195 | The AI business application battle is intensifying, Self improving AI, AI for early cancer detection, and more AI news you should know for the week ending on June 6, 2025

25:03So it's still very confusing and still not very helpful to know what's real and what's not. Now, this obviously doesn't happen in a vacuum. Google released VO3 a couple of weeks ago. VO3 is mind-blowing and out of this world. And if you haven't seen VO3 videos, go and check anything on the Internet. Just search for VO3 videos on Google or on any other thing like an AI tool. and you'll see hundreds of incredible videos that will blow your mind. And so Microsoft didn't want to stay behind, and they're releasing Sora for free, so you can generate AI videos on the Microsoft universe as well. But while they're moving forward and releasing a lot of amazing stuff with AI, and if you don't know what I'm talking about, go back to the news episode two weeks ago after their big event.

25:41Not everything is going great. And while Microsoft is making very big announcement and releasing very impressive AI capabilities, there's still a lot of turmoil internally in Microsoft, and they're making another restructuring on who's going to manage what. This is the third major reorg of the AI Microsoft team since the beginning of 2024. If you remember last March, Microsoft acquired Inflection and its co-founder and the team Inflection. If you remember last March, Microsoft acquired Inflection co-founder Mustafa Suleiman and most of the Inflection team to create a new Microsoft AI organization.

26:17Last October, they hired former Meta head of engineering Jay Parikh to be the AI apps czar. And now there's a whole new change going on as well. So the chief Microsoft corporation for LinkedIn is taking charge of the team that will build email and productivity app. Ryan Roslansky, the professional networking site chief executive officer since 2020, will tack on responsibility for the teams behind Outlook, Word, Excel, and the rest of the Office bundle. He'll report to Rajesh Jha, who is a top engineering executive whose organization includes Windows and business software. And Roslansky, as the CEO of LinkedIn, will still report directly to Satya Nadella.

26:55So he's going to be wearing two hats, reporting to two different people. Now, the Dynamic 365 corporate vice president, Charles Lamanna, will also join Jai's organization as well. So lots of big shifts in very senior leadership and the reporting lines that all are tied to how are things developed within Microsoft to connect all their office suite and other things that they're doing to AI. that's never a good sign when you do a lot of reshuffles especially when these new reshuffles are not mentioning anything about Mustafa Suleiman and his role in this whole thing exactly what does he oversee right now is it just the development of new models for Microsoft unclear but I will update as the dust settles so on one hand Microsoft is delivering amazing new capabilities and vision and on the other hand not very clear what's happening there internally right now and from Microsoft to Anthropic, we'll start with some not very good news for Anthropic.

27:48So Reddit just filed a lawsuit against Anthropic for breaching contract and using user data without consent. The lawsuit claims that Anthropic attempted to access the platform's data more than 100 ,000 times between July 2024 and May 2025, despite the digital restrictions and the formal requests that were sent to Anthropic to stop this behavior. Now, we had many lawsuits before, or mostly from different publishers, against the big model labs. But this is the first time that a big tech company is suing one of the giants for training on their data, or at least attempting to train on their data. Where will this go?

28:25I don't know. But this is just another angle where we see this battle between the people who create and own the content to the people who believes that they can use the content as they wish. If you want the exact quote from Reddit, they said, despite what its marketing material says, Anthropic does not care about Reddit's rules or users. It believes it is entitled to take whatever content it wants and use that content however it desires with impunity. Now, if you want to make the story a little more interesting, Sam Altman, the CEO of OpenAI, owns 8.7 % of Reddit, making him the third largest shareholder.

29:00And he was once one of the board members of Reddit as well. OpenAI themselves have a licensing deal with Reddit. And so this might be a play by Reddit. Now, I'm not saying Sam is involved in this, but it's a very obvious play by Reddit to get another licensing deal. As I mentioned, they already have one with OpenAI, and they have a similar agreement with Google. So I believe they'll be able to twist the arm of Anthropic to do the same thing if they want access to Reddit's information. Now, the response from Anthropic was very generic. We disagree with Reddit's claims and will defend ourselves vigorously.

29:32What does that mean? I don't know. But as this evolves, I will let you know. Staying on Anthropic, they just announced on June 5th that they're launching Claude Gov, which is a specialized AI model for U.S. defense and intelligence agencies. It is similar to and competing with a system that was released by OpenAI called Chachupiti Gov. In their statement, Anthropic revealed that Claude Gov has already been deployed for several different agencies at the highest level of U.S. national security, though they didn't specify which agencies it was deployed and exactly when. And unlike, obviously, the consumer version that we have access to, it has a lot less guardrails and it's running within a secured environment that allows the government to use it with huge amounts of data without the risk of that data falling into the wrong hands.

30:17Connecting some of the dots for you, on November 2024, Anthropic announced a collaboration with Palantir, who has been providing multiple AI platforms to the government before on AWS Secure Cloud. So it's a tight partnership between several different companies who are now providing secure AI capabilities for the government. Now, where does that lie with the company's core values when it comes to Anthropik? They're saying it's perfectly aligned with it because their current guidelines are saying that legally authorized usage for things like foreign intelligence analysis is allowed while it prohibits disinformation, weapons development, censorship, and malicious cyber operations.

31:01in my eyes this is a very slippery slope especially once you're giving the government access to a secure environment that you may not have access to even though it's your technology you don't really know what they're going to use it for and so this is a very very big problem and i shared with you my thoughts before just like the race on the civilian side the race on the military side is going to be as fierce because nobody will want to be the one that doesn't have this technology and allow the other side to have it. So having AI autonomous weapon systems and so on is something that's coming. It's coming fast and it's going to change future wars dramatically with AI making decisions instead of humans making decisions.

31:39And that opens a whole can of worms that you can think about on your own, or maybe we'll do a whole episode about that. But it's definitely not something that I'm happy about. Staying on Anthropic. Anthropic just launched Anthropic Explains, which is a blog primary written by Claude. There is human oversight that is looking at the outputs and making sure that they're aligned and making final changes and the humans are in charge of the final product. But most of the blog is written by Claude itself, and it includes articles on a wide range of things, including highly complex topics such as simplified complex code bases with Claude is one of the articles that Claude itself has written.

32:16That comes to show you several different things. One, how good these tools are becoming, and I actually really like the way Claude writes blog posts and longer pieces of content and short piece of content as well, to be fair. It also shows you the importance of this collaboration between human writers and oversight to AI writing the initial content. And that's the way Anthropic is following, at least right now, or at least this is what they're saying. I can tell you that as these systems get better and better, I think humans will trust them more and more, will verify less and less. And we're going to get information and assume it is correct and accurate and is aligned with our needs and values because we just are going to stop reviewing what it's doing.

32:51I don't encourage that. I just think that's human nature. But to show you the power of that and the level of content that it is generating, Claude's posts are gathering 200 ,000 plus unique reads in its first week. I don't know many blog writers that ever got that. Now, obviously it's Anthropic. They have a big following. Obviously there's the whole thing of, I want to read the blog post that was written by Claude to see how good it is and see if I can tell the difference. But it's still a very large number of people who have read the blog that was written not by a... Staying on Anthropic and its relationships with other companies, Anthropic just cut Windsurf's direct access to its models.

33:26So we talked about Windsurf several times in the past. They are one of the leading professional vibe coding platforms out there. There are serious rumors that they're going to get acquired by OpenAI for$3 billion. And so because of these news that apparently are materializing, Anthropic decided to cut their direct access to all their models. Windsurf CEO went to X and basically complained about the whole situation, said that it does not understand this move, said that they're willing to pay more to get access to the models, and that the fact that they cut them off within only a five-day notice is completely unprofessional and unfair.

34:01And on the other hand, Anthropics co-founder Jared Kaplan told Tech, I think it would be odd for us to be selling Claude to OpenAI, which makes perfect sense to me. And Anthropics' Steve Minch said, we're prioritizing capacity for sustainable partnership that allow us to effectively serve the broader development community. Now, to be fair, you can still use Claude's models on Windsurf, just not directly from Windsurf. You have to bring your own API keys, which is not a big deal. The problem is it's a lot more expensive than running it through the Windsurf environment. And I'm sure they're going to lose clients to other similar platforms out there, such as Cursor.

34:37From Anthropic, let's switch to OpenAI, which is potentially maybe acquiring Windsurf. And this topic is just a quick mention. A new book was just released by investigative journalist Karen Howe, and the book is labeled Empire of AI, which is following everything that happened from OpenAI's launch to becoming a$300 billion for-profit giant. So Howe began covering OpenAI in 2019 for MIT Technology Review. She has interviewed multiple people. She had access to a huge amount of documents, and she has a very unique view into everything that happened and happens in OpenAI, which is obviously a fascinating read.

35:16I didn't get a chance to read it yet, but I'm letting you know that it exists. And if you want to get a lot more inside information on what has evolved and what has happened in OpenAI from being a small research lab nonprofit for humanity to becoming one of the most powerful companies in the world, then I think it's a must read. And now to specific news from OpenAI. OpenAI just opened its Codex coding agent to ChatGPT plus users. So if you're paying 20 bucks a month, you now have access to that. Before that, it was only available in enterprise teams and pro tiers. So now it is also available to most of the people who are paying for ChatGPT.

35:51Now it's available to all the people who are paying for ChatGPT. Codex can now connect to the internet, which is also a new thing, which allow it to find pieces of code and dependencies and instructions and API documentation and so on. on its own. It also allows it to connect to staging servers and running tests with external resources and so on. So it's becoming more and more powerful. Going back to what I said in the beginning, this is starting to look more like an IDE and compete with tools like Cursor and Windsurf. And definitely it's built in the first step in order to compete with the coding capabilities within Cloud, which are very impressive.

36:26I've actually built several different things in Cloud already, mostly simple stuff, dashboards and games with my kids. But the coding and execution capability in Cloud is really amazing. And so it's not surprising that OpenAI is allowing now access to everybody to use Codex. I haven't tried Codex yet, but I am going to compare the two and give you my opinion sometime in the next few weeks. OpenAI is also planning to launch O3 Pro, which similar to the O3 that we're all using, it's going to be only offered to the Pro subscribers who pay$200 a month, and it will provide additional computing power, but also enhanced reasoning capabilities beyond the basic model.

37:04Well, is that going to be the thing that's going to drive more people to pay$200 a month? I don't know. I actually really like O3. I think it's very powerful and it's doing a lot of things, but I don't think I will pay 10x just to get a little more of that. And some good news also for the free tier users of OpenAI. OpenAI is just rolling a lightweight memory feature for the free tier. So it works very similar to the way memory works for the paying users, but it only has short-term memory instead of long-term memory, meaning it's going to bring the context from your recent conversations into your new conversation, but it's going to forget that information that was older.

37:39I actually find the memory capability to be very helpful in ChatGPT, and I hope I can have it in some of the other platforms in the same efficiency that ChatGPT does it. So this is really good news for the free users. This is not available to people in the EU, UK, Switzerland, Norway, Iceland, and Liechtenstein due to strict AI regulations in those places. And if you are a free user, you can disable the memory and control what it's actually remembering, just like everybody else. And from OpenAI to Apple, which we'll probably talk a lot more about next week because their worldwide developer conference is happening this coming week, but they just shared that they are testing a large language model with 150 billion parameters, which per them, and I'm quoting, approaching the quality of recent ChatGPT rollouts.

38:24Now, Apple is actively testing different levels of this model with 3 billion, 7 billion, 33 billion, and 150 billion parameters using an internal tool they call Playground that allows them to benchmark their models against other models, mostly ChatGPT. Now, the large model, the$150 billion model, runs on cloud computing similar to other large language models and is obviously outperforming the on-device 3 billion parameter of Apple intelligence. We talked about this a lot on this podcast. Apple's performance on the AI field has been nothing short of embarrassing so far. They haven't released anything significant.

39:01The things they have released are very small components of the bigger picture that they've promised. A lot of people have bought new Apple devices and have upgraded their iOS platforms to get capabilities that did not show up. There have been a turmoil in leadership over there. There are people removed, new people were hired, and they're basically scrambling to figure things out. The new Siri that was supposed to be released last year will maybe be released in 2026, and it might get pushed to 2027. This is not looking good for Apple. And a part of these announcements, Google Gemini is expected to join ChatGPT as a Siri backend alternative for iOS 26, with talks also going on with perplexity to potentially be a part of Siri and maybe Safari search.

39:44What does that mean? Well, as users, it means you're going to have more options, which is a good thing. It is also very clear to me that the new partnership between OpenAI and Johnny Ive is a big threat to Apple with its iPhone dominance in the market. And I think they're terrified of that. So I think they're looking for partnerships that are not with OpenAI to have some other differentiators to stay relevant. and there's also two trials going on. One is against Google Monopoly, which part of that is trying to break their$20 billion a year deal with Apple to be the default search on Apple's devices.

40:18So that might be another reason for Apple to move away from Google search on their devices to other solutions. And as I mentioned, their largest conference of the year, the WWDC 2025 is starting on June 9th, and it will be very interesting to see how much AI they're actually going to be there. I actually believe there's going to be a lot less than we've seen last year because of the very serious backlash after last year was everything Apple intelligence and they failed to deliver on everything they promised. So I think it's going to be a lot more low key when it comes to AI this year, but I will update you next week.

40:51One additional anecdote on Apple, they just dropped from third to fourth place on Fortune 500 company with UnitedHealth Group overtaking them following Walmart, Amazon, following Walmart and Amazon, which are still holding the first two places. And I heard a very interesting podcast this week that is asking if Apple is the next Nokia. Now, I know that sounds insane. They're one of the most loved companies in the world. They have the iPhone and the iPad and the earbuds and the Macs and so on. And yet Nokia was in the same exact situation before the collapse. They were the largest behemoth in the mobile cellular world by a very big spread.

41:30And I don't know anybody who has a Nokia phone right now. Why? Because they missed the trend of smartphones and they were not fast enough to adapt. And when they did adapt, it was too little and too late. Now, is the same fate is going to happen to Apple? I don't know, but it's definitely not looking good for Apple in the last couple of years. Combine that with the fact that, as we mentioned, Johnny Ive now has a partnership with OpenAI to build something that may take away market share from phones. Combine that with the fact that Meta has actually built glasses that people actually use and love versus the really sophisticated, really advanced headset that Apple developed that very few people bought because it was$3 ,000 and really heavy.

42:08Combine that with the fact that there's a trial going on against OpenAI when it comes to allowing people to open a secondary app store that Apple doesn't control and cannot take 30 % of the profits off the top and that they are facing pressures from multiple directions and facing pressures from multiple directions. And this may lead to results that are very different than what we know from Apple right Now, that being said, their stock is seemed to be holding pretty steady despite all these issues. But if you look at Apple in the past decade, there's been very little innovation. Again, the only big thing that you can talk about that was like, oh, my God, this is a new thing was the headset, which was a huge failure.

42:44And so no big innovations coming from Apple, the company who maybe had innovation as its main driver, introducing new things to the world that didn't exist before. And now one of the people that helped them lead that is running against them, and they're facing a lot of other issues. So again, I'll keep on updating, but very interesting point of view about Apple's current status. From Apple to Google, Google just announced a new cool feature for Notebook LM, which is public sharing. You can now share your notebooks with anybody in the world, even if they don't have a Google account. What does that mean?

43:14It means that one of the best tools in the world today to summarize information and allow asking questions about information, you can now make it a much more collaborative environment. Teachers can create interactive study guides for students. Startups can build product hubs that multiple people can participate in and learn about. Researchers can share findings through that. And business people can share project data and so on with other people. And so the tool just becomes more collaborative, which is great. I use Notebook LM all the time. I think it's a fantastic way to summarize information from multiple sources, be able to ask questions about it and so on.

43:47And from Google to Perplexity introduced Perplexity Labs. last week, which we talked about, which is a really cool tool. I still think it's behind the more agentic tools like GenSpark and Manus, which have way more advanced agentic capabilities compared to Perplexity Labs. And if you want to know more about GenSpark and Manus and tools like that and how to run them safely, don't miss the next episode that's coming out on Tuesday. This is exactly what we're going to dive into. But staying on Perplexity, their CEO, Arvind Sweenivas, predicts that AI agents will redefine how we interact with the web, moving beyond answering questions to take actions like booking rides, ordering foods, and everything else that we do in the internet today.

44:24Now, Srinivas aims to disrupt current search by building AI agents that will integrate seamlessly with the apps. And as we mentioned previously, they're coming up with a new browser called Comet that will have everything integrated into it. And this browser is supposed to be launched next month. So it's just around the corner. There are many other companies right now that are building quote unquote AI based browsers. It will be very interesting to see how that works, but the direction is clear. We're going to have a lot more agentic approach to interacting with the web. We're going to see less and less human traffic to websites as time goes by, and companies who rely on human traffic going to their website, whether for e-commerce or for information, needs to understand that this is going to change.

45:05It's not going to happen overnight, but over the next five years, there's going to be a very clear increase in agent traffic to data, not necessarily websites, and a very significant decrease in human traffic to these websites. And that means that companies have to start adjusting to that and figuring out what that means from data structure, architecture, and other aspects of the way they're engaging with their customers. Now, just like the potential iOS collaboration that I mentioned before, Perplexity secured a deal to pre-install its AI on Motorola's new Razr phones, and Srinivas credits Google antitrust scrutiny for loosening grip on OEMs that are now allowing them to install their search and other capabilities on the phones instead of or in parallel to Google's offerings.

45:46And from perplexity to Meta, Meta just announced in their shareholder meeting that they're aiming to automate the entire creation of ads by the end of 2026. We shared that with you a couple of weeks ago, but now we got more details on what the plan, but the plan is very extreme. And I'm quoting Zuckerberg from the meeting. In the not too distant future, we want to get to a world where any business will be able to just tell us what objective they're trying to achieve, like selling something or getting a new customer, how much they're willing to pay for each result and connect their bank account.

46:19And then we just do the rest for them. What does that mean? It means that you don't need any kind of agencies and that it will optimize automatically down to the individual level. What does that mean? It means you don't need agencies because it will create the ads, it will create the copy, it will create the text, it will create the images, it will create the video. It will create everything that it needs. It will optimize down to the personalization to a specific person based on their interest, based on their hours, based on how they consume, what they clicked in the past and so on. Stuff that no agency and no human can do right now.

46:49And it will optimize for all these things. You're obviously in a question whether you're allowing your creative to be controlled by a machine. But if that machines know how to achieve the results better than you can with your creative, then there's no problem. Well, there is a few problems. Problem number one, that you lack control on the messaging that is going to be presented on behalf of your company, which might be a problem, especially if you have a brand name that you're trying to preserve. I'm sure they will have some tools to guardrail what it can and cannot say and what brand guidelines to follow and so on.

47:19But that wasn't clear so far. The other thing is that it's going to kill, I don't know how many, but many, many, many different agencies who specialize today in doing exactly that, which is doing social media advertising on behalf of clients. There's been similar announcements already by Google and TikTok saying that they're going to follow similar concepts. So the idea of an agency that builds ad and distributes ads free because they know how to do it better than you might be the thing of the past within a few years from now. Staying on Meta, they just announced that they're going to switch most of their risk assessment from human review to AI review on Facebook, Instagram, and WhatsApp.

47:53Whether that's good or bad, I'm not 100 % sure. There's goods and bads in both of it. I'm sure AI can look at more stuff. I'm sure AI can classify things better. I'm sure AI is going to miss things that the humans would have caught. And so I really hope that's not going to come to bite us, you know where. But for now, this is the direction that Meta is taking. By the way, similar to Microsoft that we talked about before, Meta is also going through some serious restructuring in its generative AI group, splitting into two units to address what they call, quote unquote, internal challenges. So in the new leadership structure, Ahmed Eldal and Amir Frenkel will co-lead the new AGI Foundations team focusing on LAMA models and AI agents.

48:34And Joel Pinyu, VP of AI Research, will oversee the team dedicated to AI integration in Meta's consumer apps like WhatsApp and Instagram. So one more the front end, one more the back end, if you want. And this actually makes sense to me. If you go back to how we started this episode, we're talking about the fact that the application layer is going to be as important and I think in the long run more important than the underlying models. Splitting it into two teams makes perfect sense to me. And now to some very interesting announcements from smaller startup and not just from the giants. Phonely, which is a company that generates phone agents, has just achieved 99.2 % conversational accuracy, surpassing the previous king, which was OpenAI's ChatGPT4 all with 94.7%.

49:15Now, the biggest difference was cutting response time and latency by over 70%. How did they do that? Well, I'm sure they're optimizing the models like everybody else, but they also did it through partnership with Grok with the Q. So Grok that develops and provides the most advanced inference platform in the world today. So these are chips that are not optimized for training AI like the GPUs from NVIDIA, but actually are specialized in inference, meaning generating tokens. So they have what they call a multi-LORA hot swapping that enables instant model switching in real time without any latency.

49:49And they're also using Mai Tai's platform to optimize the performance of everything that they're doing. And they were able to reduce the response time from 661 milliseconds, basically over half a second, to 176 milliseconds, which is what humans do. So it reduces dramatically the ability of humans to actually know they're talking to machines. based on their own internal research, they're seeing that about 70 % of people cannot tell that they're talking to AI when they're talking to their platform. That's obviously coming from their own CEO. So I don't know how credible this is, but I can tell you for sure that these agents are getting very, very good.

50:27They're doing an incredible job in providing customer service. They're going to get really good at doing outbound sales and inbound sales and a lot of other stuff. And that's another industry that is at complete risk of elimination, which is the industry of the call center and contact center. I think five to 10 years from now, the concept of contact centers will just cease to exist. There's still going to be people doing maybe higher level, more relationship kind of things, or being supervisor to assist in stuff that the AI wasn't able to solve. But that's going to be single digits percentages of the amount of people that are currently employed in the call center industry.

50:58Another company that made a huge release this week is Heygen. Heygen is a platform that enables to create human-like avatars for any need that you have, whether it's training, customer service, onboarding, et cetera. I use Haygen all the time. I teach Haygen in my courses. It's a fantastic platform. Well, they just launched AI Studio, which allows to take the creation of the videos and the realism of the avatars to a completely different level. It includes multiple components. Most of them are prompt-based, so you can prompt how the video will look like, but they have now new functionality like Voice Director that allows a much better fine-tuned avatar speech, including the ability to do voice mirroring, meaning you can record your own voice and then it captures the nuances of how you speak.

51:40And then you can apply those nuances to any voice that you can pick from their platform. They're also allowing you more gesture control of how the actual avatar moves their hands and what kind of gestures they're going to have. And they're going to be releasing additional capabilities like prompt control over camera, motion elements, and prompt-based editing and streamline, including B-roll adding, all of that with a lot of AI capability. As I mentioned before, I really like HeyGen. I really like their offering, and I'm very excited to test this out. And I promised you in the beginning that there's going to be some additional news when it comes to the scientific benefits of AI.

52:13Well, the big one is that the FDA just approved a tool called Clarity Breast, which is a breast cancer detecting platform. So the Breast Cancer Research Foundation just announced the authorization of Clarity Breast to be used with actual patients. That's a big milestone for AI because it is the first AI platform that will be used for similar things in the public. It can predict a five-year risk for breast cancer using only standard mammograms. So today, the way breast cancer risk is analyzed is based on a mammogram with a human looking at it combined with other risks such as family history. Well, this tool actually looks at very small nuanced changes in mammograms and using that because it has so much information in the past, it can make a much better prediction of future risk.

53:02This is fantastic news for women, and it's fantastic news for the research of cancer because similar things might be applied to other ways of predicting cancer. And as we all know, catching cancer when it's in the early stages dramatically increases the chances of a successful recovery. And so I find this to be really amazing and great news. And in another breakthrough in a field that is less relevant to most of us, but it's still very interesting, New AI techniques allows researchers now to have better analysis of cosmological parameters and allowing them to better predict and understand how the universe works.

53:39These new tools were used with huge amounts of data from Sloan Digital Sky Survey BOSS data set, and it allows researchers to identify subtle patterns that were invisible in traditional methods and helping increase the accuracy of predictions and information by 30%. This is a huge increase. Again, it doesn't have any daily application to us, but if you take that to the world of research and you combine these two last pieces of news, it shows you that AI's ability to look at huge amounts of data, whether visual, numerical, or other, and make sense in that data is going to allow us to drive significant new innovations and research in many fields, which is really exciting.

54:16That's it for today. We'll be back on Tuesday, as I mentioned, talking about general agents like Manus and GenSpark and how to use them safely. So I highly recommend you check that out. I think it's going to open your eyes to what's possible today, which most people do not know. Quick reminder for the course that is coming up in August. So if you're looking to take our AI Business Transformation course, you should check out the link in the show notes right now. And if you're enjoying this podcast, please share it with other people. You can do this right now. When you're done, open your phone, click on the share button and share it with four to five people that can benefit from listening to this podcast.

54:48You're helping drive AI literacy. And I also be very grateful if you do that. And until next time, have an awesome week. you