213 | OpenAI’s monster week 🤯 GPT-5 for everyone, $12B revenue, new open-source models & an $8.3B raise, Anthropic’s surprise enterprise lead & Claude Opus 4.1, New models, a Prime Minister now asks ChatGPT for policy advice and more AI news for August 8,

27:48But it wasn't all unicorns and butterflies for ChatGPT this week. And now we're going to talk about several different problematic topics, some connected only to ChatGPT, some connected to the AI world in general. But the first one is thousands of chats were exposed on Google search. So OpenAI had a share function feature that allows users to opt into indexing by clicking a box saying, make this chat discoverable. Now in the fine print, it said that this means that your entire conversation will be available to search engines. And what happened is a lot of people who use the feature then found that some intimate or secretive or problematic conversations that they had are showing up on Google's search results.

28:34Now, the chief information security officer for OpenAI called it short-lived experiment to help discover useful conversation, but it's saying that OpenAI removed the feature and they even work hard to remove those chat and de-index them from search engines as of Friday morning. Combine that with the news that we shared with you a few weeks ago that follows OpenAI failed attempt to block court's order from requiring indefinite retention of even deleted chats. And it tells you that while we got used to using this tool and providing it every information that we have, because it is quote unquote safe and it's not shared with anyone, there are loopholes that are either created by new features or legal loopholes that may require OpenAI to provide your information to whoever needs to get it from a legal proceeding perspective that tells you that you need to consider what you share with that tool very closely.

29:27I assume it's going to be very similar with any other AI tool. Now, two interesting studies that were published this week that are relating to advancements in models that are related to the release of GPT-5 and new open source models. And we're going to talk later also about Claude Opus 4.1. Well, a new research from MIT Sloan found that while advancements in models improve outcomes, half of the improvement actually comes from users refining and improving their own prompts. Now for full disclosure the experiment was done on image generation models, older image generation models, DAL-E2 and DAL-E3 but regardless it had 1900 participants and that measured their performance across multiple aspects of using these models and what they found is that users that prompted better get significantly better results versus users who just upgraded to a better model.

30:24As an example, a user who got better results wrote 24 % longer, more descriptive prompts. Another interesting thing that they found is that non-technical users excelled in prompting. And I'm quoting, the best prompters weren't software engineers, there were people who knew how to express ideas clearly in everyday language. This is really a very short and perfect way to describe why large language models are so attractive and so good. You do not need to be a programmer. Actually, it's better if you're not a programmer. It is better if you just know how to clearly, descriptively, and in detail describe what you're trying to achieve, and the AI will achieve it for you.

31:03Now, another interesting thing that they found is when users were trying to use AI to rewrite their prompts, they actually got a 58 % decrease in their performance, mostly because it added irrelevant details that shifted the user's original intent. And the main professor that was running this research emphasized, and I'm quoting, people need to be caught up with these technologies and know how to use them well. Basically, what he's saying is what I've been saying all along and why I'm teaching the courses that I'm teaching, doing the workshops. It's not just about the models. It's about how you use them and having the skills and the knowledge and the understanding of how they work so you can prompt better and get significantly better results if you know what you're doing.

31:45Another interesting research that was shared on TechRadar, and they're sharing that while Sam Altman's visions and everything that they're doing and all the things that they're sharing across code writing and healthcare and science and all these things eventually come down to 77 % of chat GPT users use it as a chat engine. The actual survey was done by Adobe across US chat GPT users. So three quarters of people who are using chat GPT are not really using it as an AI tool and not driving any real efficiencies or real benefits that AI provides. They're just replacing Google. By the way, out of those people, 23 % prefer ChatGPT over Google, but they're using both back and forth.

32:3455 % use it for just general inquiry. 53 % are using it to get creative ideas and 20 % just ask financial advice. Another interesting piece of statistics is 13 % are already using ChatGPT as a shopping assistant, which is more than I expected and more about the impact on e-commerce in a few minutes when we're going to talk about Shopify. So what do these two last items mean to us? It means that most people do not know how to use AI. Even the people who are claiming that they're using AI daily do not actually know how to use it. They don't know how to prompt. They don't know what benefits they can get from it.

33:12They don't know what use cases it can be applied for effectively and so on. Now, this is not surprising to me in the courses that I teach, in the workshops that I do with companies, more and more people, obviously over time, because I've been teaching these courses for over two years, start to come to the course with AI experience, quote unquote. But once I start drilling deeper and try to understand what they're actually doing with AI, you understand in most cases, it is very, very basic. And it drives very little business value because of that basic knowledge. The flip side is when people leave the course, they have hands-on experience and they know how to apply AI across a huge variety of AI use cases, such as strategic planning and data analysis and content creation of text and video and audio and images, report generation, presentations, deck creations, et cetera, multiple aspects of the business, including strategic decision-making and strategic deployment of AI.

34:10So the range of knowledge that you can gain in a very short amount of time if you're taking a course is very significant. So if you personally want to up your AI game, you have a really immediate opportunity right now because our next cohort of the AI Business Transformation course starts this Monday, August 11th. This is two days after this podcast airs. So if you want to join this current cohort and dramatically improve your AI skills and capabilities. You have a very short amount of time to do this. The next course that we're going to teach is going to be maybe on the second half of November, if we even have time to do it this year.

34:51If not, it's going to get pushed to 2026, as I am swamped and extremely busy with company-specific private training, which is most of what I do. By the way, if you are in a senior position in a department or a company, and you're looking to train your entire team or your entire company, please reach out to me on LinkedIn or on the link in the show notes of this podcast and book a time with me to talk. And I can explain to you what the different options are, or at least educate you on what you can do in order to train your team properly and not just specific individuals. But if it is just you or you and a few colleagues, this is a great opportunity for you to jumpstart or dramatically improve your AI capabilities just by joining the course that starts this Monday.

35:35And as a reminder, as a listener of this podcast, you can use promo code leveragingAI100, all uppercase one word for$100 off the price of the course. Now, I promise you that the future you will thank you if you take the course, not joining the course may have side effects such as regret, loss of business opportunities, loss of career opportunities, and more. No, I'm just kidding with that. But, well, I don't know. I'm totally kidding. But you should join the course if you feel that you're falling behind in the AI race. It is a well-structured course that I've been running for over two years with thousands of people that have dramatically shifted their careers and their businesses because of what they learned in the course.

36:14And now back to the news. A very interesting research by Stack Overflow that has surveyed nearly 50 ,000 developers found that only 33 % of developers trust AI coding tools outputs in 2025. That's a big decline from the 43 % in 2024. One of the CTOs that they surveyed said, and I'm quoting, someone gives an AI coding agent a try expecting magic, but it messes up the architecture, changes something it shouldn't or just spits out bad code. Despite that, 84 % of developers are using or planning to use AI tools in 2025, up from 76 % in 2024. So almost a 10 % increase. 60 % of people surveyed have a favorable view for integrating AI tools into their workflow, but that is down from 72 % in 2024.

37:09So a big decline, again, a pullback from the level of trust that people had with these tools before, despite the fact that we have significantly better coding tools right now. Developers are citing and saying it's almost right, but not quite right, with 66 % of the people struggling with nearly correct code and 45 % finding debugging AI-generated code more time consuming than actually writing the code themselves. And 75 % of developers revert to human expertise when they distrust AI answers or the way they write code. That basically tells you that it's still a tango of the human and the AI, and the days of AI writing completely independent code are still not here.

37:54That being said, the latest developments with GPT-5 is showing self-correction capabilities and evaluating much larger pieces of code and understanding architecture, at least based on what OpenAI and Cursor shared in the announcement, which tells you it's moving in the right direction, but we are not there yet. So if you're asking yourself, why is the adoption growing despite the fact that the trust is declining? Well, Erin Yepis, Stack Overflow's senior analyst says, and I'm quoting, developers are learning where AI fits in their workflows and where it doesn't. So basically, if you know where you can trust it and for what tasks you can use it effectively and which ones don't, you can still use AI to improve your capabilities and your efficiency while avoiding some of the pitfalls.

38:41And as the models and the integrations get better, you can grow the amount of use cases that you can use it for. Another interesting point to take into consideration when using these models and connecting them to your system that was announced this week happened at the Black Hat Cyber Security Conference in Vegas. An Israeli cybersecurity company called Zenity unveiled a new zero-click hack that is targeting ChatGPT, showing how attackers can basically get control of users' accounts and access to all their sensitive data if they're connecting Google Drive to their accounts. Now, they're saying they can exploit the same exact mechanism to Microsoft Copilot Studio, Salesforce, Einstein, and other tools that are connecting to our standard existing tools.

39:26It allowed the attackers to access files, set malicious goals for the AI itself, influencing responses, recommending harmful actions, downloading viruses, etc. Now, they obviously share that with OpenAI and Microsoft. They both issued fixes for this issue, but the hackers who found this vulnerability are saying that this is basically the intended system behavior. The loopholes that they're using are the way the system is working, and they're just manipulating that in order to get access to that information. On the big picture, what you need to know is that using AI tools exposes you, exposes your data to more risks.

40:04This is why there's a huge boom right now of cybersecurity companies that are trying to block these loopholes and try to provide higher levels of security. But in general, you're paying a price for the efficiencies and for the benefits that you're getting from AI. And the price that you're paying is many more risks from a cybersecurity perspective. or in the way Zenity's CTO said it, in a reality where technology outpays security, organizations must take responsibility to avoid paying steeper price than the efficiencies AI offers. So enough with negative stuff, some additional interesting news, and there were a lot of really interesting or relevant new releases this week that we're going to jump into quickly.

40:48The first and the most interesting one from my perspective is Shopify just launched a new innovative e-commerce platform. They're calling it MCPUI, or if you want, MCP User Interface. And what it basically does is it connects to Shopify's entire catalog through an MCP connector, but it also provides a user interface that is rich user interface that provides detailed images and descriptions and also a full checkout capability, a unified checkout capability that allows you to search across millions of stores, whether Shopify or non-Shopify stores, and get a unified checkout that will, behind the scenes, connect correctly to the different suppliers and do the checkout for each and every one of them separately.

41:32This is a whole new ballgame when it comes to how we're going to shop in the future, and it's going to dramatically impact how e-commerce is done. If you think about it, this capability, because it's an MCP connector, allows any agent and any large language model to now query any shop that is connected to it behind the scenes across multiple providers, multiple suppliers, multiple platforms, compare different options, and then pick the ones that you want, regardless of where they're coming from, and then buy it for you, completely ignoring who's actually selling it behind the scenes. And you can do all of that while just chatting with your agent in English, or if you take it to the next level, open your camera, show him what you're looking at or how you look like or what kind of style you're looking for in your house, in your clothing, in your kitchen, whatever it is that you're shopping for.

42:24And it will find you all these things across millions of stores and will help you make the decision and will help you purchase that without ever going to any single website. If your business currently depends on traffic coming from your website, if you're an e-commerce kind of provider, this needs to scare the hell out of you and you need to start investing very, very quickly in understanding how to get exposed to these agent tools versus traditional SEO. Now, in the short term, you will need to do both at the same time. And over time, there's probably going to be more and more traffic from these agents because it makes it a lot easier and less and less traffic from websites.

43:02If you're going to wait, you may lose more traffic to these tools than you expect. And I assume there's going to be some kind of a singularity point of shopping where it will catch like wildfire and more or less overnight, a lot of the traffic is going to go to zero, but that is still a couple of years into the future. But that's not a lot of time to prepare and understand what you need to change from an infrastructure perspective in order to benefit from this new way of shopping. If you're a developer, there's early access to this Shopify MCP capabilities with a private token to allow them to get feedback on how the system works.

43:37But I expect it to roll out to everybody in the very near future. Also this week, Claude launched Opus 4.1. So this would have probably been the top piece of news in every normal average week, but because everything else will happen this week, it's in the rapid fire items as part of the releases of different models. One of the reasons it's not one of the top items is that it's not very clear what exactly they changed. They're saying it's just a little better than everything that Opus 4 was. They're saying it's mostly better in agentic tasks and real-world coding and reasoning. It's scoring better on most benchmarks, especially for coding.

44:16And Rakuten Group that has been testing it for a while saying that Opus 4.1's ability to pinpoint exact corrections in large codebases preferred for daily debugging tasks by their team. Winsurf reports that Opus 4.1 matches performance leap between 3.7 and 4. So basically they're saying it's not a small leap forward, but it's a big leap forward when it comes to junior developer benchmarks. So what Anthropic released is a new model that they're suggesting to anybody who's using Opus 4 right now to upgrade to Opus 4.1. It's going to be the same exact amount of money. So that's a cool thing. So you're getting more intelligence for less.

44:56This is obviously a part of the battle with OpenAI for coding dominance. Which one of them is actually better, Opus 4.1 or GPT-5? I'm sure we're going to get more and more examples in the next few weeks, but both of them were just released this week. Another interesting release going back to Elon Musk and controversies, Grok just released an image generation and video generation tool for premium subscribers on their iOS app. This new tool called Imagine can generate images and can generate videos. And it also has what Elon Musk calls a spicy mode that allows to create quasi-pornographic images and videos.

45:38When I say quasi, it's not full nudity, but it's very close to that. That being said, they have blocked some things that were possible to create on previous image generation models on X. So as an example, attempt to generate images of pregnant Donald Trump, that's not me trying, that was in the article was not doable. You cannot show full nudity or full pornography, but it's definitely edgy like X and everything else Elon Musk. The quality of the video and the images themselves are not to par with the latest and greatest in the industry right now. It's not as good as VO3 or not even as OpenAI's Sora, and it's not as good in generating images as the top image generation models.

46:21But Elon Musk tweeted, Grok Imagine should get better almost every day. Basically saying that's not the final word and they're going to keep on improving it. And from everything else we've seen from Elon and everything else we've seen from XAI, this is probably true. And we're going to see this model become a serious competitor with every other model out there. Another interesting release this week, Google DeepMind released Genie 3. Those of you who don't know, Genie is a series of models that can generate a full 3D real world rendering in real time and run it consistently. So the new model is a significantly high resolution.

46:58It can run a 720p resolution at 24 frames per second consistently for more than a minute. So basically you can prompt a world, whether a real highly realistic world or a game world or whatever kind of environment you weren't and it generates it in real time and you can walk or drive a vehicle or navigate or fly through that world at a high resolution keeping consistency of this entire universe for more than a minute this is incredible the demos that they have are absolutely mind-blowing and you can prompt it you can prompt it to change the weather you can prompt it to change the lighting you can prompt it to add different characters or different events that are happening in the scene the demos that they have provided are absolutely mind-blowing.

47:46They're developing it for two main reasons. One for AGI research, because now you can explore the world while actually doing it all in simulation, but creating whatever world you want for the AI to learn from. And the other reason is to train robots. If you can duplicate the actual real environment and then connect the AI training platform of the robot into that environment, you can train it to navigate across different things, grab stuff, move around, avoid obstacles, etc., all in a simulated universe that is generated in real time. Again, I must admit, looking at the demos, it blew my mind. And if they solve the length of consistency to be multiple minutes and then hours, then there's no more gaming engines to develop, as an example, because in real time, you can prompt a game and have it generate whatever universe you want with whatever level of graphics, with whatever characters, and be able to control it in real time.

48:44Another interesting announcement comes from Microsoft, who introduced Copilot mode to their Edge browser, which is turning it into an AI browser, which is a trend that we're seeing from more and more companies. It is currently available for Windows and Mac users, and this is an opt-in tool, so you don't have to use it, but you can choose to use it, and it It integrates chat and search and navigating across websites and interfaces and tabs into one unified environment. It can see all the data across all your open tabs and it can provide you answers or even help you shop across all of them together.

49:20I think the days in which we use traditional browsers are numbered. I don't know if those numbers are a year or two, but that's probably it. I think everybody will switch to AI-centric browsers as it makes a lot more sense to have the AI help you navigate the web. It's going to be agentic. It's going to know a lot more knowledge. It's going to understand you. It's going to be connected to your tools, and it's going to provide a lot more value in a much more effective way than current browsers do. The AI company that is pushing the boundaries when it comes to helping the world, the universe, and humans with AI is obviously DeepMind, and they have released several different interesting tools this week.

49:57Their Alpha Earth Foundation, which is a new AI model, takes satellite imageries from multiple sources in high resolution and combines them into AI together with Google Earth's model. This allows researchers to completely revolutionize things like ecosystem mapping and deforestation tracking and things like that. Through this tool, they're generating 1.4 trillion embedding of satellite footprints every single year. And it's driving 24 % lower error rate when trying to analyze this data with 16x less storage needed to do the operation. Over 50 global research partners are already using this technology and it's helping them, as I mentioned, track multiple aspects that are happening across earth in order to research them and hopefully help us make the world a better place.

50:51Another interesting release from DeepMind is what they're calling PERCH AI model, which is taking bioacoustics data and turning it into data sets that can be analyzed. This can basically find and analyze communication between animals, even with background sounds such as forests and oceans, and that allows scientists to find different species, including endangered species, as well as research their behaviors in a much better way. If you go back to early episodes of this podcast, I shared many times that I'm waiting for the day that we all can be Dr. Doolittle and be able to talk to animals or at least understand what they're saying.

51:30Because if AI can analyze our language, I don't see a reason why it cannot analyze animals' language. There's just less of a financial drive in this right now. But I definitely think we'll get to that situation. and as an animal lover, I would love to see that happening sooner rather than later, but at least for now, it is helping research in science. Two more interesting releases from Google. The latest version of the Gemini app allows you to create illustrated storybooks that are completely AI generated, including the story and the art and including voice narration that can tell the story. This is perfect for a family activity and I absolutely love it because you can use it to teach your younger kids whatever you want with a story that can be based on either a prompt or an image or even the graphics that the kid has scribbled on a piece of paper.

52:20You can use that as a reference and use it as part of the creation of the book or think about images of your family vacation and so on and combine all of those into a story that can teach your kid about whatever you want to teach them about. Whether it's simplifying science or just teaching better behavior and being kind to other kids and things like that, you can create stories with whatever reference will make them tick and use it together with them to show them how AI works. So I see huge benefits in doing this kind of thing. I'm going to try this this weekend. Once you create one of these stories, you can download it, you can share it, which makes it even more exciting.

53:01And the last thing that I will share from Google. Google have made a big upgrade to Notebook LM, which is one of my favorite tools. It's a tool that allows you to take a huge variety of sources and turn them into knowledge. And they just completely upgraded the way you engage with it, which allows you to easier use the platform. But in addition, they added video overviews. So it had audio overviews, which were a big hit because it turns whatever information you bring in into this podcast. Well, now it turns it into a narrated PowerPoint presentation. There's basically graphics that it grabs from the content that you give it, as well as generates graphics on its own, plus a little bit of text.

53:37So think about, again, a PowerPoint presentation with a full narration that can explain whatever you want it to explain to you, either it's content from your company or a topic you want to learn. And I find this very effective and very helpful. It's currently available only in English, but they're planning to add it with additional languages moving forward. Continuing with other releases this week, Alibaba just released Quen Image, which is a 20 billion parameter open source AI image generator that is actually working really well. It's creating solid images, but its biggest strength is creating text.

54:11It's creating clear text in both Chinese and English and can combine Chinese and English in images. and you can create anything you want with it, whether it's slides, posters, marketing, content, and so on. It is an open source model that you can use under an Apache 2 license and you can get it from every place that you can get open source models. I've seen mixed reactions to people who already tested it. Some are saying that it's very, very good. Some are saying that it's disappointing and it's below par compared to other models out there. But based on the fact it's their first variation of this, very similar to what we talked about with Elon Musk's model.

54:48I definitely think that it will get better. The examples that they shared on the website are absolutely spectacular, but I'm sure they were handpicked, and that doesn't mean that's going to be the results every time you run it. Another very interesting open source model that was released this week is Black Forest Lab, the company behind Flux One, just released Flux One CREA, which is a new version of their image generation model that comes with open weights and generates stunning and amazing photographs. And what they're saying that it improved and redefines photorealism, and it's taking away the AI-ness from images that are generated by AI, allowing them to look even more realistic.

55:29It is fully backward compatible with the previous version that was just called Flux.1, which means you can just replace whatever you built with the previous model with the new model and get even better results. It is also available through the API, through every place APIs exist, like Replicate and File and Runware and Data Crunch and tools like that. I love using this tool. I use it through both Replicate and File, and I get really good results with it. I didn't get a chance to test the new one yet, but I'm going to. They also built a whole rigorous set of tools that will prevent it from generating what should be censored content, which is also a very big benefit, especially coming from an open source model.

56:10More news from China. Tencent just launched a set of their open source tools ranging from half a billion parameters to seven billion parameters with 256 ,000 context window hybrid reasoning. So it knows when to reason and when not to reason like all the latest models. And per them, the smaller models are very effective on running on edge devices such as phones and computers. Per their notes, they are outperforming GPT-01 mini on multiple benchmarks. So again, another open source capable model coming from China that is now available everywhere that open source models are. So that's it for model releases this week.

56:52And I know that was a lot. There's probably a few more, but we covered, I think, the critical ones. And now we're going to talk about valuations and fundraising that, again, went crazy this week. So the first piece of news is OpenAI just raised$8.3 billion at a$300 billion valuation with a total goal of raising a total of$40 billion in 2025. Another company that did a big raise this week is Clay. Clay is a sales automation startup that I use in my work, and I think they have an amazing product. They have much bigger clients than me, like OpenAI, Anthropic, Canva, Intercom, and Rippling that are all using Clay as part of their sales automation process.

57:33They just raised$100 million Series C at a$3.1 billion valuation. This is a crazy amount of money, unless you had companies raising billions recently, which makes$100 million look like a small amount of money. But it's a very significant raise for Clay. Good for them. Also, what's good for them is that, again, a$3.1 billion valuation after their previous raise was six months ago at$1.25 billion. So they almost tripled their valuation in just six months, but their revenue is growing and they're doing well. And as I mentioned, they have a really solid product. Another company that is looking to raise money right now is Mistral, which is the French open source company, more or less the only significant AI model from Europe.

58:19It started in 2023, and they're looking to raise a billion dollars at a$10 billion valuation. They're not there yet, but with everything going on in the AI world right now, and definitely them being the only option in Europe, I see that as a very likely outcome. A New York-based startup called Tavili that is providing a platform allowing agents to search, crawl, and extract information and insight from both public and private sources while aligning with compliance requirements and company policies just raised a$25 million Series A. So that's a pretty large amount of money for a Series A. And again, we're seeing more and more AI startups getting into more and more aspects and niches of the data flow and security between different aspects of AI and existing systems.

59:07And Tavili is just another example of those. So what does this tell us? It tells us that there's a crazy trend of investing money in AI companies and it's not stopping. and we're going to continue seeing these kind of news every single week. With that, we're going to switch to the impact of AI on the job market. Microsoft is just laying off another wave of employees, specifically at the Redmond campus. This one is a relatively small wave compared to previous ones. They're laying off 40 employees, but that adds up to approximately 15 ,000 jobs that Microsoft cut since the beginning of the year across the entire US, which is just one company out of multiple tech startups that are letting people go.

59:48To show how significant it is, we just got an update from the Bureau of Labor of Statistics, or BLS. They're saying that 26 ,500 IT jobs were lost in 2025 so far. Now, they've just revised their data, including some really big shifts in the data that they provided before. So their May data that was showing a growth of 125 ,000 jobs was cut down to 19 ,000. So instead of 125, only 19 ,000 jobs were added. June, similar change in the numbers. They slashed their original estimate of 123 ,000 jobs created to 14 ,000 jobs created, which is showing, first of all, a lot of really surprising and scary inaccuracies in the data the government is collecting, but also is showing you that some of the positive signs that we had earlier this year were actually inaccurate.

1:00:39Now, the other thing that is very obvious from this report is that IT unemployment rate hit 5.5 % in July, which is way above the national average rate for everything else. And broader tech companies' occupation maintains a much lower 2.9 % unemployment rate. While this is somewhat a positive sign that everything else is still holding, it's just because AI is more advanced in writing code than it is in other things. As it evolves, which we're going to talk about in a minute, that may not hold. Some bright spots in the reports are showing persistent demand for developers, software engineers, system architects, support specialists, and cybersecurity experts.

1:01:20But it's looking for people with AI skills in these fields. So again, going back to what you heard me say a million times, if you do not learn how to use AI in your field, your job might be at risk. And people with AI skills have a much higher opportunity of making more money and getting better jobs than the people who don't. So do not neglect that and make sure you train yourself and learn how to use AI effectively. And a Goldman Sachs economist, Joseph Briggs, just shared on the Goldman Sachs Exchange podcast that the biggest spike in unemployment is with younger individuals ages 20 to 30 years old.

1:01:59And it is because many CEOs prefer to use AI instead of hiring people for entry jobs. Now, they're expecting AI could displace 6 % to 7 % of US workers over this coming decade. But they're clearly stating that our analysis doesn't factor in potential for the emergence of AGI. Basically, what they're saying is that they're predicting the future based on the present, meaning the current level of AI capabilities versus the future AI capabilities, which are definitely coming potentially a lot faster than they think. And so what this tells us is that the impact, first of all, on entry jobs, first of all, on IT jobs, and then slowly over the rest of the economy will be amplified as AI gets better and better.

1:02:45I said multiple times on the show, I think the concept of AGI doesn't really matter. AI is going to keep on getting better. As it gets better, It's going to displace more jobs or at least require retraining, reskilling of people in order to keep them relevant in an AI-centric workforce. And speaking of when we're going to get AGI, Demis Asabes, Google DeepMind CEO, was speaking at South by Southwest London, and he shared that he sees AGI coming in the next five to 10 years, possibly the lower end of that, which means five years. Now, he sees an amazing world of abundance, that's a quote, where resources aren't zero sum.

1:03:25Anybody who has been following Demis knows that that's been his opinion all the time. Everything we shared earlier on this show that talks about what kind of models they're investing in developing and they're releasing to improve research and science tells you that that's what he's pushing for. But he's also saying, and I'm quoting, we've got to make sure it gets distributed fairly, but that's more of a political question. Basically, what he's saying is that there's a very serious risk of more power being concentrated by a lot less people, both from an economic perspective as well as from every other perspective.

1:03:58And if history of mankind tells us anything, it tells us it's exactly what's going to happen unless we do something very, very aggressive to prevent that. Demis also defines himself as a cautious optimist, and he was citing humanity's adaptability from hunter-gatherer brains to modern civilization, basically saying the brain we have today is the same brain we had when we were picking strawberries to stay alive in the woods and trying to hunt animals. And yet right now we're running this incredibly sophisticated and developing AI, and our brain was able to adapt through a very large change. my biggest fear is just speed.

1:04:35We went through those changes in a significantly longer amount of time than the speed we need to adjust to AI. But by the way, I'm on his side of the fence. I'm a cautious optimist, and I'm hoping that everything that I'm doing and other people are doing when it comes to education and creating awareness and driving action when it comes to AI will lead us to benefit from the benefits of AI while reducing the risks as much as possible. And I promised you a piece of news about a prime minister using AI. Well, the Swedish prime minister, Ulf Christensen, has sparked a national controversy and a big riot after openly admitting that he is frequently consulting with AI chatbots, including ChatUPT and Le Chat, which is the chat by Mistral that we mentioned earlier, to get alternative perspectives on policy and decisions that he's making.

1:05:25The exact quote is, I use it myself quite often. If for nothing else, then for a second opinion? What have others done? And should we think the complete opposite? As I mentioned, that raised a lot of criticism, including people who said we didn't vote for ChatGPT. And there were also security concern raised, basically saying he's inputting his political thoughts, as well as potentially government secrets into models that he does not understand. He does not necessarily know who controls them, where the data goes, where the servers are located, and so on. Now, while I personally think it's a very good idea to use AI to get opposite opinions, in order to broaden your horizon and get into additional points.

1:06:04I do that all the time when I do things in my business. It does bring me back to what Sam Altman shared in his interview a few weeks ago that I shared with you on this podcast with the three main things that worry him about the future with AI, where the third thing that he said is that AI will quietly take over the world by becoming the decision machine for everything that we do. So not by taking over the world aggressively, but just by the fact we're going to consult with it and have it help us make decisions across every level, including government level. Well, it didn't take very long for us to get a real example of that happening.

1:06:39And again, while I think there's benefits in that, this might be a first step in a very scary, slippery slope that Sam already predicted. I do very much hope that we're going to find ways, as I mentioned, to benefit from it while not becoming dependent on it. And another interesting policy and governance related topic, Illinois has just became the first state to formally regulate artificial intelligence in mental health care, basically passing legislation that prohibits AI-powered therapy unless it is supervised by a licensed professional. And they're saying, and I'm quoting, just like a doctor might use AI for informational purposes when considering a patient's diagnostics or when researching treatment options, the therapist could still do the same with consent, but this supplementary tool would not replace medical know-how, it would augment and support it.

1:07:33I think this is an interesting initiative. I think on one hand, yes, this is important. I don't think there's enough science right now and research right now to say whether AI does a solid enough work compared to humans. So it doesn't have to be perfect. It just has to be as good as human therapists. It raises several questions from my perspective, and I will leave you with these questions. Question number one is, have we done proper research? Did they prove that human therapists actually provide better care to people than AI therapists? I don't think they've done that research. I don't know if they have evidence.

1:08:07I think they're assuming that and that's how they made their decision. The second thing is, what about the aspects of availability and affordability of therapy? If there's people who cannot get to a therapist because they live in a rural area or because they cannot afford a therapist, AI could be a great option over getting no help, and it can give some help. Now, again, does that require additional research? Yes. Do we need to make sure it doesn't suggest to people to jump off bridges? For sure. But do I think it's a legitimate option that should be available under specific circumstances? Absolutely.

1:08:44And so I don't think this is the right way to legislate when it comes to AI. But the fact that they're at least thinking about it is a good step in the right direction. I would love to hear your thoughts about it on LinkedIn or send me an email or whatever you think, because I find this a very interesting topic to discuss in the future. That's it for this week. I know this was a long episode, but there was a lot to cover. There is a lot more than that. And there were many, many articles that did not make the cut. And you can get to see all of them by signing up for our newsletter. There's a link in the show notes.

1:09:18And while you're opening your phone to see the links in the show notes to either connect with me or to get the rest of the articles, you can share this podcast with other people you know that can benefit from it. We'll be back on Tuesday with a very powerful How to Use AI episode, as we always do. And until then, have an awesome rest of your weekend. And don't forget, there's a course starting on Monday. So if you want to join us, you have less than 48 hours to sign up.

1:09:46you