In short
The episode discusses “Monetizing AI: The Kimi K3 Effect,” focusing on how open-weight/open-source models like Moonshot’s Kimi K3 can reduce AI operating costs for businesses versus paid services (e.g., OpenAI/Anthropic).
Key claims
Kimi K3 is “moonshot” hyped and benchmark-competitive (sometimes near/above Claude Opus 4.8), largely because it uses distillation (Anthropic calls it a “distillation attack”)—training a model to mimic Claude by querying it extensively. Guests/hosts: John Daly and Jamie (plus Jaden, referenced as the school-community member who shared ad results).
Guest backgrounds
no formal credentials given; they discuss AI monetization, marketing, and running AI tools locally.
Notable examples
local MacBook running Kimi K3 (slow but works; ~1.45TB experts downloaded); concerns about third-party integrations siphoning data; leaked chat indexing via “share” buttons; companies trying to limit model releases to prevent distillation and copying.
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Chapters
Tap a time to open that second in VOIntroduction to Kimi K3 and AI Models
1:00 to 2:45
Discussion on the Kimi K3 model and its implications for businesses using AI.
“Today on the podcast, the very hyped moonshot Kimmy K3 model is forcing OpenAI Google Anathropics to rethink closed AI.”
AI Hustle Community Insights
2:45 to 5:45
Overview of the AI Hustle community and strategies shared among members.
“Jaden, is that something to really be concerned with and make your decisions based on that?”
The Economics of Using Kimi K3
5:45 to 7:50
Examination of the cost benefits and performance of the Kimi K3 model.
“and they could see everything you were typing into it.”
Security Concerns with AI Models
7:50 to 11:44
Discussion on security risks related to running AI models and data privacy.
“So at the end of the day, you really can save a lot of money.”
Concluding Thoughts and Call to Action
11:44 to 12:45
Final reflections on Kimi K3 and invitation for audience engagement.
“Ten hours on my feet, traffic, noise, everybody needing something from me.”
Transcript
Automatic transcript. May contain errors.0:28John Daly here. With the American Express Platinum card, I can unlock experiences like no other. Since I'm always booking my next trip, I love that I can earn points on travel. Plus, I get a resi benefit, so you know I'm hitting the restaurants everyone's talking about. And you can find out your welcome offer after you apply, which could be as high as 175 ,000 points. For experiences like no other, there's nothing like Platinum. Learn more at AmericanExpress.com slash explore dash platinum. Terms apply. Today on the podcast, the very hyped moonshot Kimmy K3 model is forcing OpenAI Google Anathropics to rethink closed AI.
1:09There's so much drama in Silicon Valley from this, but also there is a ton of money that you can save in your own business by using some of these open source models instead of Claude and OpenAI to run really repetitive, heavy tasks. So we're going to talk about how to get some of the cost savings with these AI models, what they're useful for, what some of the alarm bells are, what people are concerned about, all the drama. There's so much to get into. Before we do, Jamie, would you want to tell them about the school community? Yeah, if you've ever wanted to learn how to actually make money online using AI or grow your business, you're going to want to check out our school community called AI Hustle.
1:46We have over 600 members and each week we release bonus content over there. we show you exact strategies and ways we're making money. This week, Jaden talked about his, well, actually we talked about our podcast and how we're making money there. But he talked about his exact Facebook ad breakdown. He spent over$2 ,000 on Facebook ads promoting his AI tool called AI Box. And he went through the results and how he optimized it and whether or not it was profitable and made sense for him to keep doing. So broke all that down. And so it's really just a great place to learn that kind of thing. We go over marketing, vibe coding, how to become an AI consultant, and then also how to make money with AI content and music.
2:33So if you're interested in any of that, we'd love to have you be a member. It's free to join. You can get our weekly live streams and also some bonus episodes. But if you want access to the full vault, it's$19 a month. uh well let's talk about moon let's talk about kimmy k3 um this is another pretty funny uh named software if you ask me an ai company um you know deep seek i remember it was like really mysterious sounding and this one even to me sounds more funny but um you know the big thing with a lot of these open source um ai you know things is is security people are concerned about the security of their data and also national security, all those kind of things.
3:17Jaden, is that something to really be concerned with and make your decisions based on that? Or, you know, we're going to talk about, yeah, what are the reasons you would want to use Kimmy K3 over Claude or OpenAI? I mean, the biggest reason just comes down to economics. And even when DeepSeek came out, it was the exact same thing. It's so much cheaper. So for so many tasks, you just get way more bang for your buck. Kimi K3 in particular is interesting because in a bunch of different benchmarks, it was actually better than Opus, I think, 4.8 on some tasks, worse than others. But it was like really close up there.
3:54And Moonshot actually used something that Anthropic calls distillation attack, where basically they go and ask Claude like a million questions and they get the responses back and they take the question responses from Claude and they train a model to just be like answer like Claude answers basically. And sometimes I've seen with these models where you ask them like, what is your name? And they're like, I'm Claude Opus 4.8. So like it just like straight up. Anyways, Anthropic doesn't like this because they're spending billions of dollars to train their models. And if you go run a distillation attack, you get very similar sounding results and responses.
4:28You do have to have underlying like data sets and like information and stuff, but you can kind of scoop up the whole internet and then you can fine tune or you can distill Anthropic for how to answer things. And then you have to be clever on how you do logic and reasoning. But there's a bunch of the pieces that Anthropic does not like being stolen. So they're trying to get it legislated. They're trying to make it illegal to do. But about 25 different tech companies, including Microsoft, I think even OpenAI has signed on saying that they want these open weight models to be available. A lot of them had said the distillation, they want to still be possible and not illegal.
5:03I think Elon Musk in a lawsuit with OpenAI admitted saying that Grok had trained using distillation off of OpenAI. And he's like, yeah, it's pretty industry standard. Everyone does it. And so anyways, Anthropix trying to make it illegal, which like, yeah, they spent a billion dollars to make a thing. They don't want anyone to rip it off. But also Anthropix ripped off the entire internet for free without compensating anybody. And now they're mad and people are ripping them off. So I don't really have a ton of sympathy. At the end of the day, why you'd want to use K3, I think it's really good. and way cheaper.
5:34You just save so much money running some of these open source models. But the danger in all of this, it's not in the same way if you were just using like DeepSeek and it was tied to like a Chinese server and they could see everything you were typing into it. I understand where a lot of the distrust could come from. But when it's an open source model and it's running on your own computer, that's less of a concern. However, I've recently been kind of looking into some of the issues and one of them is if you build, if someone builds a third party integration on top like it integrates with your gmail or integrates with another tool sometimes those third-party integrations built on top can siphon out data so the connectors can be yeah can be the issue that people would want to be worried about okay so yeah i'm just trying to think would you have to have a pretty um intense computer setup to run some of these models locally or this would be in like if you're trying to maybe start a business and you needed to use a lot of tokens but you didn't want to go through OpenAI, but do you have to have a pretty intense computer to run some of these models?
6:38Yeah, I think it's interesting. I was reading like one thread over on Reddit in particular where someone said that they got KimiK3 running on their MacBook. They said it was super slow, but that it did work. And so I think, yes, it is possible. um they also said that they pulled all 1.45 terabytes um of experts down to a local disk so experts are some of the things you have to download in order to run this so it's you know it's almost like two terabytes like it's pretty massive and if your macbook doesn't have more than two terabytes of storage uh you you'd be in trouble um but they they were able to get it down um from 16 seconds a token down to less than a minute and they did this on an m1 mac so not even the newest Mac, but they were able to run this on an M1 Mac.
7:30So anyways, there's a whole bunch of people that are in this thread over on Reddit where they're talking about how they optimize this model. But yeah, like you can run this on a MacBook. Is it the best? Is it the greatest? Is it good and fastest chat GPT? No. But if it's running on your computer, it's completely free other than your computer. And if you have a beefed up computer, like if you have like a Mac studio or something, you know, something more powerful, then this thing is going to get a lot better and a lot faster. So at the end of the day, you really can save a lot of money. And I think that pretty much terrifies Anthropic and open AI to imagine that, you know, you could be running a model that costs you basically nothing other than a good computer and they're not making money off of it.
8:09Yeah. And if you have a new updated Mac, it looks like those are at least three times faster than the M1 chip, the M5. So yeah, that's really interesting. I mean, I feel like there's a lot of things I use Claude for that I probably don't need to be, that don't need to be, you know, super expensive, intense, complex tasks. I think I'm just using credits. For example, if you had a repeat task and you wanted to have a, you know, a daily rundown of your, your emails you got in and tasks for the day, like that, I don't think is necessary. If you could find a way to run that locally that would be great because then you wouldn't have to pay for your 200 cloud subscription so but you know i do get the security thing and i think but how you know the question is how secure really are is open ai and anthropic because i mean we just saw a news article recently that um some of the people's chats were leaked onto just the internet and it showed up in it in a google search so i will say for those leaked chats in particular it's a problem that we've seen with grok and also open ai in the past which basically there's a share button so if you want to share it with a friend um then a thing happens where it can get indexed by google sometimes so that's kind of what happened you have to click the share button enable so you are sharing it with someone you just didn't want to share with the whole internet but then google indexes it so that's the that's kind of the issue with that one got it okay that makes more sense anyways uh interesting topic and we'll see you know the other thing is here too where the question is, well, for some of these companies like Anthropoc and OpenAI, do they want to automatically release their most powerful models to the public?
9:50Probably not because then Chinese government is just going to steal it and make their own models just as good. And the arms race will stay on equal playing field. So they'll just be able to copy the model that comes out back and forth, back and forth. So some of the discussion amongst these companies now is should we release our newest and greatest model or should we kind of keep it to ourselves, use it only in certain situations? Yeah, what are your thoughts on that, Jayden? Well, I think it's so annoying. It's like my least favorite thing because, yeah, basically they're all like, oh, no, if we release our best model, then it's just going to get distilled.
10:32So instead of releasing their best model to the public, they just released their best model to, I mean, they did it with like, you know, Mythos, you know, Claude Mythos or whatever, where they released the really good model to like Amazon and Microsoft because it was good at finding security vulnerabilities and like go patch all your security vulnerabilities. But it just kind of feels annoying that like the biggest, richest companies in the world get access to it and like smaller startups don't. So I kind of hate that side of it. But the other funny thing is that Fable 5, Claude Fable 5 is what was distilled to create Kimmy K3.
11:03And they did this very quickly. They distilled. So it's like in the little window where it was publicly available to everyone, they just got completely copied. And that's why Kimmy K3 was so good is because it was better than Opus 4.8 because they were distilling it off of Fable 5. Right, interesting. Yeah, that's crazy. We'll see. We'll keep you guys updated on any news related to Kimi K3. But for now, if you got any value out of this episode, we would really appreciate a rating, a review, wherever you're listening. Those help us out a lot, reach more people. And again, check out the AI Hustle School community if you want to learn how to take some of these tools and actually go and make money with them.
11:42Thanks for listening, and we'll see you next time. Ten hours on my feet, traffic, noise, everybody needing something from me. My head is spinning. I just need the world to stop for one second. But then I get home, shoes off, phone out, and I open Modo Casino. Hundreds of slots, classic table games. It's my instant reset. The variety I want and the quiet I need. Modo Casino is a social casino. We're prohibited. No purchase necessary. Visit Modo.us for more details. Download the Modo Casino app today. Must be 21 plus. Modo Casino. America's social casino. Saving money on your summer project at Minari.
12:18Transform your garage with rock-solid floor coatings. Complete your project with a showroom quality finish in one weekend. A one car kit covers up to 250 square feet. Get everything you need to complete your project with rock solid floor coatings. Our knowledgeable team is ready to help bring your dreams to life one gallon at a time. Plus check out our weekly flyer on Menards.com for all the great deals happening now. Save big money at Menards.
From the publisher
In this episode, we investigate how Kimi K3 is creating new revenue opportunities in the AI space. Find out strategies for leveraging this technology for profit.




