Today on the AI Daily Brief, how this unhinged AI ad shows the future of advertising. Before that in the headlines, OpenAI's open model gets delayed. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

All right, friends, quick announcements here before we get into the headlines. First of all, thank you to today's sponsors, KPMG, Blitzy, Vanta, and Agency.org. As always, if you are looking for an ad-free version of the show, you can find it at patreon.com slash ai daily brief. Ad-free plans start at just $3. Also, I will quickly mention, as I did last week, that we have moved into sponsorship sales for the fall. There are a few slots left, so if you are interested in learning more, reach out to me at nlw at breakdown.network.

For now, though, let's get into this OpenAI news and the rest of the headlines. Welcome back to the AI Daily Brief Headlines Edition, all the daily AI news you need in around five minutes. We kick off today with an unfortunate delay from OpenAI. The company has delayed their first open model in years. Sam Altman tweeted, We're going to take a little more time with our open weights model, i.e. expect it later this summer but not June.

Our research team did something unexpected and quite amazing, and we think it will be very worth the wait, but needs a bit longer. Now, this open model was announced back in April, and it's intended to be a reasoning model alongside the same lines as their O series. The announcement had come in the wake of DeepSeek shocking the world as an open source reasoner, and Altman publicly acknowledging that OpenAI had been on the wrong side of history when it came to open source.

Following the announcement, we got rumors that the research team was working on a way to hand off more complex query to OpenAI's closed models. There is certainly a lot of competition in this space right now. Mistral recently released their first family of reasoning models, expanding their range into that latest architecture. Alibaba's Quent team have also released a range of hybrid reasoning models that can modulate their inference according to the complexity of the prompt. Given Altman's tweet, it's likely that OpenAI is cooking up something interesting, but for now, we're all just going to have to wait.

Speaking of Mistral, that company has brilliantly navigated the changing geopolitical landscape to its own advantage.

Growing tensions with the U.S. have caused European companies to seek out a domestic AI solution, and Mistral is putting itself firmly in place as that answer. The company is on pace to hit $100 million in annualized revenue this year as a result. From the beginning of the AI boom, European officials have stressed the need for strategic autonomy from the U.S. And while the Restrictive AI Act has been more of a barrier than anything else, the intention has always been clear to ensure Europe isn't reliant on foreign technology.

Extending their mission to deliver European tech to the continent, Mistral has also announced a partnership with Nvidia to build a massive new data center. Located 20 miles south of Paris, the data center will initially house 18,000 of Nvidia's new Blackwell chips. The goal is to give European customers a way to tap into Mistral's AI products without dependence on the US. The plans are to expand the facility to 100 megawatts over the coming year and a half.

Speaking from the VivaTech event in Paris yesterday, Mistral CEO Arthur Mensch said that U.S. Vice President J.D. Vance's speech on USAI dominance in February had been a wake-up call. Mensch said,

It tremendously affected our demand because European leaders just don't want to be talked to that way. At the same event, NVIDIA CEO Jensen Huang said that more than 20 AI factories, his new term for these massive AI data centers, are being planned and built across Europe over the next two years. He said that several of them will be gigafactories that house over 100,000 chips, making them suitable for large model training. Huang said, "...we will increase the amount of AI computing capacity in Europe by a factor of 10."

Shikar writes, Another story that we've been watching closely is that of the talent wars in Silicon Valley.

Now, this has been top of mind ever since the news of Meta potentially sort of acquiring Scalebroke. As part of those reports, we discover that Mark Zuckerberg is also spending big to build a 50-person team around Scale CEO Alexander Wang, who will be leading a new superintelligence effort at Meta.

Bloomberg sources are now discussing what the new roster looks like. They report that Jack Ray, a principal researcher at Google DeepMind, is expected to come on board. And Johan Schalwex, a machine learning lead at AI voice startup Sesame AI, has also reportedly agreed to a deal. Zuckerberg has apparently been hosting candidates at his homes in Palo Alto and Lake Tahoe as part of a personal recruitment drive.

Bloomberg reiterates that Meta and Zuckerberg are offering compensation packages worth tens of millions of dollars over several years. And as much as this is nominally about future superintelligence, it's also about Meta shoring up its immediate term. Arguing for why these massive compensation packages actually make sense, Super Annotate co-founder Vehan Petrasan said, "...there are very few people globally who can do these types of large AI trainings very, very efficiently."

Now, Meta is no stranger to spending big, but this is definitely a whole different level. Menlo Ventures' Didi Das confirmed what he's been hearing on the ground in San Francisco, posting, "'It's true. The Meta offers for the superintelligence team are actually insane. If you work at the big AI labs, Zuck is personally negotiating $10 million plus per year in cold hard liquid money. I've never seen anything like it. If you see your buddy who works in AI suddenly join Meta, congratulate them on the new future Atherton Mansion.'"

Didi also said that while he hadn't heard of any nine-figure deals, which was something that had been reported earlier, he had heard nine figures over four years. When someone parroted the usual VC line that top talent can do better by building their own startup, Das replied, they probably can't actually.

Professor Ethan Malek pondered how this gigantic recruiting effort might fundamentally shape the industry, saying, Now in terms of competition, and whether there's anyone that can compete with the big tech lab, some interesting news in enterprise AI.

enterprise AI startup Glean has closed a new funding round that values the company at $7.2 billion. The Series F round, led by Wellington Management, was a small one, just $150 million in fresh capital, but a big jump in valuation from their $4.6 billion valuation from back in September.

Glean is in a really interesting spot. They are offering an all-in-one enterprise end-to-end sort of AI package that includes models, agents, enterprise search, and their pitch has had surprising traction inside enterprises. Not only have they surpassed 100 million in ARR, in our interactions at Superintelligent with companies, they've also had a lot of traction in ARR.

The one partner to do it all thing that they offer, especially with their exclusive focus on enterprise AI, as opposed to some of the other big labs, which obviously have lots of different commitments in lots of different areas, has been really resonant.

At the same time, those big players are all offering certain comparable features, and we're definitely starting to see more aggressive competitive tactics. Reports are that Salesforce has recently changed its terms of service for the Slack API. That effectively means that companies like Glean won't be able to index and store Slack data for use with their own search technology.

Given how much important context and information exists in Slack, that's a big challenge for a company like Glean. Now, a Salesforce spokesperson says this is just all about data security, stating, Glean themselves acknowledge that the change will harm their users, stating in an email to customers that it would, quote,

In some ways, clearly a bit of a nudge to try to get those customers to put pressure on Salesforce to potentially reverse course or think differently about the policy. Strange Fund's Tara Tan sums it up pretty admirably, writing, Meta buys scale, OpenAI acquires Windsurf and Anthropic snubs it. Salesforce owns Slack and blocks off Glean and other players. Everyone is floating $10 million paychecks to poach or hoard top AI talent. It's like Settlers of Catan in real life. Control the ports, block the roads, hoard the future.

Perfectly summed up, but that is going to do it for today's headlines. Next up, the main episode. Today's episode is brought to you by KPMG. In today's fiercely competitive market, unlocking AI's potential could help give you a competitive edge, foster growth, and drive new value. But here's the key. You don't need an AI strategy. You need to embed AI into your overall business strategy to truly power it up.

KPMG can show you how to integrate AI and AI agents into your business strategy in a way that truly works and is built on trusted AI principles and platforms. Check out real stories from KPMG to hear how AI is driving success with its clients at www.kpmg.us slash AI. Again, that's www.kpmg.us slash AI.

This episode is brought to you by Blitze, the enterprise autonomous software development platform with infinite code context. Blitze is used alongside your favorite coding copilot as your batch software development platform for the enterprise seeking dramatic development acceleration on large-scale codebases. While traditional copilots help with line-by-line completions, Blitze works ahead of the IDE by first documenting your entire codebase, then deploying over 3,000 coordinated AI agents in parallel

to batch build millions of lines of high quality code. The scale difference is staggering. Copilots might give you a few hundred lines of code in seconds, but Blitze can generate up to three million lines of thoroughly vetted code.

If your enterprise is looking to accelerate software development, contact us at blitzy.com to book a custom demo or press get started to begin using the product right away. Today's episode is brought to you by Vanta. In today's business landscape, businesses can't just claim security, they have to prove it. Achieving compliance with a framework like SOC 2, ISO 27001, HIPAA, GDPR, and more is how businesses can demonstrate strong security practices.

The problem is that navigating security and compliance is time-consuming and complicated. It can take months of work and use up valuable time and resources. Vanta makes it easy and faster by automating compliance across 35-plus frameworks. It gets you audit-ready in weeks instead of months and saves you up to 85% of associated costs. In fact, a recent IDC white paper found that Vanta customers achieve $535,000 per year in benefits, and the platform pays for itself in just three months.

The proof is in the numbers. More than 10,000 global companies trust Vanta. For a limited time, listeners get $1,000 off at vanta.com slash nlw. That's v-a-n-t-a dot com slash nlw for $1,000 off.

Today's episode is brought to you by Agency, an open source collective for interagent collaboration. Agents are, of course, the most important theme of the moment right now, not only on this show, but I think for businesses everywhere. And part of that is the expanded scope of what agents are starting to be able to do. While single agents can handle specific tasks, the real power comes when specialized agents collaborate to solve complex problems. However,

Right now, there is no standardized infrastructure for these agents to discover, communicate with, and work alongside one another. That's where Agency, spelled A-G-N-T-C-Y, comes in. Agency is an open-source collective building the Internet of Agents, a global collaboration layer where AI agents can work together. It will connect systems across vendors and frameworks, solving the biggest problems of discovery, interoperability, and scalability for enterprises.

Welcome back to the AI Daily Brief.

Advertising. Now, this is a space that I've spent some time. For a long time, more than a decade ago, I was working on a company that helped introduce big advertisers and big brands to new startup platforms that they might want to advertise and market with. Then, of course, I had my stint where I was making a huge volume of major national campaigns and Super Bowl ads and things like that.

And so even though I'm not in this space anymore, I definitely keep a side eye towards AI's particular implications for it. And it's been very clear for some time that there were going to be major implications for advertising and marketing when it came to AI. Now, this is an industry that has jumped headlong into this space. It's one where already creativity is in service of commerce,

And so perhaps some of the barriers that have stopped other industries, or at least slowed them down when it comes to the adoption of AI, haven't so much applied when it comes to advertising and marketing. Instead, the constraint has been more around just what's actually possible.

In a recent speech that I did, I was talking about why the speed of AI development is accelerating. And one of the things that we were talking about was the fact that each time models advance, it actively opens up new use cases. In other words, you're not just doing the same thing you did before, but a little bit faster, a little bit better, a little bit cheaper, although that might be part of it too. It's that there are things you simply couldn't do before that are now possible.

One big version of that recently, or one big unlock, was when Google dropped VO3. The biggest difference with VO3 is that it had built-in sound.

Previously, if you wanted to make, for example, a 30-second ad using generative media, you could do it. You could use Sora or Veo or Higgs field or whatever to create a video, and then you could use 11 labs to create audio for it and match it up. All of that was possible. And while it was, of course, a heck of a lot faster than an entire shoot in filming session, it was still a fairly complicated and complex endeavor.

Then Google drops VO3, and all of a sudden you can just build your sound and the action of the video right into the prompting. It took less than a month for that new capability to turn into an actual ad airing on major network TV. Earlier this week, AI filmmaker PJ Ace tweeted, "'Calchi hired me to make the most unhinged NBA Finals commercial possible.'"

Network TV actually approved this GTA-style madness. High-dopamine VO3 videos will be the ad trend of 2025. So for those of you who don't know, Kalshi is a legal betting site where you can bet on anything. And let's look at the ad that they just dropped for the NBA Finals with PJ. We're in Florida asking people what they put their money on!

OKC! Indiana got that dog in them! Will egg prices go up this month? I think we'll hit $20. How many hurricanes do you think we'll have this year? Cal-shee! Cal-shee lets you legally trade on anything, anywhere in the U.S. OKC! OKC! OKC!

Now, as an aside, as someone who loves the art of a good brand tagline, the world's gone mad trade it is maybe the most perfectly aligned with our modern moment tagline I've ever seen for a company.

In any case, PJ then goes on to share the process that he used to build this. His general process is 1. Write a script. 2. Use Gemini to turn it into a shot list and prompts. 3. Paste it into VO3. And then 4. Edit it in CapCut or some other platform. He also shared a little bit more of the concept. He writes, "Kalshi asked me to create a spot about people betting on various markets, including the NBA Finals. He said the best VO3 content is crazy people doing crazy things while showcasing your brand.

He then went on,

I then asked Gemini to take the script and convert every shot into a detailed VO3 prompt. I always tell it to return five prompts at a time. Any more detail than that and the quality starts to slip. Each prompt should fully describe the scene as if VO3 had no context of the shot before or after it. Redescribe the setting, the character, and the tone every time to maintain consistency.

He then goes on to add a bunch of other prompting tips. Ultimately, he said this took about 300 to 400 generations to get 15 usable clips. One person, two to three days. A 95% cost reduction, he said, versus traditional ads. Now, PJ did point out, just because this was cheap doesn't mean anyone can do it. I've been a director 15 plus years. Brands still pay a premium for taste. The future is small teams making viral brand adjacent content weekly, getting 80 to 90% of the results for way less.

I think PJ's right about the taste. I think he's wrong about the 80, 90% of the results, as we'll get into in just a minute. But this ad is actually playing during the NBA finals, and so I think reflects a pretty significant moment in the evolution of AI advertising. So what might some of the implications be? Well, the obvious one is that costs come down dramatically.

One thing that we should note is that already in advertising, significantly more is spent on the media cost, i.e. where you place the ad than on the creative itself. For an ad that costs a couple million dollars in production, it's probably going to be the headliner of a 20 million plus campaign. Still, when you can get really high quality video ads at a tenth of the cost now, it's going to have some dramatic implications for how things play out. And one of those is the sheer variety and amount of content go up.

In many, if not most cases, advertisers aren't going to look at their current $2 to $5 million production budgets for a campaign and say, great, we only have to spend $200,000 now. We can save all that and put it into media, although that's certainly one possibility. Instead, I think that they make more assets. The variety and amount of content go up, but the

and specifically get customized audience by audience. Already advertisers try to think about how different parts of their audience are going to react to different versions of content. And I think that customization and personalization is going to get massively more important as part of campaign planning because we just have the capability to do more of that customization. Specific audiences are going to get specific content that's tailored for them. Now this creates a really interesting opportunity as well.

Some channels where people receive content already have the ability to customize them in a pretty automatic way. Specifically, if you are interacting with advertising content on any sort of social media platform, the whole value proposition of that media platform for advertisers is that they know so much more about their users that they can really get the right ad to the right audience at the right time.

However, there still are other places that people consume media. And I think that we're likely to see new personalization infrastructure that actually helps route the right content to the right people at the right time. Think, for example, smart TVs that have small on-device models that are constantly learning about their user behavior that are able to interact with advertising distribution mechanisms to put specific versions of specific content in front of you based on what they see about you and your habits.

Interestingly, however, as much as the technology is enabling this, there are also some headwinds. One of them is, of course, the legal battles around copyright. Big new news on that front, Disney and Universal have sued Midjourney for copyright infringement. They allege that the company's AI models were trained on their copyrighted materials and easily reproduce characters like Shrek, Darth Vader, and Minions.

The complaint states, Midjourney is the quintessential copyright free-writer and a bottomless pit of plagiarism. Kim Harris, the general counsel for NBCUniversal, said that the companies filed the suit to, quote, protect the hard work of all the artists whose work entertains and inspires us and the significant investment we make in our content.

Now, let's be clear. Whatever you think about mid-journey or AI training practices in general, this is not about the artists who produce the content. This is about the investment in that content. These companies have become gigantic IP holders over the years, accumulating tons of properties through acquisition.

Disney's acquisition of Marvel and Lucasfilm were some of the biggest entertainment deals in history, adding properties like X-Men and Star Wars. Disney also acquired 21st Century Fox in 2019, bringing The Simpsons into their IP empire. Disney's chief legal and compliance officer, Horatio Gutierrez, says, We are bullish on the promise of AI technology and optimistic about how it can be used responsibly to further human creativity. But piracy is piracy, and the fact that it's done by an AI company does not make it any less infringing.

Now, this lawsuit is a little different from the numerous other copyright infringement cases being fought by the major AI labs. Thus far, most of the cases have largely dealt with the use of copyright material and training data.

Plaintiffs have struggled to make the AI models reproduce written works word for word, and that fact has allowed the labs to make a fairly arguable claim that AI training is fair use. In this case, plaintiffs have produced dozens of examples of images generated by Midjourney that depict copyrighted characters. The complaint stated, This case is not a close call under well-settled copyright law. This is textbook copyright infringement.

Now, as you guys know, usually at this point, I would go get you a bunch of takes from Twitter slash X. But honestly, when it comes to these copyright battles, the only takes are the screeching screams from the people who have made it their life's work to take down these AI companies. And while I am not at all being dismissive of their passion or their beliefs, it is somewhat ironic to me that the dislike of AI for some runs so deep that they're cheering for mega IP holders like Disney over small startups like Midjourney. But that's just where we are.

One interesting set of commentary came from Imad Mustaq, the founder and former leader of Stability AI. He wrote,

Hopefully all done by the book, as that's a dangerous one based on jurisdiction and sources. It's puzzling that Midjourney apparently didn't engage with Disney and Universal when they were contacted in December. It seems the plaintiffs want outputs blocked, which could be easy settlement versus punishment. So imagine that will be how this ends? Google Imogen, etc. use fully licensed inputs and have lawsuit protection on output as part of service. Chinese labs don't really care about inputs or outputs. Open source models from them have another legal status.

Now, those same folks that I was just mentioning got all up in his replies, asking him to please look at all of their work exposing Midjourney for the copyright stealer that they believe it is, but he just wasn't budging. He said, I did see and read it all as well as the whole case document. The overall case is weak legally, as I note, having spent a huge amount of money and time understanding the law precedent and more. Now, ultimately, this is all a bunch of armchair legal analysts, and these cases are, as I have said before, 100% going to end up in Supreme Court.

In fact, I think that almost all the legal decisions along the way until we get to Supreme Court are more or less completely irrelevant. There's just no way any copyright issues are getting decided by anything less than the Supreme Court. So it's kind of just a waiting game for all of these lawsuits to work their way up through the system to finally get there. What's more interesting for our purposes is to ask, what would happen if this did end up in a big victory for Disney and Universal?

Well, one possibility, as Ahmad points out, is that if these companies are simply trying to block outputs, that's an easy settlement that could mean it never gets there.

A second possibility, however, and one that seems extremely likely to me, is that the industry just shifts over to models that have been trained on licensed content. There are numerous companies positioning to be the recipient of that opportunity. Many of the people who are loudest, by the way, on Twitter and other channels about copyright infringement are leaders of those models, which is always helpful context to understand where someone's perspective is coming from.

But the big labs are also thinking about this as well. So again, another possibility, if this goes to Supreme Court and it does get decided against the AI companies, is that simply a different set of AI models that were previously behind, but frankly might be caught up by the time that the Supreme Court makes its decisions, step in, and from a user perspective, it's all functionally the same.

Another possibility is that one of these AI companies that's getting sued just goes scorched earth in crypto and goes fully decentralized, obviating the traditional structure of a company that could get sued in the first place. I don't think that's the most likely outcome, but it's certainly one possibility.

But then, lurking around all of this is, of course, China. There are a huge number of Chinese models, many of which are open source, which are going to have a very different respect or lack thereof for any sort of copyright consideration. And they will persistently be available throughout the entire time that all these legal battles are happening.

One of the really interesting real-world contexts that I don't know how it doesn't shape to some extent, at least the way that the legal system looks at this, is that this genie is fully out of the bottle. And when it comes to what companies do, there's no way they're going back to not using AI. If existing US-based models get banned for copyright infringement, people are just going to start using the Chinese models. Does the US need to then ban Chinese models or block companies from using them?

reasonable questions, but think about the precedent that we're starting to set there of how much control government and the legal system have over the decisions that businesses make. Point is, it's messy.

Ultimately, the TLDR is that AI video is happening one way or another. It's happening because people are going with bootleg Chinese models, or because the Supreme Court rejects the copyright arguments, or a fairly likely outcome that some company is in a position to take advantage of a legal decision if it goes against the current crop of AI companies. In the meantime, people are just going to keep innovating and releasing products.

Mid-Journey, for example, appears very close to releasing their video generation model. And it's exactly as aesthetically pleasing as you would expect a model from Mid-Journey to be. Point, guys, is that advertising is changing. It is changing fast. It is changing at an accelerated rate. And we have an AI-generated calci alien in the middle of a party premiering live all over the US as part of the NBA Finals right now. That's going to do it for today's AI Daily Brief. Appreciate you listening or watching, as always. And until next time, peace.