Kimi K3's Edge in the Anthropic Battle  🔗 **Source:** [Kimi K3's Edge in the Anthropic Battle](https://share.snipd.com/episode/47de7d2c-2ccb-41e4-8691-a50a579496e5) ## Highlights - **Using AI Box MCP Inside Claude To Generate Ads** - Jaeden demoed AI Box's MCP inside Claude to generate images, audio, and video simultaneously for ad variations. - He created 20 ad variations in seconds by having Claude call AI Box to produce images and pull them into templates. Transcript: Jaeden Schafer Podcast today. First, I wanted to tell you about the most exciting feature I have added to AI Box, which is an MCP, basically a connection that allows Claude or any other AI tool to get all the other AI Tools inside of it. Particularly exciting for me is that I've added image generation inside of Claude. You can also do audio and video, but image is what I'm really excited about. I've spent the entire day making a Facebook ad campaign and every single ad I created with this tool was using the AI box MCP. Essentially what you do is when you're using Claude, all you have to do is say, generate an image of XYZ. You can put the regular prompt you'd put into chat GPT or something else and just say, use AI box, or it really just understands that that's a tool that can generate images. So you could just say, generate an image of XYZ. Claude will go use your AI box account, generate the image, pull it in and open it right there in your Claude desktop. So you can see the image inside of Claude. It feels magical. I'm not going to lie. And the thing that I love about it is that it can do multiple images simultaneously. So I'm like, I work with Claude to come for like a template of an image that I like for the ad, but it needs, you know, images and little graphics inside of the ad. So I work on that and then I'm like, perfect. This looks great. And I'll make 20 different variations of this, you know, change the title and headline and change these graphics, go use AI box for all of that. So it uses AI box to generate all of the little images and graphics and pulls them into its template it made. And it spits out 20 of them. And it does it in like a few seconds because it can do simultaneous image generation on AI box. That is what I'm super excited about. If you want to try it out, it is AI box dot AI slash MCP. ([Time 0:01:00](https://share.snipd.com/snip/7ed8ccca-1d0f-4bbd-8014-6e2fbabcd10a)) - **Apple Alleges Large Scale Recruiting Of Design Talent** - Apple sued OpenAI claiming OpenAI hired 400+ former Apple designers and misused trade secrets via its chief hardware officer. - Jaeden Schafer notes Apple named OpenAI's chief hardware officer and alleges employees brought secrets and even retained laptop access. Transcript: Jaeden Schafer Apple and open AI. I think a lot of people are talking about this right now, particularly because the, the, the story broke that Apple has passed in video to claim the world's most valuable company title. So now we have the biggest company in the world coming against opening eye, which is looking to IPO soon. Apple hit a $4.88 trillion market value on Friday. And the reason for this isn't just Apple growing, but also Nvidia falling. Nvidia went down 3%, which basically ended their run as the top in the entire market, which they've been sitting there for quite some time. Apple's valuation, while it hit that, it surpassed NVIDIA's valuation of 4.8 trillion. NVIDIA shares fell, like I said, Apple's up 22% this year and NVIDIA has gained about 7%. It's interesting because we're seeing NVIDIA that had such a good run for so long, and we're shifting out of that being the hottest segment. And a lot of Wall Street has rotated towards memory and infrastructure. Companies like Micron Technologies and SanDisk are becoming really popular. And so it feels like maybe NVIDIA isn't just the hottest, sexiest company. Now, while this is happening, Apple is doing incredibly well in the stock market and their lawsuit of OpenAI is pushing forward. They are alleging that OpenAI hired over 400 former Apple employees and they misused trade secrets through their chief hardware officer. And of course, with the IPO that's coming up, this is something that OpenAI obviously doesn't want to have, you know,ishing their name Apple named opening eyes chief hardware officer In this whole complaint and they and this is all happening at the same time the opening I has released their first piece of hardware it is a $230 keyboard for codex or coding tool which I have purchased by the way so I'll give you guys a review when that comes in the mail I'm actually really excited about that was it a spur of the moment purchase yeah I actually saw the news Drop that this thing was available, and I was covering it on the AI Hustle podcast, and like mid-podcast, Jamie starts talking, and I'm like, okay, I got to buy this keyboard, because It said while supplies lasted, I knew the supplies were not going to last, so I went and purchased it. Side note, OpenAI also released a basketball that they're selling, and it just looks like a weird gradient color basketball with the OpenAI logo on it. That also said supplies last. I saw this, and that came out, I guess, a few hours ago. I went and looked, and that was already sold out. So the basketball and the keyboard are sold out, but it looks like OpenAI is making these physical products, which is kind of funny. Will this save them from their IPO woes? I don't think so. But it's going to be interesting because that was their first piece of hardware that they're releasing and they're working with the former Apple iPhone designer, Johnny Ive, to come Up with a new undisclosed piece of hardware. Some people say it's basically like a smart speaker on your computer. In any case, I've looked over a lot of the accusations Apple's made. It looks pretty bad. Basically, they brought over a new chief of hardware and he came from Apple and they gutted the entire design team of Apple. They hired like 400 people from the design team of Apple. He used to be over there at Apple, so he knew everyone. He pulled them in. And when he pulled them in, Apple's allegation is that he told all of them as they were coming to OpenAI, this is how you get around all of the basically when you quit Apple how you get around All of the you know retaining all the information so bring little components give us the secret words tell us what's going on at Apple and in some case they said that one of the employees Kept like a laptop and use some like VPN loophole to keep access to some data at Apple that's kind of Apple's lawsuit in any case. I mean, it's an important lesson though. When your employees leave, everything in their head also leaves with them. So what type of confidential information do they move over? That's what the lawsuit is going to be ([Time 0:02:32](https://share.snipd.com/snip/6c9470f9-6b7f-4976-a396-ea2165e6686c)) - **Kimi K3 Matches And Beats Anthropic On Benchmarks** - Moonshot's Kimi K3 is an open-weight model at 2–3 trillion parameters that outperforms Anthropic on several benchmarks. - Jaeden highlights its front-end developer benchmark lead and Moonshot's valuation jump to $31.5 billion. Transcript: Jaeden Schafer Moonshots Kimmy 3 is looking to beat Anthropic right now. And on one important benchmark, which is a front-end developer benchmark, it is ahead of Fable 5 even. So it's really doing well. It's a 2 to 3 trillion parameter openweight model. And with all of this, it's coming from a Chinese lab. And the Chinese lab's valuation, Moonshot, jumped to $31.5 billion in two months as they're coming out with this model. And, you know, it's beating even Anthropoc on a bunch of different benchmarks. This is the largest open weight model that has ever been released from China. They raised $2 billion at a $20 billion valuation back in May, and now they're valued at 31 billion. So their valuation's increasing, their model's getting really good. This is all, you know, coming to a time with Anthropoc, they're getting kind of held back by the government. It feels like we're getting a lot of regulation, these 30-day kind of holdback periods where the government can go and review the models before they release them to the public. None of that is happening in China, and they're releasing all these models, you know, open weight so people can actually get them and fine-tune their own models on them. A lot of enterprise buyers prefer this because if you can grab something like Kimi K3, put it on your own hardware. It's not tied to China. It's not tied to moonshot. They don't see into anything that you're doing. You fine tune your own stuff on this system. Um, and then you can basically access it for quote unquote free, but you're, you're using it with all on your own hardware. ([Time 0:06:16](https://share.snipd.com/snip/ad966c6f-76eb-46ff-aff3-89a5c8337b36)) - **Open Models From China Accelerate Enterprise On-Prem Adoption** - Chinese labs can release large open-weight models without regulatory holdbacks, accelerating enterprise adoption that wants on-prem fine-tuning. - Jaeden explains enterprises value running Kimi K3 on their own hardware to avoid vendor inspection. Transcript: Jaeden Schafer Really good. This is all, you know, coming to a time with Anthropoc, they're getting kind of held back by the government. It feels like we're getting a lot of regulation, these 30-day kind of holdback periods where the government can go and review the models before they release them to the public. None of that is happening in China, and they're releasing all these models, you know, open weight so people can actually get them and fine-tune their own models on them. A lot of enterprise buyers prefer this because if you can grab something like Kimi K3, put it on your own hardware. It's not tied to China. It's not tied to moonshot. They don't see into anything that you're doing. You fine tune your own stuff on this system. ([Time 0:06:58](https://share.snipd.com/snip/b86128dc-c7bf-484c-9473-12f860a2e8fb)) - **Use AI To Rapidly Audit Regulations** - Use AI to audit and streamline large rule sets to save massive staff time. - Jaeden cites Governor Kathy Hochul using AI to review New York's laws, turning a five-year task into months and finding outdated rules. Transcript: Jaeden Schafer Said that this was a huge productivity jump you know she said it's 30x productivity jump on what she was able to do. So she gets some points for looking like the cool governor for using AI to streamline. And honestly, I hope every state does that. So good on her for that. Bad on her for banning data centers for a year. Although it doesn't really matter because every other state will get them instead, ([Time 0:10:00](https://share.snipd.com/snip/5d9fa336-5766-4394-bd00-5fadf9da89eb)) - **Notebook LM Becomes Gemini Notebook With Code Execution** - Google rebranded Notebook LM to Gemini Notebook and added secure code execution tied to user documents. - Jaeden reports 30 million users and syncing with the main Gemini app to run code for data analysis against Google Drive. Transcript: Jaeden Schafer State will get them instead, I guess. Okay, Google has renamed Notebook LM to Gemini notebook. I mean, it makes sense. They're trying to keep with the branding. Google loves changing the name and changing the products and killing the products of everything. I'm just happy that notebook LM is still alive. They've added code execution to it. It's funny because they just basically had a runaway product that was random. They released tons of products. Notebook LM just caught fire for some reason. I think a lot of students just would throw their textbooks into it and have it do a podcast talking back to them. They liked that. But in any case, it can now do code execution. It has a secured cloud computer that can write code against your own documents for data analysis. The app has 30 million users. 600,000 organizations are using Notebook LM. Honestly, it blows my mind. I mean, I've played with it. I've used it. I know people that religiously use it. It's not part of my go-to, but it blows my mind that they have 30 million users and now it's called Gemini Notebook. So they're just giving Gemini a better name and a little bit of a good PR boost with this popular tool. The code execution is live today, so you can go use that for Google AI ultra subscribers and for workspace customers. It's rolling out to all pro users over the coming weeks. Notebooks will now sync with the main Gemini app, which honestly I'm all for that. I hate it when there's all of these like different tools and they're not connected, especially when it's from the same company. Like Google really should ([Time 0:10:20](https://share.snipd.com/snip/e1c4ccc0-2a21-4933-b677-6c95404a3584)) - **Visual Reasoning Still Trails Text Models** - Visual reasoning remains a frontier where models lag human toddlers, motivating startups like Xlorian focused on multimodal understanding. - Jaeden notes ex-DeepMind researcher Andrew Day raised $55M for a visual AI startup before shipping a product. Transcript: Jaeden Schafer He did all of this before the startup shipped anything. Dai is basically betting that visual reasoning, teaching AI to understand images and video the way the current systematic text handles is the next frontier. And he basically turned down a whole bunch of really high offers to take strategic investors like NVIDIA who understand how expensive AI research actually is. So NVIDIA is one of his investors, a bunch of other people who got turned down. Dai spent over a decade at DeepMind on foundational work that all later influenced ChatGPT. Alorian's valuation to capital ratio is way more aggressive than even thinking machines from a former co-founder of OpenAI, which was one of the largest seed rounds in US history. Google, OpenAI, and Anthropoc are all working on multimodal understanding, which is a recent Stanford benchmark. And that benchmark in particular showed that frontier vision models are still lagging and they're worse than a toddler on basic visual reasoning tasks. ([Time 0:12:20](https://share.snipd.com/snip/231234e9-6c00-487c-94ff-703d03f2d50a)) - **Databricks Valued At $188 Billion As AI Platform Powerhouse** - Databricks raised funding at a $188 billion valuation, positioning it alongside top AI platform companies. - Jaeden explains Databricks sells data and infrastructure for training and serving models and acquired Mosaic ML to strengthen offerings. Transcript: Jaeden Schafer Databricks has landed funding at a $188 billion valuation. They just closed this funding round. They made it basically in the same league as OpenAI and Anthropic on paper anyways. The data and AI platform company, they're not just building AI models. They're also selling a lot of the infrastructure that companies use to train, deploy, and run models in production. Databricks right now is competing with Snowflake, if you've heard of the company, on data warehousing. And they do, you know, they're also competing a bunch of other big players on modeling serving. They're also strengthening their acquisition. They acquired Mosaic ML, which I covered on the podcast back in 2023. I don't know why, it's kind of funny. That particular episode I covered from the San Francisco airport, which is funny to me. But anyways, the company has basically not done an IPO. So they're not publicly listed. They've just done these private funding rounds. It's kind of wild that we have these massive AI companies that are doing that right now. At this new valuation, they're now valued higher than most established enterprise software companies. ([Time 0:13:20](https://share.snipd.com/snip/f9e37c8e-8a16-4a54-b14f-5aebb4a25b7c))