environmental for sure, picking up plastic. That's where we need to focus. All
Katie Wilson: Mm.
Jackie: right. Okay, so let's talk about one more. In Arizona, a 10-year manslaughter sentence was tossed out because an AI video of the deceased victim influenced the judge's consideration in the sentencing phase. an Arizona Court of Appeals says that the man has to be resentenced since the statements from the victim weren't real, totally AI generated, and it violated due process. his due process rights. So red flag or green flag on AI generated videos in court. Yeah.
Katie Wilson: Yeah, this is pretty wild. Like if you see the actual video, it's jarring. Like it's the victim actually like speaking to camera, but the video and the voice and likeness are all a I generated. And I don't know who wrote it. I'm assuming his family did. but you know, I'm a red flag on this as much as I like, I'm really pro-victims' rights and wanting people to be able to represent themselves. However, right now a victim statement is left to the families. They're real people that are alive, they can talk about the impact. and really I I think let's leave it to them versus AI videos of the victims. Then we're putting words in people's mouth. and like you see from a legal standpoint, it can violate certain terms potentially. So yeah, I I'm not entirely comfortable with this being used in court.
Jackie: Yeah, good point. Red flag, condolences to the victim and the family, of course, but it might end up, you know, a loophole in some way using these AI generated videos for someone who's actually guilty to go free. So we definitely need to be careful with this in court. So red flag
Katie Wilson: Totally.
Jackie: on that. so as we wrap up this episode today, as AI usage expands, it's clear that not every task needs the most expensive model. A lower cost model may handle routine steps very well, while premium models can be reserved for tougher decisions, and sometimes a small open weight model can do the whole job. So orchestrating the right mix of models for the job is where we're at. Today, at least, things are always changing. And like every major technology wave before it, competition is already driving costs lower and broadening access. And all of us need to keep the conversation going and just stay informed. So good seeing you today, Katie. And just a reminder, where can everyone find you?
Katie Wilson: Yeah, you can find me on LinkedIn at Katie Wilson or Serious Moonlight Consulting on Instagram or subscribe to my newsletter. Sales isn't a dirty word on Beehive. How about you, Jackie?
Jackie: Yes, so please find me on LinkedIn at Jackie Gutierrez and on Instagram and TikTok at Business Growth Gal. And please follow Bop Boundaries at Bop Boundaries on Instagram, TikTok, YouTube, and LinkedIn. Thanks for listening, everyone. Make sure to subscribe wherever you get your podcasts. We're available on all platforms and also YouTube. Bye, Katie.
