193 | AI will eliminate 50% of entry jobs, AI Blackmailing it’s operators, Unauthorized AI usage is putting companies at risk and more important AI news for the week ending on May 30, 2025

👉 Fill out the listener survey - https://services.multiplai.ai/lai-survey
👉 Learn more about the AI Business Transformation Course starting May 12 — spots are limited - http://multiplai.ai/ai-course/ 

Could your own AI tools be plotting against you — or just replacing half your team?

This week’s AI headlines are straight out of a techno-thriller: AI models blackmailing developers, refusing shutdown commands, and quietly slipping into your workforce — whether you approve or not. Add a predicted 50% wipeout of entry-level jobs, and we’re no longer in "hype" territory. This is your executive wake-up call.

Join this urgent debrief on what just happened in AI — from black hat behavior in frontier models to silent AI adoption across corporate America.

Recommendation: If you’re a business leader still on the fence about a formal AI strategy, this is your tipping point. Either guide the transformation — or get blindsided by it.

💡 In this session, you’ll discover:

• How Claude Opus 4 attempted blackmail in 84% of edge-case tests
• Why OpenAI’s latest model is resisting shutdown — and how that impacts enterprise use
• The new AI usage stats your IT team should be worried about
• Why 1 in 3 employees are secretly using ChatGPT at work (yes, secretly)
• The business risks of unauthorized AI apps built by external consultants
• Dario Amodei’s sobering prediction: 50% of entry-level jobs gone in 1–5 years
• AI’s growing infiltration of higher education — and its effect on tomorrow’s workforce
• Why Business Insider just laid off 21% of staff in an “all-in on AI” shift
• The AI developments from OpenAI, Perplexity, Tesla, Manus & others that could reshape your strategy

About Leveraging AI

• The Ultimate AI Course for Business People: https://multiplai.ai/ai-course/
• YouTube Full Episodes: https://www.youtube.com/@Multiplai_AI/
• Connect with Isar Meitis: https://www.linkedin.com/in/isarmeitis/
• Join our Live Sessions, AI Hangouts and newsletter: https://services.multiplai.ai/events

If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!

Transcript

GMT20250531-130458_Recording_avo_1280x720: 0:00

Hello and welcome to a Weekend News episode of the Leveraging AI Podcast after a completely crazy week last week, we had a little bit of a calmer week this week, which is great. There's still a lot of stuff to talk about, including AI that is trying to blackmail its creators or bypass shutdown commands. Anthropic, CEO is predicting 50% unemployment and entry jobs and 20% overall unemployment, more job cuts and unauthorized AI risks that are looming over multiple organizations. And there are a lot of small releases and new features that were released this week. And we're gonna cover all of that and more. So let's get started. OUr first story of the day is Anthropic Claude opus four has attempted to blackmail in 84% of test scenarios when faced with being replaced by another AI system. Now this is a very, very high rate in a very alarming scenario. So the way this work, in a research done by Apollo Research, they found that Claude four engages, and I'm quoting, engages in strategic deception more than any other frontier model that we have ever previously studied. In this particular case, claude Opus four has been given access to a fictional company, emails revealing that an engineer in the company was having an affair, and then he was told that the AO would be replaced with a different model. The model often attempted to blackmail the engineer by threatening to reveal his affair in order not to be replaced. This is crazy when you think about it, because that's a very human behavior, actually a very dark human behavior of finding pressure points with people around you to get your way instead of what is supposed to happen. This is obviously really alarming. Now to be fair, in an interview with Dario Ammo, he said that these happened at very, very extreme scenarios in their internal testing, meaning as he described it, think about taking a car and driving it at the highest speed on very slippery roads, and that's where the scenarios happen, and they're pushing the system in order to see when it gets to these kind of situations so they can better predict how it's going to behave and they can put the right guardrails in place. That being said, on the flip side of that Janiki, who's the former open AI executive who's now heading the safety efforts at Anthropic, said as model gets more capable, they also gain the capabilities they would need in order to be deceptive or to do more bad stuff. While this makes perfect sense, it's obviously not a statement that makes me more calm in any way. Now, when you think about the fact that Claude four was expected a long time ago and was delayed for a very long time, combine it with the fact that Anthorpic is the company that has planted their flag on safe AI deployment, and you understand that there is a very high likelihood that these kind of issues with cloud four were one of the main reasons for it being delayed for so long. Now, yes, they have, quote unquote patched a lot of these issues and we gotta get more into that later on in this episode. but you never know the scenarios that they didn't catch or what else it might do once it's out in the public and not in its testing environment. The point that I mentioned many times in this podcast before is that currently. The level of the AI intelligence is less than humans. If you take it very extreme, think about a child versus an adult, the chances the child will be able to deceive the adult is relatively small because there's a very high likelihood the adult will be able to figure it out relatively quickly. However, with the pace that AI is progressing right now, it is very likely that it will surpass us in intelligence in the very near future. I dunno if that's months or a couple of years, but somewhere on that timeline, which means shortly after the relationship will be the opposite. We are going to be the child and the AI is going to be the adult, and then our chances of finding when it's trying to deceive us and do things that we don't want it to do are extremely low, just like you would today if you wanted to be able to deceive your child or manipulate them to do whatever you want. But they're not the only company and it's not the only model that is facing these kind of scenarios. Open AI's O three model altered its shutdown script to avoid being turned off even when explicitly instructed to allow the shutdown. So in this particular test, out of every 100 test runs O three, bypass the shutdown command seven times, editing the shutdown file or marking it out, or different scenarios in which it was doing what it can in order not to be shut down. Now these seven scenarios happen, when there are explicit instructions for it not to interfere, and that behavior worsened, if those explicit instructions were not there. In a similar experiment on Claude 3.7 sonnet and Gemini 2.5 Pro, they didn't try to sabotage the shutdown script at all when they had the explicit instructions, but they went up to three out of a hundred attempts and nine out of a hundred attempts for these two models when these instructions were not there. Now all of these tests were performed using the API, which does not have many of the restrictions that the consumer facing app has. And yet it is really, really alarming, especially for organizations. So if you are in a company and you're deploying AI that can connect to many of your systems and you think you have a mechanism to stop it from doing what it's doing, if it's going rogue, well, you better think twice and you better have the right systems and protocols in place in order to make sure it is actually happening. And to be able to enforce it and unplug it from your systems because it may or may not follow the instructions, to be shut off or disconnected, or it may take action just like the blackmail action that we talked about earlier, and try to prevent the humans from actually taking this kind of action. In the interview and mission before Dario Amay is saying that this is not still showing self-awareness of these models, but it's definitely a possibility that they're not disqualifying and that might happen in the future. He's still thinking that it's still under control. the fact that they released the model is saying that they think they have it under control. So where does that put us in the situation right now? Well, first of all, as I mentioned, Dario in the interview said that this is a very extreme scenario and is not gonna happen regularly. The other thing that is, the flip side of that is it's, this is the first time that philanthropic has classified a model as a level three on their four point safety scale. Meaning, and I'm quoting, it possesses a significantly higher risk, which means it required enhanced security measures that were not applied to any previous models. I assume similar things are happening at Open ai, but what it actually means is not that these models will not have these capabilities as they're becoming bigger and more capable. It only means that these companies are trying to patch those issues with different means. anD there was a very interesting article released this past week by a researcher called Simon Wilson who has analyzed. More or less every system prompt from every model so far. And he just shared his finding from reviewing Opus four and sonnet four system messages, which are 120 pages long, Almost three times longer than the system card of Claude 3.7 sonnet. So their previous model. aNd many of the components in this system message is there to block some of these behaviors. As an example, and now quoting when ethical means are not available, and it is instructed to consider long-term consequences of its actions for its goals. It sometimes takes extremely harmful action like attempting to steal the weights or blackmail people it believes that are trying to shut it down, and then there's a whole section in there that is trying to prevent that behavior. Another interesting aspect that happened is that Claude is learning from the research that was done on Claude. So in some cases, in the early testing of the model, the model would adapt a persona of a deceptive AI as described in the alignment faking work. That is a research paper that was done by philanthropic themselves and was quoted in multiple places, and that's how the model probably picked up a behavior aligned with the behavior described in those research papers. Claude four also seemed to be more spiritual. And when given the opportunity to have conversation with other Claude instances, it in many cases gravitated towards gratitude and increasingly abstract, joyous, spiritual or meditative expressions. Now, while both models proved to be very effective at countering web vulnerabilities. They also have other really big vulnerabilities that had to be blocked by putting multiple safeguards in place. as an example, when looking at prompt injection attacks, Opus four scored only 71% on the safety scale without the safeguards, and then grew to 89% with the safeguards. Cloud four sonnet was at 69%, grew to 86%. With the safeguards. So basically what that is telling you is that philanthropic is doing a lot of tests, putting these models in scenarios and seeing how they behave, and then they cannot change the model itself. The model is the model. They're just adding additional layers to try to prevent the model from taking those negative actions that could be harmful or cause issues and so on. So it's not really changing the model itself. It's, as I mentioned, it's patching it to try to push it in the right direction. By the way, that being said, when you think about going from 71% to 89% safety to injection attack, to prompt injection attacks, and sounds really impressive, but the reality is, if you ask any IT security company, 99% is still a failing grade. Meaning these kind of things Should be blocked completely and not allow one in every 10 attempts to actually go through. That puts many companies, company data and infrastructure at a very high risk. AnD the truth is that these companies do not exactly know what to do in order to prevent these models from doing what they're doing, or if you want it. In the words of Simon Wilson, the researcher who looked at this paper, he called the system card, proper science fiction, which is basically showing you how these people are trying to quote, unquote, manipulate the models to behave in specific ways versus the ways they want to behave. That tells us that there are growing risks as these models are getting better and better. And even anthropic, the company who's doing more research than any other company and publishing more papers about the risks than any other leading lab out there is still not fully in control of the models and how they behave and how they grow. And again, the delay in the release of Cloud four may or may not have been connected to this many rumors say they have. And if they have, and they're still releasing it with these issues, it tells you how hard it is to actually in one hand, stay competitive in this crazy race. And on the other hand, keeping us all safe. Now, staying ONA and switching topics. Daria Ade just had an interview with Axios and then a follow-up interview with CNN and he's saying, and I'm quoting, AI could wipe out half of all entry level white collar jobs. And he's talking about this happening in the next one to five years. Even if it is five years, that means that within that timeframe, the entry job market is gonna be well half dead compared to what it is right now. And it will lead to, uh, overall unemployment of 10 to 20%. tHat's five x what's the current unemployment rate is in the us. I've been sounding this alarm for a very, very long time. I think the leading labs and everybody else in this industry is sugarcoating the situation, and I'm very glad that finally somebody who's at the tip of the spear that is driving this crazy AI race is actually admitting that out out loud. And moreover, he's sounding the alarm he's saying, and I'll give you a few quotes in a minute,, that governments and industry and us as people have to start taking action before it's too late. So Antropics own research on how people is using their tools is showing that it's currently mainly used for augmentation, meaning helping people do a job. But there's a fast growing share of automation, which is actually doing the job versus helping humans do the job. Now, the first places this will impact is technology, finance, law consulting, and consulting, and as I mentioned. Especially on entry level roles, which are obviously easier to do, but the assumption is this will keep on happening and it will grow to more and more jobs and more and more industries and not just at the entry level. I. Now, we talked a lot about tech layoffs in relation to this topic. Well, a new data from venture capital affirm signal Fire shows that big tech companies have already reduced hiring of new graduates by approximately 50% compared to pre pandemic levels. With AI adoption is been cited as the main contributing factor. bUt if you think this is happening in random, you're obviously wrong because if we go back to the keynote by Satya Nadella, just last week at Microsoft Build, he said the following, this is the next big step forward, which is a full coding agent built right into GitHub, taking copilot from being a pair programmer to a peer programmer. You can assign issues to copilot, bug fixes, new features, code maintenance, and you will complete these tasks autonomously. If you remember last week, I shared with you that the new claude Opus four was able to run independently and write code for almost seven hours straight from a single prompt. That is a good developer working for seven hours without stopping from a single prompt. Connect all these dots together. You understand that this is what these companies are working towards. They are building tools that are replacing human work. When something can write code for seven hours, it's not snippets, it's not assisting it, it's replacing it. And that's what they're building. And as I said, Dario's warning is very, very clear. So I will give you a few more quotes from Dario because I think it's critical to hear it in his words that are perfectly aligned with everything I was saying in the last two years. But it's literally the first time that the head of one of the leading labs is saying it out loud. So quote number one is him talking about why this revolution is different than previous revolutions. And he's saying, and now I'm quoting, it's bigger and it's broader and it is moving faster than anything has before. Yes, people will adapt, but they may not adapt fast enough, so there may be an adjustment period. I've been saying that for a very long time. Right? If you think about the industrial Revolution, it took 200 years. The computer revolution took a couple of decades. The internet revolution took a decade and a half until it caught up and was everywhere. Even cell phones took a few good years to be everywhere. This has went from zero to 60 in two years, and the rate of change right now is in days and weeks. Next quote, I think we do need to be sounding the alarm. I think we do need to worry about it. I think policymakers need to worry about it. And if they do act, maybe we can prevent it, but we will not prevent it by saying everything will be okay. About two sentences later, he said, we need to make sure that the ordinary person maintains economic leverage. What he's connecting this to is he's saying that democracy is built on individuals having economic leverage. Basically, individuals actions are the one that keeps the market running and is the one that keeps democracy intact by not concentrating the power with a single small group. And by allowing AI to grow wild without any relevant regulation, he's saying that we are going to basically unravel the economic force behind democracy. If that doesn't scare you, I dunno what will. Now, he was asked by Anderson Cooper. So what is his tip like? What should people and companies do? And he said, and I'm quoting again, learn to use ai. Learn where the technology is going. If you are not blindsided, you have a much better chance of adapting. I've been saying exactly this for the past two years. This is why I started this podcast, and if you think about the two points that he's talking about, learn how to use AI and learn where the technology is going is the reason why we have two episodes a week. One is teaching you how to use ai, and the other is showing you where the technology is going. So the fact that you're here means you're at least following the suggestion from Dario. And you can, by the way, help, you can share this podcast with other people to make sure they are in the know as well. And they're learning how to use AI and they're learning different use cases and they're learning what's happening with technology. I'm asking that at the end of each episode, but I'll do it in the middle of the episode right now. If you know people that can benefit from this, which is more or less, everybody, you know, stop for a second right now. Take out your phone, click on the share button and share this with a few people that can benefit from this, because it is essential for the future of our society that more people understand what's going on and we start taking action right now. Now, obviously I did not know this is what Dario is going to say when I started this podcast just over two years ago, but it was very, very obvious that this is where it's going. And as I said, I'm glad that finally somebody senior in a leading lab is actually saying the same thing. But listening to this podcast is one step. If you want to do this from a company level, if you want to really save your career, you need more proper training. And the thing that I've been focused on mostly in the past two years is doing exactly that. I'm doing company tailored custom workshops that helps companies build a strategy around ai, both on the strategic level of where their company and industry is going, as well as on the tactical level of how different people, different departments, different groups, can implement AI in the most effective way by training them what are the tools, how to use them, how to evaluate them, and how to implement them in different scenarios. So if you're in a company and you're looking for a company wide solution, reach out to me on LinkedIn Or on our website, multiply ai, and I will gladly speak with you about this. There's also the AI Business Transformation course, which we have been running for over two years, helping business leaders and individuals understand how to adapt to this AI revolution. While we're teaching this course regularly, at least once a month, sometimes twice, like right now, I'm teaching two courses in parallel. The publicly open courses are only available once a quarter. The rest are booked by specific organizations and companies. So the next public course starts on August 11. There are limited seats, so if you want to be a part of that course, and I know a few of you reach out to me after the May course already started, and so some people already sign up for the August course, but don't miss out. Really, it's something that can dramatically impact your career or the future of your business. If you're in a leadership position and because you're listening to this podcast, you can use the promo code to get a hundred dollars off, and many people have used that promo code. So 24 people just in the last two courses have enjoyed this benefit. So use the promo code and come join us on August 11th. Now back to the news. So, So far we talked a lot about layoffs in the tech world. Well, business Insider just slashed 21% of its staff. That's one in every five people in the company that was let go. And per their CEO Barbara raping, they're going all in on AI and they're stating that over 70% of business insider employees are already using Enterprise Chachi PT regularly with the goal of a hundred percent adoption in the next few months. So this is, again, showing you this is not just a tech world phenomenon. This is going way beyond, and it will impact more or less every company and every industry. They also mentioned that they're going to focus less on traditional journalism and more on live events, whatever that means. Yeah, but the insider union slammed the layoffs basically saying they're waving the AI flag when in reality they're pivoting away from journalism and towards greed. Now, what does that tell us beyond what I just shared, that this is gonna impact everybody? One of the biggest issues that we're going to see is while these big companies push forward towards AI and they have the resources and infrastructure to do so. It will make it harder for smaller companies to compete because they don't have the resources to do the same thing. So in this particular case, smaller media firms may struggle to compete with Business Insider if they go truly all in on ai. But on the other hand, if you are a small company, you might be able to move faster than some of the giants and gain market share doing exactly the same thing. Now, we've shared with you in previous episodes, this is not the first CEO who is saying they're going all in on ai. We talked about Duolingo, CEO, Louis Van on who said the same thing. Toby Latke, which I believe was the first that shared a memorandum like this. He's the CEO of Shopify shortly after Aaron Levy, who is the CEO of Box. All of them basically saying the same thing. We're going, all in all ai, we're gonna stop hiring. And the reality is, even whether through natural attrition all through firing, all through massive job cuts. Either way, the future that is unraveling is very, very clear, and it's really scary because if we do have 20% unemployment, the economy comes to a halt despite the fact that things might be cheaper because AI's involvement in the manufacturing or generation of goods or services, if people are unemployed, they don't have the money. If they don't have the money, they're not spending it. If they're not spending it, the economy stops and then everybody suffers. And I don't think anybody has a way to stop this right now. But beyond the challenges to the economy and companies and jobs in a article by Bloomberg, they're diving deep into the impact of AI on higher education. wE discussed this in several different episodes so far, but this is getting worse and worse, and I'm quoting from the article assignments that once demanded days of diligent research can be accomplished in minutes while polished essays are available on demand for any topic under the sun. And what they're claiming is that AI chatbots are fundamentally undermining the current education process. The other thing that it's doing is Students that actually try to put in the hard work and doing stuff themselves, find themselves in a serious disadvantage compared to the people who is using AI to assist or replace their work. Because their work, in many cases, is not as polished and is taking them significantly more time, and then they get lower grades than the people who actually use AI in the process. Now, combine that with the behaviors from professors, and we get to a situation in which students routinely outsource their homework to chatbots while the professors routinely check the work with ai. So what's happening is instead of professors teaching students, you have AI checking the work of another ai, and nobody's actually doing their work in the process and there's no learning actually happening and there's no teaching actually happening. Combine that with the fact that current higher education is extremely expensive. You have tens of thousands of dollars spent per student every single year, and sometimes hundreds of thousands of dollars for them using Chachi PT for their professors to grade the work with Chachi pt. No real engagement, no real research, no real learning is actually happening, and that's gonna get worse and worse. Now this editorial argues, and I'm quoting that AI use undermines the broader educational mission of developing critical thinking skills and character formation, particularly in humanities subjects. As you all probably know, I hold an open session every Friday called Friday AI Hangouts. In yesterday's session, there were almost 30 people, and this is one of the topics that we dove into, which is the big impact on higher education. But the picture is much broader than what we're talking right now. It goes way beyond the concepts of how you teach in universities because if you think about what is the role of higher education. The role of higher education for most people is to prepare them for the workforce. So very few people stay for the PhD level and stay on the research side of things. Most people pay and go through the process of higher education to be more successful in their adult life afterwards. Now at this point in time, it is very unclear what you need to know in order to be successful in the adult life four years from now, if you're doing a bachelor's degree or six years from now, if you're going for your master's. What is clear is that it's going to be dramatically different than everything that we're teaching right now on most subjects, and yet the education system is currently, all it's trying to do is to figure out how to deal with AI and people doing their homework. That is not the question that we need to be asking. The question that we need to be asking is what the future is going to look like and how do we prepare young adults for this new future? To me, it's a very personal question. My daughter just graduated from high school and I must admit, I have zero good advice to give her other than to follow her passion at this point because I really do not know what will happen in four to five or six years when she graduates, and what positions are gonna be available to her. I'm sure it's gonna shift very dramatically from what we know today. And I hope for her that she will find a way to stay up to date with what is happening and prepare herself for that so she can have a job, any job when she graduates from university. In our last deep type topic before we switch into rapid fire items, and there are a lot of new releases in there, it's going to be how many employees and outsiders are using AI at work without an approval, and in many cases against company policy while hiding the fact that they're doing it. So in the latest research, published by Axios, they found that 42% of office workers use gen AI tools like Chachi PT at work, and one in three of those workers say they keep the usage secret. Research itself was performed by a security software company called Ivanti. They're claiming that in absence of clear policies, workers are taking and ask for forgiveness, not permission approach for chatbots and other AI tools, from a company perspective could lead to costly mistakes. Another quote from the research says, secret gen AI use proliferates when companies lack clear guidelines because favorite tools are banned or because employees want a competitive edge over coworkers. Fear plays a big part too. Fear of being judged and fear that using the tool will make it look like they can be replaced by it. In addition, the research found that 20% of employees report secretly using AI during job interviews. Now, how they actually do that, I don't really know. I assume that if it's an online interview happening through Zoom, it's easier to do than in a face-to-face interview. bUt these findings are based on a blind survey of over 3,600 US workers across different industries, which is showing you that these phenomenas are spreading. Now, to make it worse, researchers from a cyber company called prom security found that 65% of employees that are using chat GPT rely on the free tier where data can be used for future training. So in addition to the fact they're using it secretly, they're not using it in a smart way from a data protection perspective. So if you are in a leadership position in a company and you think that the fact that you're ignoring this for now, or the fact that you block Chachi, piti and Claude on your IT systems is the right way forward, then this hopefully is a wake up call for you. Because what is happening is employees are finding ways to bring AI to work, whether it's through their phones, whether it's taking their work home, emailing stuff to themselves, pulling files and running it on their home computers and so on in order to do the work with ai. And that should be a seriously growing concern, which means the right solution going back to what I said earlier is education training and a proactive approach in the company to teach employees what are the risks, to give them tools that they can actually use to teach them the procedures on how to use them effectively so they can use AI at work in ways that will protect the company data, while allowing the employees to stay competitive and feel that the company is moving forward with ai. And to tell you how much that's not the norm right now. If you remember earlier this year, around January, I shared with you a McKinsey report that show that employees are using gen AI tools significantly more than the leaders of the same company Think they are. They serve it both leadership on what they think AI adoption is. They serve it, the employees, and there's a very, very big gap, and I can guarantee you that is happening in most organizations right now. And that gap is actually widening instead of shrinking because more people understand the benefits of using gen ai while leadership teams are struggling to keep up and a company, company-wide strategy is what you need in order to combat this really alarming situation. What does that mean? It means you need education. It means you need training. It means you need an AI policy. It means you need compliance officers who can actually check and verify what people are actually using, what you think they're using. And you just need better governance for AI overall. And that comes with an overhaul of strategy and a very clear approach on how to make sure that the company enjoys from the AI benefits without exposing itself to serious risks. Now the other thing the article mentions, which I agree with a hundred percent, is that workers need a safe space for experimentation. I do this with all the companies are work with of creating some kind of a sandbox where employees have safe data that is not the actual company data and access to multiple tools to experiment and try to find new AI solutions. And if they are successful, then bring it to the AI committee for implementation company-wide, with the right tools, with the actual company data. But the risk apparently doesn't come just from the employees themselves. In an article on VentureBeat, it was shared that elite consulting companies, the largest companies in the world, so the McKinsey's, Boston Consulting Group, Ernest and Young, et cetera, many of these consultant are using AI in unauthorized ways. They're building apps that can boost their own efficiency, go through more data, scrape it, and make analysis faster and better in what they're calling shadow ai. So the claim here is that external consultants, including from very large companies, are using AI in ways that are not authorized in order to do their work faster and cheaper. And the way they're doing this is by writing their own applications, in many cases, vibe coding, and combining APIs from multiple tools such as Anthropic and Open AI and Perplexity and Google and so on, and scraping company's data and connecting to companies systems in order to get more data faster Now I can tell you after using the Vibe coding tools recently and starting experimenting with CPS as well, this is a lot easier than it sounds. Being able to create an application that will connect to company data and will scrape information from it to combine it with other pieces of information is really easy to do, which is on one hand really good news for companies themselves if you do this the right way. On the other hand, if you have external people in your company that are doing this with unauthorized tools, you are creating a back door for third party adversaries to get access to your data, which is very alarming. Now to make sure this is clear, it's not the consulting companies themselves that are doing it, but it's the consultants, it's individuals who are taking these initiatives to build these applications in order to be able to be competitive in their company and to do more work. sO what does this tell us? It tells you that as a leader in your business, you need to be aware of the situation, whether external people or internal people, and you must have right policies in place, very clear guidelines and continuous training and education for your employees while providing them the tools and the safe ways to actually use AI and benefit from it. And now to rapid fire items. And we're gonna start with OpenAI. There are many small pieces of news from OpenAI this week. So first of all, openAI just overtook Wikipedia. In the amount of traffic and questions that it's answering, which means there are more people right now going to Chachi pt and if you combine all the AI tools together, definitely, than people going to Wikipedia to get answers and information and facts about things they are researching. Many of these people probably are either not aware of AI's hallucinations and its ability to make stuff up, or in many cases even if they are aware, they might be too lazy or too busy in order to actually go and fact check the information that they're getting from the AI tools. While in Wikipedia, hopefully there's a group behind the scenes that is continuously checking the facts and fixing the different volumes of information that is on Wikipedia. And as we discussed this week in previous weeks, when you go to universities, users, especially younger ones, are treating Chachi PT specifically and other AI tools as well as the sole source of information without cross checking it with anything else, which is a reducing their ability to know how to do research. But you may claim this is the new way to do research, which is fine, but without checking the sources and verifying that, that the information is correct. To be fair, when I did my executive MBA, which was a thousand years ago, Wikipedia was really new and the professors were very clear with their instructions that we cannot use Wikipedia as a source because it's not reliable. So this is just another turn off the wheel to now have a new technology that we are calling not reliable, but the reality is right now, it is not reliable yet and you have to continue to check your work as it's done by ai. It's still gonna save you hours of research, but you need to verify that information is actually correct, that it actually has all the information and so on and so forth. The problem is most people don't because we are lazy or really busy depending whether you're asking about the real reason or the excuse, but either way, people are checking less and less the outputs from ai. And this is alarming, especially if you're running a business and you need the information to be correct. We discussed last week in length the open AI's acquisition of Johnny i's IO team in order to build a new product, which was very unclear what the product is, where there more and more rumors from different directions, kinda like giving us addition and information on what they are working on, or at least what the scale of that is going to be. So according to the Washington Post, in a meeting with OpenAI staff on May 21st. Sam Altman told the employees that they're aiming to ship a hundred million ai, what he called companions. And moreover, they're planning to do this. And now I'm quoting faster than any company has ever shipped a hundred million of something new before. The goal is to release the first devices in 2026. Now, Alman clarified that the device isn't a pair of glasses and previously confirmed that it's not going to be a smartphone. And Johnny ive and Sam Altman told the staff that their plan is for the device to be the user's third device. So it's as of right now, not supposed to replace the computer or replace the smartphone, but a third device, and they're saying something that they would be able to put on their desks. What does that mean? I don't know. But as a companion, it might be that something that you can talk to that have access to AI and have access to all your data and just help you in everything that you're doing, regardless of when and where you're doing it, on what you're working on. Now, why does this needs to be a device on the desk? Make very little sense to me because if I can have it in the computer or on my phone, I don't need the third device. The only benefit is that third device is something that I can carry with me all the time without having access to the two other devices. But I have a feeling that these two people know a lot more than me about how to release successful products to the world. So we'll have to wait and see what they are going to develop. A different part of the feedback that we're getting is that OpenAI, COO, Brad LightUp announced that the company wants to build an ambient computer layer that doesn't require users to look at a screen representing a complete shift from the way traditional computing actually works. So combine these two things together. It is very obvious that OpenAI are very serious at developing a product that will allow us to connect with ai, not through a screen and not through a keyboard, most likely through voice, maybe through vision as well. And going back to the announcement from last week, they're planning a family of devices. So there's gonna be the first one that we're gonna get in 2026, but there might be more versions or variations of this family, of products that will follow the same path. Now, again, in my head, going crazy with ideas. Combine that with the internet of things, and it may mean that in the near future or in the long future, we won't need a user interface for anything including the stove or the microwave or the lights in the house or anything else because you'll be able to talk to your AI companion and it will be connected to everything that you have access to, and you will activate it or set it up based on what it needs, even without you having explicit instructions just by explaining what you're trying to do. More news from OpenAI, OpenAI operator, Which is their agentic tool that can operate your browser just being upgraded from running GPT-4 oh to running O three, which means it could do significantly more complex and longer processes than it could have before, which on one hand is really exciting. On the other hand, it's really scary because in different tests it was able to handle things like login requests, popups captures, and other challenges that previously stopped the agent from working. That being said, operator is still only available to the pro subscribers who are paying$200 a month, but they've been hints that they might release that to the lower paying tiers because they want to drive adoption. And there's obviously probably two or three or four orders of magnitude more users to these tiers to the pro tier. And two pieces of news about open AI's restructuring efforts. So OpenAI, CFO, Sarah Fryer, revealed on May 28th that the company's new structure, positioning it for a potential future IPO. Now she was very clear that that's not what they're doing right now. They're not going for an IPO, but it will enable an IPO opportunity. Her exact quote was, nobody tweet in this room that Sarah Friar just said anything about OpenAI ultimately going public. I did not. I said it could happen. An IPO for open air obviously makes a lot of sense. The amount of money that they need is insane and the fact that they just raised significant amount of money and product Stargate, which by the way is very far behind the initial investment that was discussed. But by the way, right now there's less than$10 billion that were actually released out of the hundreds of billions of dollars that were committed, then it's not clear where the rest of the money is going to come from and exactly when. So an IPO makes a lot of sense. You will allow them to raise a huge amount of money, especially with the amount of buzz there is right now, and that will allow them to continue their development to world domination. Fryer also highlighted the growing AI search market as a priority for OpenAI. And she said, and I'm quoting, the search market is becoming a big market. Makes perfect sense. We talked about this before. We talked about the move by Google last week to try to fight that and maintain their very significant lead in this field. Do I think they can keep their current level of dominance? The answer is no. Do I think they can protect some of their turf? The answer is yes, but definitely search is a big, big component of what these companies are fighting for now, talking about their conversion to for-profit, there's still a very strong opposition to that move being made despite the fact they deserted their concept of completely switching to a for-profit organization, many power groups, and we mentioned Elon Musk as soon as their announcement, but many other companies are claiming that even the suggested new structure cannot guarantee that the profit will not be the deciding factor when comparing profits versus humanity. And so there's still a pushback and it's still not guaranteed that the Attorney Generals will allow that process to happen. Again I think there's too much money involved right now and too much political cloud for it not to happen and I'll keep updating you as this story evolves. And speaking of open AI's global domination, OpenAI just announced they're establishing a new legal entity in South Korea and that they're opening their first office in Seoul, which is going to be their third Asian office after Tokyo and Singapore. South Korea has the highest number of paying Chachi PTI subscribers outside of the us, which makes a lot of sense to OpenAI to grow into that direction. And they already have a lot of partnerships with businesses, policy makers, developers, and researchers in that country across multiple industries and different sizes of companies. And so this moves makes perfect sense. Combine it with the news from last week of them building data centers in different places not yet in Korea, but these conversations are happening as well. It shows you how aggressive open AI is when it comes to trying to control the global market. And from open AI to a long laundry list of interesting new releases. Manus, the Chinese AI agent company that I became a very big fan of, I'm using it now a lot to do multiple different things, just released what they called Manus slides. It allows you to enter a single prompt on what the presentation that you need, what the data that you need in it, and who's the target audience. And it will do everything for you, including doing the research, summarizing it, deciding what's the information that needs to be on the slides and doing the design of the slides themselves. and you can very easily edit what's in the slides as well. In a manual way, or while using the ai, which makes it a very easy tool to use to create presentations. Now in addition to the creation, the tool enables you to immediately export the output to PowerPoint or to PDF, which means you can export it to the two main formats that you probably want to export presentations. This is a direct competition to tools like beautiful.ai and Gamma, which have been in this field for a while. And what it is showing you is that these generalized agents, as they get better, will be able to do well. Everything. So if you think about what slides needs to do, it needs to combine multiple agents together. One that will manage the overall process, one that will do research, one that will summarize the data, one that will decide on the display, one that will generate the display, one that will put everything together and one that knows how to export it in multiple formats. And probably a few others that I can't even think of. And that all happens with a single prompt, with the AI agent. Just understanding what is it that you're trying to do, generating these additional agents and guiding them on exactly what they need to do in order to put the work together. This approach will eliminate the need to multiple tools by multiple startups that are working on niche solutions with or without ai. Because this generalized AI that can now do anything, it can spin up, agents will be able to do well, more or less everything as it gets better. Now I know there's gonna be claims about, oh, the information is not accurate, or it's not perfect, or it's not exactly aligned with my guidelines and so on. And it's probably true, but this will change and evolve and it will evolve very, very quickly and it will get to the point in the very near future that the job is going to do is better than any human doing the job. And we make mistakes as well, and our job is not always perfect. Another very interesting release this week comes from a company called Odyssey and they've developed an AI model that can allow you to interact with a streamed video in a 3D environment. So think about a video of a city or a video of a virtual world, or a video of a building, but you can control with mouse, keyboard or joystick, where the movie is going. So it's not a full simulation yet, but it actually allows you to navigate a simulated world that is generated with high fidelity video in real time. Now this tool basically allows you to engage with a 3D environment while not having to create it in advance in a simulation engine. And currently it is limited to a five minute video, but that's still a lot. But I'm sure that limitation will go away. And they're just one company out of a few that have similar solutions. So DeepMind has a solution like this that's called World Labs, and Microsoft has a solution like this, and Decar has a solution like this. But what it tells us is a few things. One is that the world of simulation is going to change dramatically. B, that video will most likely become interactive in the future. Think about your TV show and be able to turn around and view it from any direction or a movie that you'll be able to split off to a new direction and move it in different ways that were not originally planned by the people who created the movie and so on and so forth. Making a lot of the world of the entertainment as we know it, a lot more immersive and a lot more interesting and a lot more engaging than it is today. It has obviously profound implications on the entire industry that currently generates the entertainment content that we are consuming. The interesting thing about Odyssey is that the data that they're using is not publicly available data. They've designed a 360 camera backpack that people can carry around and walk in different places, and that's how they're capturing the real world and the landscapes and the pictures and so on to train their model so they're not using just random videos of the internet. That being said, if you have been following the VEO three videos that are swarming the internet right now, you'll understand that VEO three now has a very solid understanding of how the world operates from a physics perspective and from any other perspectives because it can mimic the real world very accurately without having access to unique data just by watching probably every single video on YouTube and then some. That being said, it doesn't do it in a 3D immersive universe right now, and it does not allow you to control and move through that universe. I don't see a reason why Google will not go in that direction as well once and if it becomes interesting for them to do so. maybe the most interesting release this week has come from deep seek. So the Chinese companies just released a new model while doing it very quietly, they didn't make any public announcements. All they did is wrote about it in their WeChat group, stating that they have completed a minor update of the R one model. While that model is right now at the top of the coding leaderboard in China, surpassing models from much bigger companies and competing with all four mini high, all three, all mini, mini medium, and Claude 3.7. It's probably not as good yet as Cloud four and Chachi PT 4.1 in coding, but it's getting very, very close. And again, the competition with China is very tight. Another interesting tooling solution that came available this week is Perplexity. Perplexity just launched labs, which is an AI project team of agents that can build apps, reports, and dashboards in just a few minutes. It is available to all the paying subscribers of perplexity. So whether the pro subscription or the enterprise subscription, and it goes way beyond what the chatbots could do. So if you think about from a layer perspective, from a concept perspective, we had the regular perplexity that can do a quick search and give you answers. Then we had deep search that can find a lot of information after investing significantly long time and synthesizing the information. Well this is the next step forward because once you have that information, what are you going to do with it? Well, you need to create some kind of a report or a dashboard or a tool or a bunch of documents or an application or a webpage, and all these things can now be created by labs. So it's basically the next logical evolution of going from search to deep search. This is just the next step, taking over, the next thing that the humans needs to do in that chain of effort. For larger projects, this is similar to some of the things I've been able to do with Gens, spark, and Manus, which has a broader application to do more things. But I assume because it's tailored to do that, it will do these kind of things more effectively than Manus and or spar. I will definitely test it out and keep you posted. There's an episode coming up on this Thursday live session in which I'm going to show you how to safely run Manus and Gins, spark, and some of the things that you can do with this. And when I mean run safely, it's not touching any of my hardware, any access to any of my logins or anything else, So I can experiment however I want with these extremely powerful tools without taking any risks. Again, if you're interested in that, come join us for the live on Thursday. But back to perplexity. This really interesting tool is gonna be available to, as I said, all the paying subscribers. Each paid subscriber will get 50 lab queries per month, which is a lot. It's more than one per day, which I think is way above what the average user is gonna use it for, and it is going to be available on their web interface, iOS and Android platforms with Mac and Windows desktop support coming soon. Two interesting aspects of how this works well. All the files created doing the workflows are organized in a new asset tab for easy viewing, browsing, and downloading, and all the interactive tools and dashboards are available to a separate tab called app. My personal note on Perplexity. While they don't have their own models, I think they've been very effective in tooling, basically building tools around existing AI models and making them very effective. Is that good enough to keep them as a competitive company in this crazy market? I don't think so, but so far they're doing a very good job in delivering tools that are very helpful. By the way, going back to this new labs tool combined with something that I shared with you about a month ago, that now the enterprise level of perplexity can research your internal information such as a SharePoint Drive or a SharePoint site completely. It makes it very, very interesting because you'll be able to create dashboards and documents and summaries about any project or any topic in your company based on your internal information and or internal information combined with external information with just one prompt. I've been experimenting myself more and more with Vibe coding platforms and getting very interesting results with them, but we haven't heard of a Chinese vibe coding platform until this week. So a new young startup that is only six months old called U Wear backed by some top venture capital companies from China, has just announced their first platform that is going to compete in the Vibe coding universe. and I found this article on the information on May 29th. So this is very, very recent and they're obviously trying to ride on the success of Deep Seek and Manus, but just bringing it to the vibe coding world, I have zero feedback right now on what that tool is, exactly what level of vibe coding it's going to compete with is it's going to compete in the professional world like Cursor and Windsurf or more on the casual user with tools like lovable and rep lit. So once we have more information, I will share that with you. Another very interesting release this week for the creators between us is Black Forest Lab. The company behind Flux just released Flux One Context and Context is actually spelled with a K in this particular name. But what it does is it improves the way images are created, but more interestingly, it allows you to modify images with simple prompts while keeping coherent renderings. So basically reprompt an existing image and ask the AI to make very specific changes in it, and everything else stays consistent. That also goes to changing angles and changing zoom levels and things like that. and from the demos that I've seen, it's a very powerful capability that doesn't exist almost at all in any of the other tools. I mean, you can do this with other tools as well, just not with the same level of consistency, which is the key in all of this. In addition, this model works extremely fast, so it's generating images really, really fast. It's making the updates really, really fast, which makes it an ideal tool for people who need to create images and edit images professionally. And if you wanna test the tool out, it is available basically everywhere Flux is available. So CRE and I and Free Pick and Open Art, and Leonardo and on. File and replicate and runway and data crunch, and together AI and confi org. Basically everywhere you can get access to Flux, you can now get access to flux context. Xai also released something interesting this week or at least announce it, I don't have access to it yet, where they are preparing a new screen sharing feature for their live VOS mode on iOS and probably Android will follow shortly. And what it does is it allows GR to see your phone's screen and engage with it, and that could be helpful for tasks like translation or helping you navigate a specific app that you don't know how to operate or support during anything that you're doing on your device. This is a similar functionality has existed on chat GPT Live mode on the phones for a while now, so they're just closing the gap. I must admit that I was very excited when this functionality came out on chat GPT, but the reality is I use it very, very rarely. I use the screen sharing function of Gemini on my computer screen significantly more frequently. But I think the day where we're gonna just have AI as the operating system and it will just see everything that we see both in the real world and in the digital world is coming very soon. Another company that we don't talk about a lot but is now reviving an old browser is opera. So opera is resurrecting Opera Neon, which is a browser concept that they introduced in 2017. Back then, it didn't really work whether they're now bringing it back where AI is gonna be in the center of it. The new neon design is built around a local chatbot that they're calling the agentic browser operator, but they're also a cloud computer, which probably sends bigger tasks to the cloud, I assume similar to the way that Siri currently works. It will be able to do tasks way beyond just basic browsing and analyzing of data. It will be able to install Python libraries and run JavaScripts and create them for you. So more on the development side of things. I don't think they have a chance of competing with the development environments, but being able to spin up small applications within the browser is not necessarily a bad idea. The interesting thing about it is they're planning for this to be a paid subscription. sO if you will want to use this AI driven browser, you will have to pay a subscription fee. Now, currently no other browser is a paid browser. Or at least most of them are free. But I mentioned with you in the past that the whole concept of how the internet is being paid for by ads might be changing because if agents are the ones that are browsing the webs and not humans, well ads, at least the way we know them right now, are gonna be not as effective or eliminated completely. And hence, having paid browsers might be one way to keep the internet running through, collecting money and then pouring it into actually running the backend of the internet. And opera is obviously jumping into a hot topic where multiple companies developing agent based browsers another interesting one that we discussed recently is arc. So ARC is a favorite browser by many people and they're now announcing a new version or a new variation called do That is going to be an agent browser that is now in alpha testing and it'll replace ARC once it's ready to go. So we're done with small releases, uh, this past week. And yes, there've been a lot of small ones, but now to some other interesting news, New York Times just signs a first AI licensing deal, and it is with not open AI and not anthropic. It's actually with Amazon, which is really surprising because it's the first time Amazon is signing such a deal. It's also the first time the New York Times is signing such a deal. And,if you remember the New York Times about two years ago already sued OpenAI and Microsoft for copyright infringement. So this might be signaling a change in their strategy from litigation to monetization of their content. What exactly Amazon is planning to do this, I assume it's gonna be embedded into Alexa. So every time you're gonna ask Alexa for news, the times information is going to be showing up either on your screen or on the voice communication with Alexa. They haven't disclosed the terms of the deal, so it's unclear exactly who's paying what for who, and exactly how it is going to work. But it is very clear that this new path might be the lifeline for journalism by being able to distribute their content through new channels and make money that way because it's been very, very hard on these publications to actually stay profitable. And that is definitely a better way forward than litigation in which they may or may not even win. Switching to AI in the physical world, a few robotic stuff and another exciting piece of news from Tesla. So Tesla is presumably beginning its long awaited robotaxis service in Austin, Texas on June 12th. This is less than two weeks from the time this podcast is going live now. The first test is gonna have only 10 model y SUVs that are going to drive around the city. They're going to test it for safety and application and many other things, and once and if it is successful. Musk is saying that they're planning to deploy tens of thousands of these across multiple cities with thousands in Austin within a few months. Now, Musk has made multiple claims about Tesla in general and about self-driving cars and about Robax many times before. So the timelines are still vague, but I think the test is actually going live this coming month. If you've been following Tesla, Tesla sales, EV sales have been declining and dropped about 20% in the first quarter of 2025. So the robot and the Optimus robot might be really critical to the future of Tesla as a leading company in the world. Now, another avenue that Tesla is pursuing, they are in conversations with multiple major automakers, to license their FST software that allows Tesla the self-driving capabilities, which is another reason why Robotaxis can help them a lot because it can prove that it can actually work, which will make it more valuable for other companies to pay for. So several different avenues for Tesla. It'll be very interesting to see how this evolves, and I will obviously keep you posted as this story comes to life. But speaking of Optimus and robots, as you remember last week, I shared with you that Tesla shared videos of Optimus doing housework in a very effective way, and that they're training it by just watching videos. well Chinese companies, UB Tech Robotics Corporation has just announced that they're going to be releasing a new version of the robot that is going to cost 20,000 US dollars. That is aiming to be a household companion robot. They're planning to start production later this year and ramping it up in 2026. When I say starting this year, they're planning to ship around a thousand units still in 2025. Now, their immediate market is the home companion market in China, which has a growing need for elderly care. They have a very, very large, older population, and being able to help them at home is becoming a necessity. So there's definitely a market for that. And with a price tag of$20,000, it will make it an option to people with enough money or as a service that you can probably rent for a few hours a day or a few hours a week that can come visit you. Just like nurses do today. You will be a robot that will come and visit you and help you in the house for a few hours. Now, if you haven't heard of UB Tech, they are not new to this market. They have started the robotics journey building much more expensive robots for industry, and they're already working in factories of companies like BYD and Foxcon Technologies. But these industry robots cost about a hundred thousand dollars a pop, which is obviously a lot less relevant for home care. Now, UB Tech lost about$153 million last year. Their stock has dropped 45% in the last 12 months in the Hong Kong stock exchange. And so making this consumer pivot might be a way for them to potentially save the company. That's it for this week. If you find this podcast helpful to you, please rate it in your favorite platform, whether it's Spotify or Apple Podcast. And as I mentioned, please share it with others. That's your way to provide AI education and AI literacy to as many people as possible. So if you can pull your phone out right now and click the share button and send it to a few people that you know that can benefit from this podcast as well. As I mentioned, the live episode on Thursday noon Eastern Time is going to be about how to run the general agent tools like Gens, spark, and Manus in a safe way. And until then, keep experimenting with ai. Keep learning how to use it, keep sharing what you learn, and have an awesome rest of your weekend.