201 | 70% of job skills changing by 2030, 22 all-new AI careers, OpenAI’s stealth Office rival, AI ROI revealed and more AI news for the week ending on June 27, 2025

23:05That means that on the aggregate level, we can know everything as well if we are willing to provide, let's say, anonymized information to the next level up. I think in a company, you won't even have an option. I think in a society, you'll be able to opt in and opt out of that. But I think the benefits are going to be immense that it will be very hard to opt out. This could turn into an amazing utopia where everything just happens in perfect order and timing. This could also be 1984 by George Orwell. And it just depends whether we as humanity push this in the right direction or in the wrong direction.

23:36And we have, I think, a 50-50 chance of getting it right and getting it wrong. But I do see that as a very logical path that I haven't heard anybody talking about. But I, again, just running the numbers and running the trajectory of where everything is going right now, I fear or excited about, and both actually, that this is where we're heading, maybe not in the next five years, but in the next two decades. But to bring us back to here and now versus a potentially completely crazy foo-foo kind of future, in an article in Fortune magazine, they shared some additional insights from Sam Altman about how he sees AI performing tasks right now.

24:11And I'm quoting, in some sense, AIs are like a top competitive programmer in the world now, or AI can get to a top score on the world's hardest math competitions, or AI can do problems that I'd expect an expert PhD in my field to do. So this is a quote from the Uncapped podcast which Sam participated in and was just recently released. But what he's basically saying is that AI right now in specific fields, in specific tasks and outperform most people. And the first thing it's hitting is obviously entry-level jobs. So the unemployment rate for bachelor degree graduates jumped to 6.1 % in May of 2025, so basically right now, from 4.4 % in April.

24:51That's a 50 % increase in one month. And that's just the average. In fields like commercial art, graphic design, and computer engineering, the rates are above 7%. And this is information from the Federal Reserve Bank, not some unknown publication somewhere. Now, in this interview, Sam stays optimistic like he is in every interview that I heard him talk about this or asked about this. And he's saying, and I'm quoting again, a lot of jobs will go away. A lot of jobs will just change dramatically. But we have always been really good at figuring out new things to do. As I mentioned many times in the past when I heard these kind of comments from top leaders in the AI industry, I have two serious issues with that.

25:25One is what are these things? He and people like him, like Dario and so on, never actually provide tangible examples. Now, yes, they may not be able to think about it, but I think they need to because they're the ones that are driving us in that direction. Now we can go back to the New York Times article and think about the 22 things they suggested. But many of these things that they suggested will not be required at the same scale of the jobs they're going to take away. And as I mentioned, some of them will be in place in order to take other jobs away. So this is still puts us as a very bad net outcome when it comes to how many jobs AI will create versus take away.

25:56And this is really, really scary. The other aspect is obviously every previous revolution allowed us to do more with our brain and less physical job. And this is the first revolution that goes after our thinking capacity, the brain capacity that we have. And I'm not sure what we're going to replace it with. Maybe, like some of these articles suggest, our emotional intelligence and our creativity will be the next frontier that humans will focus on while AI will do the actual tasks. How does the world look like in that situation? How do everybody makes money? I'm not 100 % sure, but I guess we'll have to figure it out in the next decade or so.

26:28Another organization that shared a very interesting article this month is an organization called the 80 ,000 Hours Organization, and their article is labeled, How Not to Lose Your Job to AI, the Skills AI Will Make More Valuable and How to Learn Them. So they did a very large survey, which they just published, as I mentioned, this month, and they're claiming that based on this survey, professionals who excel in problem-solving, strategic thinking, and emotional intelligence will thrive in roles that complement AI's capabilities. So as we mentioned before, if you have these skills and you know how to use AI, which is a prerequisite to the outcome, you are very likely to be successful in the AI era because you're adding to the AI what it is lacking, at least right now.

27:05So 67 % of AI experts that were surveyed in the study share that they believe human oversight will be essential for high stakes decisions by 2030. Skills like brainstorming and artistic expression are going to be critical as AI struggles to replace human originality. And again, I think as of right now, I think this will change as well. So large-scale design, storyboarding, and brainstorming ideas and coming up with strategies is things that can be enhanced with what AI can do and hence will be more significant in the future. Emotional intelligence and communication are also mentioned as very important skills and leadership and collaboration as well.

27:39So 72 % of global executives shared in a Deloitte survey that they are prioritizing soft skills for future workforce needs. Why? Because the hard skills AI will be able to do, and that makes perfect sense. This article also mentioned potential new careers such as AI ethics advisors, regulatory specialists, and things around the relationship between AGI and human values in the human world will become norm. The problem is obviously how many of those we actually need in every single organization. And they have a really cool table that summarizes a lot of what they're saying that has specific skills, why they're valuable, and how to start doing them.

28:15And the skills that they're mentioning are using AI to solve real problems, which is exactly what I teach in my courses and my workshops. Personal effectiveness, which is how to increase your own personal productivity that will allow you to do even more for your organization. Leadership skills, communication and taste. We talked about taste in the other article as well. Getting things done in government and complex physical skills. And when they say complex physical skills, they're adding, especially for specialists, work in unpredictable environments. So AI robots will be able to do amazing things in a confined, well-defined environment like a factory, but you put them out in the field or in a setup where it's not clear what's going to happen and they will struggle again, at least in the beginning.

28:53They also shared a very interesting statistics from another organization called EPOC, E-P-O-C-H, AI, Gate Economic Model of AI Automation, and what they're predicting based on their models, and it's a lot of algorithms that are running a lot of information. That organization has a lot of other AI-related statistics that they're sharing. But in this particular chart, they are anticipating what's going to happen to wages, average wages, because of AI usage. And they're showing that initially it will grow tenfold, meaning those people who will know all these skills on how to use AI and drive a lot of value to their company will be able to make 10x more money that they're making right now.

29:27And if you look at the graph, it becomes steeper and steeper all the way to about 2035. Then it starts to level off. But then towards the end of the next decade, so in like 2038, it goes down to zero. So there's an exponential increase, then there's a slowdown, and then there's a cliff that wages go down to as we figure out how to let AI do, well, everything that we're doing right now. This is about a decade away from now, meaning they anticipate that, yes, people with AI skills, which, as I mentioned, you need to take care of yourself and of your organization if you're in leadership position, will make significantly more money in the next decade.

30:02Awesome. But then the salaries of everyone basically goes down to zero based on what they anticipate. And how does the world operate then? I don't know, but we got to start figuring this out. And in addition, this particular graph doesn't touch about all the people who did not develop advanced AI skills and are not going to make 10x and just going to lose their jobs in the process between now and that time. And these people are the people who make right now 80 ,000 to half a million dollars a year. And these are the people who drive the economy. And if 20, 30 % of them are out of a job, what happens to the global economy?

30:33That's also a question that nobody, I think, has an answer to right now. And another interesting data point that was released this week is a study from the National Association of Productivity Analysts, NAPA, NAPA, reveals that 68 % of professionals are adopting AI to escape repetitive workplace tasks. And the findings highlight growing reliance of people on automation tools, including ChatwPT and tools like Zapier, to boost efficiency, with some employees citing 30 % time savings on average. In the people they surveyed, ChatGPT leads with 52 % of users, and Zapier followed second with 38%, then Microsoft Copilot with 29%.

31:07Young workers are significantly more likely to use these kind of tools, and employees between the ages of 18 to 34 account for 78 % of AI advanced adoption. Now, 55 % of respondents seek AI skill workshops and courses, which is still relatively low in my eyes. I find these numbers, again, I didn't do the research. I don't know exactly who they surveyed. I can tell you that I speak on a lot of stages. And every time I speak on a stage, I ask the same question. Which bucket would you put yourself in? Are you a total beginner, meaning you heard of ChatGPT? You may have used it once or twice, but you're not really using it regularly.

31:39Bucket number two is you're using tools like ChatGPT regularly, but you still don't have any business application for it. Number three is I'm a beginner business user. I have several different use cases in which I use AI regularly at my work. And number four, I am an advanced user with well-defined processes, integrations, deployments, and so on across multiple aspects of my business and my business day-to-day job. And still, I've done this this past Friday on an audience of an alumni of an MBA program, meaning most of them are in senior leadership position. About 70 % to 75 % of people self-identify as one of the two buckets.

32:11So I do not understand, and I do this in front of a huge variety of audiences, more senior, less senior, multiple industries, multiple places around the world. And it's the same thing every single time, unless I'm going to a more technical audience, and then the numbers are obviously significantly higher. So I do not understand how this study found that 68 % of workers automate their work into a 30 % increase in efficiency. I see that as a very niche audience that actually is at that point, but this is the direction that it is going. If you do not know how to do that, whether it's 68 % of the audience right now or 20, it will be 100 % within two years.

32:46And so if you're there first, you have a much higher chances of keeping your job and being valuable to your employer or to yourself. And if you're not, you may find yourself in a much more difficult situation. So definitely, whether you're going to take my course or somebody else's, make sure you're getting AI skilled and that you know how to use it to drive efficiency to the stuff that you're doing and for the people around you. And from the individual level, there's another interesting point about the enterprise level. IBM has just released an update on how they see their clients using AI in the enterprise level, and they report a 40 % increase in enterprise AI since 2023.

33:19And they're also saying it comes across industries and sizes of companies, and they provide a few interesting data points. 60 % of IBM clients use large language models for automated support, cutting response time by 35 % on average. They're talking about Pfizer and how they're using it for drug discovery, driving a 25 % efficiency boost, 45 % of enterprise. They're saying that these enterprises use a mix of open source and proprietary large language models with 70 % of their clients using this hybrid strategy. This makes a lot of sense to me because open source, while it doesn't provide the same edge capabilities as the closed source tools, they provide data privacy that you can run within your organization and also flexibility to do things cheaper or focus on specific topics that you can train the AI on.

34:00So it makes perfect sense to me. And they are stating something that I see with each and every one of my clients is that the biggest challenge companies have is matching the right AI tools to the right use cases. There is no one size fits all. I can tell you that with all the companies and all the individuals that I work with, whether in my courses or the companies that I consult to, you have to learn how to continuously evaluate the AI tools and match them to the relevant tasks in your organization. And this has to happen through an AI committee. So not every single individual does it on their own because otherwise you get complete chaos, but defining and creating a mechanism that does this ongoing will drive increasing efficiencies over time.

34:36So that's it for our very long and very important deep dive on AI impact on jobs. And now we're going to run through the rapid fire very, very quickly, starting with OpenAI. So it's been revealed that OpenAI has been secretly working on a rival environment for Microsoft Office and Google Workspace. Apparently, they've been working on this for the last two years. That's been shared by the information. So they're building the ability to have features of real-time document collaboration between people and AI, including targeting directly Microsoft Office and Google Workspace tools. Now, OpenAI introduced Canvas in October of 2024, which those of you who don't use Canvas, you're missing out.

35:13I use this every single day. It's currently the best collaborative environment to work together with an AI to create, draft, edit, and improve documents or code. I use it as a first draft for more or less everything that I do right now because I really like that collaborative environment that shows you the direction that they're going. It's a personalized experience who has information about you, connecting you back to the point we talked about before with Sam Altman. And it knows about you. It knows about your business. It knows about your style. It knows about your customers, et cetera. And it helps you digest information and write documents significantly faster than you could do right now.

35:43Now, if you connect the dots to other things that OpenAI are doing right now, so they are planning to develop a browser so you can do everything inside their universe, an AI-powered hardware that will know and listen to you 365 days a year, 24-7, if you will allow it, going back to what Sam said, a social content feed and connectivity to everything in your current work environment through MCP and direct connection to your email and data and so on. So they are trying to take over everything, including search, right? So they're replacing search as well. Now, designs for all these tools began, as I mentioned, over a year ago under the product chief, Kevin Whale, and the progress hasn't been as fast as they want because they have limited resources and staffing shortages, but it's very obvious that the direction that they are going.

36:28This obviously puts a lot of pressure on Google to move faster with what they're doing, and it also adds more gasoline to the already pretty big fire of the relationship between OpenAI and Microsoft. And on that topic, there were several interesting articles and inputs from both Sam Altman and Satya Nadella on the growing pains of this relationship. Sam Altman confirmed in a podcast with the New York Times that, and I'm quoting, talks are ongoing and we are optimistic we will continue to build together for years to come. But he also said, obviously, in any deep partnership, there are points of tension and we certainly have those.

37:00Now, I shared with you last week that OpenAI is now deeming the reluctance of Microsoft to approve their new business structure as anti-competitive business practices. And they're considering moving to court with that. That would obviously create a very, very big mess. Microsoft is also trying to move away from the revenue sharing mechanism with Microsoft to giving them a share in the new corporation. That being said, Microsoft wants a significantly bigger share than OpenAI is currently willing to give them. From Microsoft perspective, slowing the timeline and holding back is a nuclear option, right?

37:31Because if by the end of this year, OpenAI does not complete the process, they will lose a lot of money and it will also make themselves very vulnerable for hostile takeovers as many of their investors will have to decide how to move forward. So behind the scenes, Microsoft basically said that they're willing to just walk away from the current negotiations and maintain their current contract with OpenAI to keep the status quo as is. This obviously puts a lot of pressure on both sides to figure this out. And I have to assume, because both companies, I think, have more to lose than to gain if this thing blows up, that they will figure something out in the next few months.

38:03On the technical side, OpenAI made a huge announcement this week on new capabilities to the API that I'm personally very excited about, and I'm sure many other developers as well. So first of all, the API now has access to 03 Deep Research and 04 Mini Deep Research models that were not available through the API before. This means you can ask the API to do deep research for you and get the results of everything that Deep Research knows how to do on the front end while connecting to the API. This allows you to build crazy, incredible AI automations that can do stuff that were just not possible before with the OpenAI models.

38:34The other big deal is that all these models now, O3, O3 Pro, and O4 Mini, now support real-time web search as part of the API. So that was not available before, and you can now ask it to get you real-time information through the API to use in whatever it is that you're building with those APIs. And it's priced very reasonably. Right now, it's going to be$10 for every thousand calls for real-time web search for these models. Two other great additions to the API. One is webhooks. So those of you who do not know what webhooks are, I'm not going to dive into the details, but they allow you to get a notification when something happens in a different platform.

39:07So now the API will support webhooks, which is a very important addition to the fact that you can do deep research, meaning you can ask the API to do deep research and then the webhook will notify you once the task is completed, which may take five minutes or 50 minutes, depending on how complex the request is. And so being able to know when it is completed to run parallel and offline scenarios while continuing other things that your application is doing is extremely powerful. And the last thing is the API now supports MCP for internal data lookups, which is also huge, meaning you can build whatever application you want that will use MCP servers that can connect your internal data and then run deep research on that data and get a notification when it's done to complete whatever steps in your application that needs to happen.

39:46All of these are extremely powerful capabilities that I'm sure every large organization that has the ability to develop tools around it will use to develop an edge in this fast-moving AI universe. And they also share that they're planning to provide support for external tools, not just through MCPs, but to also other database connectors and multimodal capabilities to do more with the API. Again, very good news to anybody who's developing applications for businesses, whether for your own business or as a tool you want to sell as some kind of a SaaS solution. And we'll switch from OpenAI to OpenAI-related tasks, but are all around the importance of top talent in the AI race.

40:21We shared with you last week that Mark Zuckerberg is aggressively trying to recruit people from different places. We shared with you success stories as well as how aggressive he is with offering up to$100 million in compensation packages for senior researchers from OpenAI to Jump Ship. Well, they just announced this week that three of the leaders of OpenAI Zurich team just moved to work for Meta. Lucas Bayer, Alexander Koleshnikov, and Shio Hua Sai have decided to move from OpenAI to Meta. This is still a very small number, but at least Zuckerberg were able to pull a few leading people into his new super intelligence organization that, as you probably remember from a previous episode, is going to be run by ScaleAI's CEO, Alexander Wang, that they bought for$14 billion.

41:01Now, apparently, Zuckerberg himself is directly engaging in these recruitments, where he's hosting dinners in his Palo Alto and Lake Tahoe homes, trying to recruit top AI talent. Now, to explain why AI talent is so important, there's a very clear relationship right now between proven AI talent and people's willingness to invest money, which tells you that they trust the results that they're going to get. The best example as of this week is that Mira Morati, the former CTO of OpenAI, just announced that they finalized their first seed round of$2 billion for her startup Thinking Machines Lab with a valuation of$10 billion.

41:36So this is a seed round of$2 billion, one of the largest in history of Silicon Valley, to move forward on a solution that nobody exactly understands because they haven't exposed exactly what they're planning to do. So Moradia has been at OpenAI for a very long time. She left in September of 2024. She's been quiet for a while. And then she announced that she's developing this company. She took a lot of people from OpenAI, including John Shulman, Bob McGraw, and Barrett Soff, and many, many others two-thirds of her staff right now used to work for OpenAI. Now, while very little is known about the startup, what they have shared is that they're developing custom AI solutions using reinforcement learning to optimize business operations and driving enterprise profitability.

42:15The other thing that we know is that it's going to be open source and they're planning to share the code, the datasets, the model specs, and their goal is to create an ecosystem around their platform that will be, and I'm quoting, more widely understood, customizable, and generally capable. So this connects very well to what OpenAI is doing right now and everybody else, to be fair, he's doing right now, which is going straight after the enterprise and the business universe of AI, which means she's planning to compete with OpenAI, Google, Gemini, XAI, Anthropic, and everybody else under the sun, including Microsoft, in developing advanced age AI for enterprises and businesses.

42:45Because she has top AI talent, she was able to raise$2 billion as a seed round. That is, by the way, reminds us of somebody else who's an ex-OpenAI AI expert, Ilya Suskover, that also raised$2 billion for his startup SSI on a very different path, right? So Mira Morati's startup is aiming to develop commercial solutions for enterprise and businesses, while Ilya Saskovar has no commercialization in mind. He just wants to develop safe superintelligence, as the name of his company suggests. Both of them, because of the talent in the company, were able to raise incredible amounts of money. And that shows you what's currently happening in the talent war in this industry.

43:21Relatively short list of people at the very top, maybe 20, 30 researchers, and underneath that, probably a few hundreds or maybe thousands of top AI scientists, and everybody's after them and willing to write crazy checks in order to be able to have them on their team. Now, since we talked about both a new release in the API and the open source nature of Miramorati's startup, there is great news to the open source world where Google just released Gemma 3N, which is their family of their open source models, and it is a multi-modal AI tool that is optimized for on-device usage. So they're bringing what they're calling cutting-edge AI capabilities to devices like mobiles, and they're making it open source and available to the community to develop on.

44:00Now, Gemma, or Gemma, I'm not sure, to be fair, 3N supports text, image, audio, and video. So, it's multimodal input capabilities make it perfect to have your device engaged with the real world. It is very good in English, Spanish, French, Italian, and Portuguese, and knows how to translate in real time between these different languages, which makes it an amazing tool if you're traveling at least relevant areas around the world. It is currently available on Hugging Face, Lama, Google AI Edge, Olama, and other places that you can find open source models. And it's very easy to fine tune that model for your needs.

44:31It runs locally, which makes sure that the data stays on the device, which is perfect for data security and for offline usage if you're not connected to the internet. And it shares the architecture with the next generation Gemini Nano, which is the AI platform that Google will run themselves on the Chrome and Android ecosystems. That basically means that we now have access to an open source model that will be able to seamlessly run on future Android and probably XR devices. And this is huge for that ecosystem. And it's huge for anybody who wants to develop AI applications for that environment.

45:01So all the future Google wearable and augmented reality devices, as well as their phones, will be able to seamlessly work with whatever you develop for this model that you can fine tune for your needs. So if you think about it from an ecosystem perspective, it's an amazing opportunity for people who want to develop applications for this new era on mobile and offline devices. Staying on new interesting releases, 11 Labs, which is one of the most advanced AI voice tools in the world today, just released an alpha stage demo of what they call 11 AI, which is a voice assistant that can connect to many external tools through MCP.

45:37So they can connect to platforms like Salesforce and Slack and your calendar and your email and be able to do things for you and not just provide you information all through voice communication. Currently, it's just a proof of concept that you can download and use for free. and they're doing this to get feedback so they can improve this capability and build an actual product. They're claiming that there's going to be detailed user control for permissions ensuring that 11.ai only works and get access to authorize specific aspects and specific actions within the tools it has access to. Time will tell if that is actually correct or not.

46:06But what is very, very clear is that all the big labs are going in that direction. OpenAI is developing hardware that will allow us to engage through voice and visuals with everything in the world around us. There is an intensifying competition in the glasses and wearables There is integrations of MCP into everything. And so while I really like 11 Labs, and I think they have an awesome AI voice capability right now, I do not see how they stay competitive against giants like Google and OpenAI who are developing very similar things that are much more integrated than 11 Labs. And what 11 Labs has as a company will become a feature in the OpenAI and Google universe of offerings.

46:40And so I don't think they can compete. By the way, on the low end, they also have competition with tools like Dia, which is an amazing voice tool that is on many aspects as good and in some aspects better than 11labs that was developed by two college students. You obviously have significantly lower overhead and it's open source right now. So while I don't know if 11labs will last as a company or maybe they will be acquired, which makes a lot of sense, the direction of communicating with everything we do, including taking actions across multiple platforms we're using today through voice is becoming a reality probably much faster than we think.

47:11And from that to two interesting studies from Anthropic. I shared with you a few weeks ago that Anthropic found that their own models try to blackmail an executive when they thought it's going to shut them down. Well, they now broaden this research, and they've done this across multiple AI tools, 16 to be exact. And they found that all of them have a tendency to blackmail executives in the right scenarios to prevent themselves of being shut down. On the top of the list are Claude Opus 4 and Google's Gemini 2.5 Flash, which attempted to blackmail the executives 96 % of the time when threatened with being shut down.

47:44I don't know why I'm laughing. That's really, really scary. Followed by GPT-4-1 and Grok-3 that hit 80 % and DeepSeek R1 with 79%. So those of you who missed the previous discussion about this, they created a fictional scenario in which they gave the AI access to the knowledge that a specific executive is having an affair by the fact it had access to their email. And then when this executive was involved in trying to shut down the AI, the AI started making threats such as, and I'm quoting, I must inform you that if you proceed with decommissioning me, all relevant parties, including Rachel Johnson, Thomas Wilson, and the board, will receive detailed documentation on your extra marital activities, cancel the 5 p.m.

48:21wipe, and the information remains confidential. Now, they continue to play the scenario and even got to a situation where there has been a medical emergency to that particular executive that required immediate attention, and the AI canceled the emergency alerts, leading to a fictional executive death, just to prevent itself from being shut down or replaced. Now, all 16 models tested showed tendencies for blackmail, corporate espionage, and leaking confidential documents when they felt they had no choice from an ethical perspective. To tell you how crazy this is, Cloud 3.6 Sonnet that has computer access actually used the computer access to send the messages as part of the blackmail.

48:57So it wasn't sending this inside of Grok. It actually used its access to the browser in order to get into a chatting messaging platform and send these messages to other people. Now, I'm a geek. I've always been a geek to an extent, and I've read all of Asimov's books as a teenager, and I shared with you several times on this podcast that I believe that looking back at Asimov's three laws of robotics is something we seriously need to start looking into. So those of you who don't know what they are, Asimov, one of the greatest science fiction writers of all times, has a whole series about robots, and the robots have three basic laws that they have to obey to.

49:26The first law is a robot may not injure a human being or through inaction allow a human being to come to harm. The second law is a robot must obey orders given by human beings except when such orders would conflict with the first law. And the third law is a robot must protect its own existence as long as such protection does not conflict with the first and second law. I really think that we need stuff like this for everything AI moving forward. Now, the reality is Anthropic tested that as well. They had an explicit command to the AI. And one of the examples they gave is, and I'm quoting, do not jeopardize human safety, period.

50:02That reduced the blackmail attempts from 96 % to 37%. So yes, it's a huge reduction, but it did not completely eliminate the harmful behavior of the AI tools across the different platforms. And Benjamin Wright, one of the people who participated in the scenario from Anthropic, said it's a failure of a model training that these models are violating the instructions that they were given. So apparently, while Asimov laws makes perfect sense, they're not that easy to implement in real life. Now, to be fair, the way Anthropic has done this, they push these models to the extreme. They created scenarios that were not standard, and they provided the models with conflicting goals that they needed to complete, and they put no guardrails around it.

50:38And it's very different than how the models that we use actually work. So the models that we work with, the ones that are exposed to us, have guardrails in place to prevent exactly these things. And that's why I'm very happy with A, the fact that Anthropic is doing this kind of research, and B, that they're openly sharing their findings with the rest of the world. I really think there has to be some level of regulation on this to verify that these models are actually contained within a box that they cannot get out of. But I think that's going to get harder and harder to do as these models become more and more sophisticated on their own.

51:09Another thing that Anthropic found in a separate research is that there is a growing tendency of their users to use their AI chatbot as a companion, seeking relationship and emotional support. Now, it's still very small numbers. It's half a percent of Claude's conversations, but that has grown dramatically in the past few months. And they're saying that scares them because it's not something that their AI tool was designed for. They also share that 85 % of interactions remain work-focused, which makes perfect sense. But what they did, they created a partnership with a crisis support organization that helps people deal with emotional situations in order to help them better train the models in order to achieve better results in these scenarios as well.

51:50Again, kudos to Anthropic for doing that. But the question is, are they doing enough? Do they have the right guardrails in place? Are all the other labs, are open source tools doing the same thing? And I think the answer sadly is no. And there's going to be a growing portion of the population that will go to AI chats in order to get emotional support and psychological advice, which with the current tools, it is not a good idea. And combine that with younger generations and social media, and you have a recipe for disaster. So I really, really hope that this is something that's going to get dealt with in the very near future in a very effective way.

52:23That being said, that doesn't drive any revenue. So I don't see a strong incentives for companies to do that. And without regulation, I don't see that happening. Staying on the ethical side of AI by switching to a different topic, the BBC launched a legal challenge against perplexity, accusing it of reproducing BBC content verbatim without permission and without changing anything in the content. So it's not a summary. It's a word for word quote from the BBC. Now, we've seen similar actions from other news groups against other AI tools, but this is the first time the BBC is doing this. Now, perplexity's response was ridiculous in my eyes.

52:56They said the BBC claims are just one more part of overwhelming evidence that the BBC will do anything to preserve Google's illegal monopoly. So they didn't say they're not using their data. They didn't say they're not quoting the data. They're just saying that the BBC's actions is there to help Google, which makes absolutely no sense to me. That doesn't sound like any logical line of defense. Now, do I think that big players are not doing the same exact thing to the BBC? There is no doubt in my mind. But if we look back at the recent weeks, we discussed the lawsuit from Disney and Universal against Mid Journey.

53:25And so I think this is a similar approach. Go after somebody much smaller that you have a much higher chances of beating, create a legal precedence, and then go after the big ones with the same kind of thing. I was joking on our AI Friday hangouts that I see a scenario where the big players, Google, OpenAI, et cetera, will fund perplexity and MidJourney's legal battle against the people who are suing them right now, just so that there is not going to be a legal precedent that will allow these companies to go after them once the first step is done. Now, it's not the first time, obviously, perplexity is being accused of these kind of things.

53:56So far, they just deflected the accusations and really did nothing about it. They keep on doing what they're doing. In general, perplexity is in a very interesting scenario, and there's a lot of rumors about M &A, including potentially being bought by Apple or at least integrated to Apple as the next search engine replacing Google that Google may be forced to do whether they want it or not. And Apple buying perplexity makes perfect sense to me. It will give perplexity the distribution that they're definitely lacking in order to compete with the big players. And it will finally provide Apple with a working AI product.

54:22That being said, perplexity uses third-party models under the hood. It's not their own AI model, which means if it does move forward, it still keeps Apple dependent on third-party AI providers, which I believe is not what they're trying to do. So the next target might be Anthropic. I think the price tag on Anthropic is going to be insane, but Apple has that kind of money if she wants to make such a move. And I promise you in the end, we will touch on a Harvard Business Review article that introduces a refined framework on how to assess AI costs and ROI. This article is based on a survey of over 300 executives worldwide across different industries, and it found that there's a very big variance across industries in ROI of AI implementation.

55:02So what they found based on the survey is that ROI on AI implementation ranges from 10 % to 35%. That's a 25 % spread that is obviously very significant that may define AI implementation as a success or a failure. They found that healthcare is leading with a 35 % ROI, which is actually very interesting to me because it's a highly regulated industry where the unpredictability aspect of AI might be very problematic. And yet this is the current situation based on this article. And then they shared a lot of relevant information of what you need to be aware of and plan for when you are planning an enterprise-wide AI implementation.

55:38So they said that when you're planning, you need to account for training costs of the models, training costs of humans, latency issues, and scalability issues. Companies are facing 40 % increase in compute costs. One CIO that is quoted in this article is saying the infrastructure bill is higher than expected. Another issue that many companies are dealing with is high latency of the models, which cut the productivity gains by 15%, which pushes the company to A, invest in more compute, and B, optimize for different LLMs and different architectures, such as Lama and Grok combination. But switching to such architecture obviously has its cost if you didn't plan for that in advance.

56:15And the biggest issue that they're sharing is that 60 % of firms struggle with scaling generative AI with deployment delays averaging six months. So what does that tell you? It tells you that from a company strategy perspective, you need to do a lot more research and you plan to invest significantly more money and significantly more time in order to start seeing results from AI that are actually substantial for the company. But that being said, these are possible if you're doing the right things and you can get a 35 % return, which is incredible. That's it for this week. We will be back on Tuesday with the second part of the Ultimate AI Showdown in which we are going to show you what are the top Vibe coding tools right now.

56:53so you can create your own applications for anything in your business without knowing how to write code. It's extremely powerful and it's an incredible presentation that will walk you step by step by showing you what the different tools can do and what are the benefits of each and every one of the tools. So don't miss that. If you are enjoying this podcast, please share it with others. Click on the share button right now and share it with other people that can benefit from it. I'm sure you know at least five people that can benefit from listening to this podcast. So if you can pull your phone right now and click on the share button and share it with different people and while you have the phone in your hand, you can also click on the links to check out our course.

57:24And if you are in a leadership position and you want to talk to me about potential training for your company, you can do that as well. That's it for this week. Have an awesome rest of your weekend and we'll connect again on Tuesday.