233 | ChatGPT porn chat is coming, Job quake: 100 million roles at risk this decade + Mechanize’s “total workforce automation”, the biggest lessons from AgentForce, AI cancer-fighting novel breakthrough, and more AI news for Oct 17, 2025

28:10In December, as we roll out edge gating more fully and as part of our treat adults user like adults principle, we will allow even more like erotica for verified adults. Okay. First of all, let's talk about what happened on X and on Reddit as a response to this announcement. And a lot of people went very, very negative, somewhat nuclear responses to this announcement. The general consensus was that OpenAI has said in the beginning, and then later on now that they're trying to shift into a for-profit organization, that their main goal is to making the world a better place with AI. And yet now they are giving us another social media platform, one that might be even worse than the existing social media platforms.

28:59And they're talking about the release from last week of Sora 2 as a social app and a porn chat. One user, an ex said, and I'm quoting, I don't remember Pornhub claiming they're going to make the world a better place. Another user shared a video interview of Sam Altman from 10 weeks ago, in which he said he's really proud of OpenAI for not falling into the trap of trying to capture people's time and attention on ChatGPT, but actually trying to provide as much value as possible. And when it was pushed by the interviewer of an example, what does that mean? He literally said that he's very proud that they still did not create a sex chatbot on top of ChatGPT.

29:40And here we are 10 weeks later when they are building an app that will capture as much of your time and releasing a sex chatbot on top of ChatGPT. Now, all of that is true, but I want to give you my personal opinion. I shared in length last week what I think about the social media app that they released with Sora 2. It scares the hell out of me. and it scares the hell out of me because it is the first time that a social media platform can generate the actual content. If you think about it, yes, humans are asking Sora to generate it, but Sora can generate it on its own. It doesn't need you to tell it.

30:16All it has to do is to follow the behavior. And as I mentioned last week, potentially with the camera, seeing your face and your expression and what you're about to do, and then serve you a new piece of content that would be perfectly aligned with your personal needs, wants, and wishes at that exact second. Now, is the technology there right now to do this? No, because it takes a little longer to render the videos, but this is what's coming. And that is really, really, really scary. Like personalized fabricated content to the wishes and needs of the individuals, completely ignoring everything else just to keep you on the platform as long as possible.

30:53This is a glimpse of where this is going. But the flip side is on the adult chat question. I have no problem with that aspect as long as they can really verify that it is adults. Now, we have had porn sites forever, like since the beginning of the internet. And it doesn't matter whether I agree with the fact that we have porn sites or not in modern society. This is the reality and they are there and they're available. And in most cases, any 13-year-old kid that has a phone or 9-year-old kid or girl that has a phone can access them because there's no real verification of age when you go to most of these websites.

31:32By the way, a friend told me that. I don't really know it. And the other aspect of this is I have a very serious issue with OpenAI, Anthropic, Google, or any other large language model lab being the moral guardians of the world. Why should they decide what is acceptable and not acceptable, especially across the entire world with different societies, with different beliefs, with different moral codes and so on, why should they decide what is right and what is wrong? So I have no problem with them allowing people to do whatever they want within what is legal, not what they think is moral, within what is legal, because this is how modern society works.

32:13And we treat adults as adults and we allow adults to make adult decisions. And we allowed it during the internet era, and there's absolutely no reason we will stop doing it in the AI era. And I would love to hear your opinion about this. I would love to hear the opposite side of this conversation. So find me on LinkedIn. Again, you know my name. I'm Isar Matis. I'm on LinkedIn every single day. And tell me what you think about it. I'm obviously going to post about this as well, so we'll have the opportunity to respond to that. This wasn't the only issue with OpenAI in the last couple of weeks.

32:46They, as we mentioned, really Sora 2. Sora 2 has an opt-out option from an IP perspective. So you can create videos of any person, any known actor, any character from cartoons around the world, et cetera, et cetera, et cetera. And if the companies or the people want not to be included, they need to send a formal request to OpenAI saying that they want to opt out of their likelihood and IP being available on Sora 2. That obviously raised hell from many different directions, especially the Hollywood creators. So the Motion Picture Association, agencies like WME, CAA, UTA, which representing many, many different big stars, and the SAG-AFTRA Association as well, all has been going very loudly against this approach by OpenAI.

33:39And their claim is that OpenAI do not own the rights to any of that IP and hence should stop this immediately and replace it to an opt-in option where companies will be able to, if they want to, add their IP into the Sora universe, but that the way it is right now is a clear infringement of existing copyright law. Sam Atman said a few things. One of it is that there are different organizations that are complaining to OpenAI that their IP is not being shown enough. Basically seeing it as promotion, is that true or not? We will never know. But that was the claim that same outline made. But the other one, he said that they're going to be releasing an update to Sora 2 offering the rights holders of this IP.

34:24And I'm quoting more granular control over character generation, whatever that means. But this is coming potentially around the corner. As a reminder, Disney, Universal, and Warner have sued MidJourney for very similar issues. And when they did, I told you it's a very smart move because they have a much better chance of beating MidJourney in court versus OpenAI because OpenAI has significantly deeper pockets to fight them. And if they win against MidJourney, then now you have a precedent that you can use in the trial against OpenAI or VO3, et cetera, et cetera. Now, how will this resolve itself?

35:03I think this will make its way to the Supreme Court. because I don't think any of the sides will stop before that. And I'm not sure that the Motion Picture Association or any of the other groups, and they're probably unified together to go at this against OpenAI and the other labs, I don't think they will win. And the reason I don't think they will win is because this has been happening forever, meaning the user that uses the tool is the one responsible for whatever they are creating. If I am now going to use Canva, and I'm going to copy pictures off the internet of known characters and so on.

35:40I'm going to create an ad and I'm going to use it on social media without getting the copyrights I need in order to create that ad. And let's say I've used Disney characters. Are Disney going to sue me or they're going to sue Canva? They're going to sue me. They're not going to sue Canva. Now you can take it further back in time. When people were creating images with paint and pencils, when I create an image and I sell it. And let's say I'm making a lot of money from it. And it's an image of Mickey Mouse. Again, sorry for sticking with the Disney example. I'm from Florida. That makes it very easy for me to think about.

36:12But let's say I created a really beautiful painting of Mickey Mouse and I'm selling it. Will Disney go after me or will they go after the people who created the paints or Hobby Lobby for selling me the paint? No, they're going to go after me, the person who is using the tool in order to infringe on their copyright. I have a very strong feeling that this is exactly the card that OpenAI and the other labs are going to play. And to be completely honest, I think it's the right interpretation of the situation. The fact that I have a tool that allows me to break copyright law doesn't mean I have to use it to break copyright law.

36:50And again, this is nothing new. I could have done this in Canva. I could have created a video before, just with a lot more work. I could have created images before just with a lot more work, a little more work, again, if I'm using Canva and just existing pictures from different places around the internet. And so this is my concept. Now, my view of this, and stay with me for a minute, I promise it will make sense. I think this is a cross between Spotify and Web 2.0. So let me explain what I mean. If you think about Spotify as we know it today, as the place we consume music, and it doesn't have to be Spotify.

37:23This could be Apple Music or Google Music or Alexa or wherever it is that you consume music through streaming. This is the current state of things. Everybody is more or less happy with it. So the creators have a huge distribution channel they never had before. And they're making sense on every time their songs are being played, but it's played on a very wide audience and people get to know them. And because they get to know them, they buy tickets to their concerts and so on and so forth. Is that very, very different than what it was before? Or yes, if you remember, we used to buy albums and CDs of these creators.

37:56And the first generation of distributing music digitally was the scariest thing that ever happened to them. So those of you who are old enough like me, remember the peer-to-peer services, the P2P services that allow to exchange music online and BitTorrent files and servers where you can go and get literally any song you wanted without paying anybody anything, just by downloading it. And then obviously the biggest name that many of you probably know is Napster, right? So Napster took that concept that was kind of like an underground thing and make it into a company where you can go into their website and get any music that you want completely pirated.

38:32That took a while, but it all went away and we found a business model and a usage way that satisfied both sides. I think now, what does that have to do with Web 2.0? Oh, well, Web 2.0 was the first time that allowed anybody to create content on the web. Again, those of you who are old enough, who lived in the early days of the internet, the entire content on the internet was created by really large companies. And there was no social media and you couldn't post anything. You just consumed content off the internet. Again, for those of you who are younger, that doesn't make any sense, but that was the first decade off the internet.

39:08No individual created content and posted it online, but now it is the norm. So what I hope that will happen, and to be fair, I think that will happen, is that the owners of the IP and the creators of the platforms will find a way to satisfy everyone's needs. In other words, I will be able to create with AI anything that I want, and I will pay a little bit of money for the AI platform, and some of that money will go to pay the owners of the IP of the things that I am creating because the platform will be able to track it and distribute the revenue accordingly. This means that there's going to be more Disney characters out there and more songs by specific musicians right now and more pictures by known artists and more so on and so forth and they will make money off of it.

40:02They will become more famous and people will then consume more of their stuff. Now, does that put at risk the overall mechanism because then people may consume only this? Maybe. Do I think there is a way around this? No, I don't think there's a way around this. And the reason I don't think there's a way around this is because of the open source universe. Even if we block Sora 2, the open source tools will allow to do the same exact thing. And even if you block this variation of that variation of the open source tool, any individual can download the open source platform to their own computer, remove the limitations of IP creation, and then use it as they wish.

40:41And so I don't think there's a way to block it. And I will think the IP holders will have to work together in this new universe, just like they did with Napster and with BitTorrent servers. And we will get to the Spotify era of creation of everything. So instead of just consumption, anybody will be able to create anything. And if there is IP involved in it, you may need to pay a little more to create it. And part of that money will go to the owners of the IP. Now, will this happen? I don't have a clue. Does it make sense to me that will happen? Yes. Do I want it? Do I wish it will happen? I do.

41:17I don't know about the big owners of the IP right now. But again, as I said, I don't think they have much of a choice. Now, staying on this topic of Sora 2, Japan's government just issued a call for open AI to, and I'm quoting, refrain from copyright infringement in Sora 2 related specifically to anime characters, which they defined as, and I'm quoting again, irreplaceable treasures that Japan boasts to the world. and they're emphasizing what they're calling the irreplaceable role of anime to global culture and economy so where do we stand right now on open ai we stand in a situation that in two weeks they literally went from we are all for the future good of humanity we are not going to do things that will just try to keep you on the app as long as possible to as i mentioned giving us a new really scary social app and potentially the ability to do porn chats on chat GPT.

42:20I already told you what I think about it. But since we mentioned a government and the Japanese government approaching OpenAI, I will mention one more thing on this segment that is not completely related, but it is tied to regulations and so on. And that is California just passed a call called Senate Bill 243, and it was signed into law on October 13th. And what this bill does is it requires chatbots to clearly notify users that they are AI and not human and mandate annual reports on the safeguards in the situation that they have on negative things, such as suicidal ideation. The idea here is obviously twofold.

43:01One is to make it very, very clear to users that they're talking to an AI versus a person in order to reduce the potential dependency or trustworthiness of these tools. And the other is to generate a way to monitor what's actually happening in those platforms, especially when it's life-threatening, meaning suicide prevention. Now, do I agree with that? And do I think it will play a significant role in the future? So yes, I obviously agree with that, right? It puts reasonable guardrails without really stopping the technology from moving forward. And I think it makes perfect sense. Do I really think it will play a significant role in doing what it is trying to do?

43:42I think time will tell. I think telling people it's an AI versus a human doesn't matter. I think people who chat on these chatting platforms know it's AI. They don't think it's a human. And even if you keep on reminding it to them, it doesn't matter. Because if you develop emotions as an outcome of a conversation, the fact that somebody will tell you that your emotions came from something that's not human will not make a difference we get emotional when we watch a disney cartoon movie just think about how you felt when you saw bambi for the first time and the fire happening and what happened to his parents right we shed a tear and we become very very emotional and we know it's not real it's a cartoon on a screen and yet it drives an emotion if that emotion continues for a long period of time, you will develop some kind of an emotional attachment to that thing, regardless of whether it is human or not, despite the fact that you know it's not human.

44:38So I think it's not going to prevent that. I think it has no chance to prevent that just because the way we are wired. As far as showing people what is the impact of that and forcing companies to share publicly what is the overall situation from a social impact perspective, especially at its extremes, I think that is very, very important because it will force companies to take the right measures to prevent these things from happening. And now to the third deep dive of this week, and that's what happened in AgentForce this week. AgentForce is what used to be DreamForce, which is Salesforce's largest conference of the year announcing all their biggest announcements.

45:16Again, they've changed the name last year because they wanted to show the world that they're all in on AI and focusing on developing agents. Now, why is this very important and why is it a deep dive topic? Because it connects directly to the first segment that we had and how will AI change software and or how we work. And so in the first agent force, Mark Benioff has been very loud and vocal about how fast this shift is going to happen, how quickly it will take over the world, and how generic AI tools like Chachupiti are bullshit and has no value in the real world. And this year he acknowledged and had a complete shift in his rhetoric.

45:56He is now stating that utilizing AI effectively and implementing it on a enterprise level takes time for companies. And it's not a quick flick of a button and then you're up and running. He also agrees with what is, I think, a consensus that AI technology is definitely outpacing consumer adoption. And as I mentioned, he acknowledged to an extent the real competition. And that was kind of like the underlying clear tone that it is very obvious that Salesforce sees the moves by OpenAI and Anthropic as a threat to their business as they're growing their influence and as they're integrating into more and more tools.

46:41And it's now more or less, if you can't beat them, join them kind of approach versus this is bullshit. It will never be relevant. you have to stick with what we are giving you. Now, Salesforce shares are down 26 % year to date, and a lot more than that since its peak on January 28th, which is definitely not a good sign as far as what people thinking about the chances of Salesforce figuring out how to deliver value and how to grow sales in the current situation of things. Now, what Mark Penioff says is that he suggests that Salesforce competitive advantage lies in its deeper customer relationship and the fact that they hold, and I'm quoting, fundamental backbone of their mission-critical operations, end quote, which led them to announce their new version of the product that they called Agent Force 360.

47:33So what is Agent Force 360? It is basically their way to integrate everything, Salesforce, all the different components under one umbrella from a data perspective, and allowing the users to build agents that can connect and talk to all these tools at the same time, or as they are calling it, it's the connective tissue that brings together sales, marketing, commerce, Slack, Tableau, MuleSoft, and everything else, Salesforce, under one agentic layer. This is obviously an extremely powerful promise if it is actually doable. So let's talk first of all, what is the goal of Agent Force 360, and then talk about what is actually included in it.

48:14What they are creating is an agentic enterprise solution end-to-end with all the data and all the different tools that they have, which is a very powerful promise that I think every enterprise that is using Salesforce or that is using any other set of tools would like to have. Now, is it complete? No, it's not complete because there are many tools that are not Salesforce that organizations are using. You still have an ERP, you still have Outlook or Gmail to send your emails and you still have databases for other platforms and you have other tools, but at least on the stuff that is within the Salesforce universe, it is supposed to unify all of it.

48:51So what comes with it? It comes many different tools. One of them is called AgentScript, which is allow you to prompt and create scripts that allows users to create agents. There are many different ways to create agents. I'm not 100 % sure on the differences yet. But underlying underneath all of this are the reasoning models from Anthropic, OpenAI, and Gemini that can power your agents to be able to be very powerful and capable because of their thinking capabilities. They also created AgentsForce Builder, which is another new tool. And most of these tools are going to be released to beta in November.

49:26So they're not released yet. And this tool is allowing users to build, test, and deploy agents from a singular spot. They also are releasing AgentForce Vibes, which allows you to vibe code different applications that talk to all these underlying tools, capabilities, and agents. They're also releasing Data360, which is a new variation of DataCloud, which is a intelligence layer for their data platform that allows agents to have access to clean and reliable data and context for everything that you're building. and they're also releasing more than 300 industry-ready agents, so stuff you don't have to develop.

50:04You can just take them off the shelf, and the goal is to make them plug and play into different industries like healthcare, financial services, and manufacturing. Now, when I mentioned if you can't beat them, join them, if you go back to last year's rhetoric about OpenAI and ChatGPT versus what they announced this year, so this year versus what they announced this year, you see that there's a very, very, very big difference, And what they announced is that users of OpenAI ChatGPT that has Salesforce licenses can now connect their ChatGPT and give it full access to everything on the Salesforce app data, including CRM records, and it can control agent force tools straight from ChatGPT.

50:47Or as I said, ChatGPT will be able to, and I'm quoting, tap into powerful enterprise-grade AI right at their preferred surface environment. Basically, you'll be able to use ChatGPT to control and understand and connect with everything on Salesforce. Now, they have positioned Slack as the home for all these agents. So basically, Slack becomes your communication world with the agents behind the scenes. And so any communication that you want to do with any data will be done just like you're doing it today. But instead of talking to humans, you will talk to agents behind the scenes. I think this makes perfect sense.

51:22instead of trying to introduce a whole new way of communication, let people communicate as they're communicating today, just provide them significantly more value through an existing platform. And I think that's a very smart move by OpenAI. I already have several different integrations of AI into Slack and not natively, but just stuff that I created, including NA10 and make and stuff like that, that talk into specific channels that I have in Slack. And it's very useful and intuitive and making it out of the box makes a lot of sense. Now, the current statistics is not very promising and it's not very positive, which might be one of the reasons why their stock keeps on coming down this year.

52:00But they currently have 12 ,000 customers that are using or experimenting with AgentForce. So we don't really know how many of them are in a mature phase of actually using it. Now, that sounds like a pretty large number, but the reality is they have about 150 ,000 customers. so after over a year of trying to deploy this and getting into less than 10 percent where many of these and again we don't know how many are just experimenting and not actually deploying agent force that's not a very positive sign as far as adoption of this new technology but again agent force 360 is a whole different beast and animal both in means of its ability to connect to the data as well as in the organization's ability to easily create and deploy agents so this might change it moving forward.

52:49Now to show how amazing AgentForce is, they gave multiple examples of large organizations who are already deploying and getting benefits and immediate ROI by deploying AgentForce. These included Dell and FedEx and William Sonoma and Pandora and PepsiCo and others. So they are showing specific cases in which large organizations have deployed it and are seeing real ROI, we're obviously going to see more and more example like this moving forward. But as of right now, I think Salesforce are making the right moves to provide value. And if we connect it to the very first segment that we had in which we shared that in theory, Salesforce will be able to be your sales team and your marketing team, then if they move in that direction and if they do that effectively, they can not just save Salesforce, but make it one of the most successful companies in history.

53:45Time will tell if they are going in that direction and how effective will they be in actually deploying it in a successful way. But they definitely have the opportunity because they own the data and they're connected to everything across many, many different companies around the world. Those of you who have taken my course, I have my five rules for success in the AI era. And one of them is that there are two ways to win in the AI era, one of them is having access and owning the proprietary data. And in this particular case, it is very clear that Salesforce owns the proprietary data of a huge amount of companies around the world, which gives them in the perfect position in order to leverage that data to allow these companies to be more effective with that data, which may or may not come at the price of the employees of the companies.

54:33And now to our rapid fire items. And I'm going to start with Apple that took another hit this week. Kay Young, who just recently were promoted to head of Apple's AKI team, has left Apple for Meta after just a few weeks in the new role. So the AKI team stands for Answers, Knowledge, and Information Group that was tasked with enabling Siri to fetch real-time web data and conversational queries, basically making a smarter AI-based Siri so the person who was just made in charge, just left for Meta. This is just another nail in the coffin of Apple's AI initiatives. It literally just doesn't look good again and again and again.

55:15And this will eventually come to bite them and will start impacting the stock price of Apple because yes, it's been going up so far. I don't really understand why. Yes, they can release new devices and devices are okay. I guess they're better than the previous models, but this is not the new game. And in the new game, they have lost time and time again. I'm really surprised they haven't bought one of the leading labs or at least perplexity, which was in conversation with them so far in order to solve this problem. I don't think they will have a choice. The other choice would be to say, we're not in the AI game at all.

55:49Basically just announce it and saying, we are going to keep on building some of the best hardware on the planet and we will integrate other people's AI into it in the most effective way. I think that will take a short-term hit in their stock, but I think in the long run, that may be another approach to basically say, we're not even trying to compete on this. We will just keep on building hardware. That's what we know how to do. Or like I said, go and buy Anthropic, which is going to be very, very costly, but doable by Apple. But that's not the only person who has left a known company to join Meta this week.

56:22Andrew Tulok, who is one of the co-founders of Thinking Machine Labs, the company that was founded by Mira Morati from OpenAI who raised a crazy amount of money before even announcing exactly what product they're going to release. Well, he's leaving for Meta as well for, as a spokesperson for the company said, personal reasons. Now, this is not the first attempt of Meta to bring him over to the Meta team. The first attempt was in August with a crazy offer that presumably was about$1.5 billion dollars in stock and incentives performance over time. Now, Meta said that's inaccurate and ridiculous at the time that the article in the Wall Street Journal shared that amount.

57:09But at that time, Tulok refused that offer, whatever the offer was. It obviously was very significant, and now he is moving over. So those personal reasons might be, I'm going to become a billionaire overnight. So that's a good enough personal reason to change from one company to the other, especially when the other company already has a product and distribution and compute a lot more than your current company has. Now, a little bit of history, why Tulok is so important. He spent 11 years at Meta's Facebook AI Research, and then at OpenAI, and then obviously now at Thinking Machines. So he brings a real serious depth and experience and knowledge about the Meta's Facebook universe back into the team.

57:55In a very interesting piece of news for this week, OpenAI and Broadcom announced a multi-year partnership to generate 10 gigawatts of custom AI accelerators for OpenAI. So this is now custom chips. This is not OpenAI buying standard off-the-shelf chips from NVIDIA. These are going to be built based on their blueprint and the deployment is supposed to start at late 2026 and continue through 2029 with new investments that are coming from unshared sources at this point. This is going to be the first time that OpenAI will have its own processor after trying to do this and announcing of trying to do this several times in the past.

58:37Now, this is a really broad partnership with ARM Design Specialized CPUs to pair with these Broadcom chips. TSMC is going to fabricate the chips on their latest technology and SoftBank will obviously play a role in this in financing the whole thing. So a very large partnership that is somewhat built to hedge against the complete dependency on NVIDIA right now in future growth. And we have seen other similar partnerships in the past few weeks announced by OpenAI with other tools, but this is the first time this is going to be a chip that's going to be developed to the blueprint that OpenAI will put together.

59:18But OpenAI are not the only company who made similar announcement this week. Meta is partnering with design company Arm to create AI capabilities and AI chips to Meta's needs. And this obviously aligns with the crazy development right now of multiple new data centers. Meta themselves has several different large projects. The two biggest ones are Prometheus, which is a huge multi-gigawatt data center that is supposed to go online in 2027 in Ohio, and Hyperion, which is a five gigawatts that is supposed to go live in Northwest Louisiana in completion by 2030. And so having their own design chips will obviously allow them to do things their way and reduce cost and reduce dependency on other companies.

1:00:03And I think we're going to start seeing more of that across the board with companies, with the large labs developing their own chips in order to have less dependency on NVIDIA and third-party providers. There was a very interesting article this week on the information, which is an online magazine that I really like. They bring a lot of good food for thought and information from different sources, including about AI, and they're sharing about the current huge dilemma that Amazon has right now. I shared with you last week that Walmart signed a multi-year agreement with OpenAI, and they are becoming the first really large company to allow checkout of any Walmart product on the ChatGPT platform.

1:00:44Basically, you can go and search any product that Walmart has on ChatGPT and be able to buy it and check out right there and then without ever visiting the Walmart website or app. This took Walmart's stock up almost 5%, which in Walmart's value is a crazy amount of money at the day of the announcement. And at the same time, Amazon stock fell 1%. And if you look at a longer period of time, Target's stock went down 34 % year to date. So where does that put Amazon? Amazon today is definitely the kingpin of e-commerce in the US. They're holding about 40 % market share of US e-commerce. That's a crazy number.

1:01:28And now they need to make a choice. They either do the same thing that Walmart did and allow people to shop on Amazon right on ChatGPT or allow 800 million weekly users on ChatGPT right now, which is growing every week. So there's going to be a billion plus probably as we get into 2026. Allow them to only shop things from Walmart and not have access to Amazon, which will definitely eat into their market share. The biggest problem I think Amazon has with this is that a big part of Amazon revenue doesn't come from the revenue sharing they have on every specific product, but comes from ads on the platform.

1:02:12People who want to promote their products on Amazon are paying a lot of money for placement. And if they go down the path of allowing people to shop Amazon products on ChatGPT, and obviously later on other platforms as well as agents, they will lose a part or maybe all of that revenue over time, which is definitely not good from a stock price perspective. Now, how will this turn out for Amazon? I don't know. It will be very interesting to see how they play this. I will say something about Amazon's previous attempt at this. So if you think about it, most people before Amazon were shopping, and still many people do, shopping for goods on Google, and then from Google going to the different sites.

1:02:58And Amazon were able to change that paradigm, and many, many, many different people in the world, myself included, and definitely in the US, are going to Amazon to look for stuff and not looking on Google at all. So it's maybe the only case that in the Western Hemisphere, people are going to a different platform than Google to search for stuff. And that is the Amazon example. Can they do this in the open AI and agent era? Time will tell. I would argue that in the long run, they don't stand a chance because once everybody's used to using their own agent to do everything for them. And when I say everything, I mean everything digital.

1:03:37Then they will go to that agent to find stuff to shop. And that thing will have to find it on the internet. And then Amazon will not have a choice but to shift to that model. But in the short term, it will be very interesting to see how they play this. Now, last week, I shared with you that OpenAI is very clearly going after world domination across literally everything they can put their hands on. So this week, it's became known that they are pitching many companies to partner with them on sign in with chat gpt so similar how we have today sign in with google or sign in with microsoft or sign in with amazon with the biggest one obviously being google you will be able to sign into third-party platforms using the chat gpt connection now connect that with the announcement from last week with the announcement of apps in chat inside of chat gpt and you understand where this is going if by default you're signing into the application is with ChatGPT.

1:04:34And behind the scenes, you can connect to the app and use it within the ChatGPT ecosystem that will evolve more and more and more. Then OpenAI has their fingers into more and more and more places, and they will become the solution that people will go to for basically everything they need. There are similar solutions to that. In China, if you think about it, they have single large apps that you can use for everything, including payments and logins and transportation and literally everything you want. And it seems like OpenAI are moving in that direction in a very aggressive way. Connecting back to what we talked about in the beginning, Zendesk is moving very aggressively to develop agents on the Zendesk platform.

1:05:15And in their AI summit, they shared that their new autonomous AI agent is designed to independently resolve 80 % of customer support issues that they are seeing right now. So think about it, they have the data of exactly what is happening from a customer service perspective across, well, a huge amount of companies that are using the Zendex platform. And now they can build agents to handle these situations, to build it to spec of exactly what the needs are, which will allow them, going back to what we discussed in the beginning, to become the workforce for customer service versus just the tech SaaS platform to hold the data for customer service.

1:05:57Now, in addition, to be able to handle the 80%, they're also rolling out co-pilots that will allow the remaining 20 % of the people to collaborate with AI to deliver customer service resolution faster, better, and cheaper. And they've announced an admin layer to control all the agents, a voice-based agent, and a lot of analytics in the backend to show exactly what is happening. So whether we like it or not, this is coming and it is coming very, very fast. And even if they're exaggerating by 100 % and it's going to solve only 40 % of use cases across the entire user base of Zendesk. It is very, very significant and will obviously improve over time.

1:06:39A company called Definity, spelled with a capital D and then Finity after that, has just released Caffeine, which is another vibe coding platform. They're claiming it is slightly different than the existing vibe coding platforms. They have their own development language for a while. So this company is not a new company. And the goal is to allow people to very quickly build web applications and deploy them at a enterprise level of certainty and security. Whether that is really different than what we have right now or not, I don't know, but that's another Vibe coding solution. And we're going to have a bunch of these very successful.

1:07:16My go-to right now are mostly Lovable and Replit, but there are many others out there. And this is just another contender in this very hot part of the competition of the AI market. Staying on new releases, Grok, so Elon Musk's company just released Grok Imagine version 0.9 is a image and video creation platform. And the interesting thing about this is that it generates videos extremely fast from either text or images. So from a speed perspective, you can generate video on Grok faster than any other platform out there. And the very initial examples that I've seen are showing very solid realism and motion as far as the quality of the video that it generates.

1:08:07I shared with you in the last couple of weeks that Grok4Fast, their smaller, faster model, is currently the most efficient model more or less across the board. So from a speed and cost to value perspective, they are alone in the top right side of the quadrant of being able to deliver value fast and cheap and this is just another example that only on the image and video creation side of the universe will it be able to compete with sora 2 and vo3 and the next generation of tools time will tell but they're definitely doing the right things to place themselves as the most cost effective fast solution in the AI race as of right now.

1:08:48And speaking of VO3, Google just unveiled VO3.1, which is a jump forward from the previous model that is now really, really old, and it was released just a few months ago. So they released two models, VO3.1 and VO3.1 fast, just like with the previous version of just VO3. It is available through the Gemini API, the Google AI Studio, and Vertex AI, and it's also available on the regular apps of Gemini and Flow. It provides enhanced audio and visuals compared to the previous model. It allows to use three reference image characters to create character and style consistency across video and shot. It includes the ability to extend scenes, so you can take an existing video and extend it over time up to a minute or longer by generating new clips to continue the original clip so basically allowing you to create longer and longer videos which is something that most of these tools did not know how to do you had to create the separate cuts and then bring them together in an external editor and now you can create much longer sections and segments that was not doable before definitely not at that quality it creates a smooth transition between a first frame and last frame of a video and knows how to work with it including the audio that has to come with it and they have a few new features that did not exist before that are going back to the tooling side of things with the ability to crop and zoom and do things that were not possible before.

1:10:16And that goes back to what I said many times before. We are getting to the level where these tools are getting to the point that they're all really, really good. And what's going to make the difference is not necessarily the marginal difference in the quality of the output, in this case video, but how easy, effective, and useful the tooling is going to be. Meaning how easy will it be for a user to get the output that they want versus how good the quality is, because the quality is going to be more than acceptable across all these different tools. And these are the first moves by Google to move more in that direction.

1:10:50Now, the cool thing, if you want to learn how to use Google VO3 very, very quickly and get very powerful capabilities, Google also released VO3.1 Prompting Guide. So you can go to the guide, copy and paste example, and use their five-part prompt formula to create incredible outputs that look completely realistic or not realistic, but basically look however you want. And there's going to be a link to that in the show notes and obviously in our newsletter as well. Now, Microsoft is rolling out a big update to the Copilot app on Windows. and the idea is that you will be able to seamlessly produce new documents and so on out of thin air instead of just helping you with editing and stuff like that before.

1:11:40And this includes full Word document, Excel and PowerPoint presentations, as well as later on PDF files directly from chats and prompts, bypassing basically the current apps. So this is released for initial testing. It's still not widespread yet, but this is the direction that it is going. So the dream that I thought will come true sometime in late 2024 might finally happen in 2026, where there's going to be one unified solution for Microsoft and one unified solution for Google versus standalone apps that are somewhat useful on very, very specific use cases, but definitely not as useful as they can be if they can access everything across the board.

1:12:23and I'm really interested to see it finally coming together. Myself, as a Google user, I'm obviously more interested on the Gemini side of the Unified platform, but there's definitely a huge audience for a unified co-pilot that can get to all your data across everything Microsoft and create outputs across the entire Microsoft universe. Hugging Face did an interesting analysis on all the different things that are currently available on Hugging Face. They released that on October 13th, and it includes the platform's 50 most downloaded open source entities that account for over 80 % of all of the downloads on the platform.

1:13:04So the 80-20 rule works, right? It's just 50 entities, but they account for 80 % of the downloads across thousands or tens of thousands of things you can currently download from Hugging Face. So they come from 20 companies, 10 universities, 16 individuals, and they together drove 36.45 billion downloads. Now, small models are crushing it with 92.48 % of overall models are less than 1 billion parameters, which is showing you that while we're all looking and chasing the big models, under the hood, behind the scenes, people are building different applications on open source tools are actually preferring much smaller models because they're fast and cheap to run.

1:13:49Now, from a modality perspective, computer vision is at 21%, audio is at 15%, multimodality, so tools that can do all of them is at 3.3%. I think it's just not as developed on the open source world yet, but this gives you an idea with the obviously on top is text models still. English dominates the overall languages of these different models. US entities from the number of entities are leading, but this month was the first month that Chinese models were downloaded more than U models. For the first time ever, there has been more downloads of open source Chinese models than U.S. models, despite the fact that the U.S.

1:14:28have many more entities on the list, way more than everybody else. What does that tell us? It gives us an idea of what's actually happening behind the scenes when developers want to choose open source models for the different things that they're developing, showing you that there's a very live and healthy open source environment and showing you that China is rising very aggressively on that field on purpose. All the big releases that they have done have been open source and they have placed themselves as the leader in open source model development. And now for the final and really exciting news that I promised you for the end of this episode, which is Google Gemma powered AI tool that was custom created for this purpose, has been able to create a breakthrough in potentially providing a new path for cancer treatment.

1:15:17So I'm not a scientist and I'm definitely not an expert on cancer, but I'll try to explain to you what I understand from reading this paper. So the model is called C2S dash scale 27b model, which is designed for single cell analysis. And what it's trying to do is the following. Apparently there are two types of cancer. There's cold tumors and then there's tumors who are active. And these cold tumors evade detection of our immune system and T cells. So the idea was to come up with different types of chemical compounds that will trigger these tumors in order to being detected by human T cells and hence attacked early in the process and hence potentially helping fight cancer in earlier stages.

1:16:08And what this model was able to do is it was able to identify specific candidates as far as compounds that can do this, checking over 4 ,000 different options and then recommending the ones that are most likely to do that. and the ones that the AI has found have been then tested in an actual lab with an actual cell and was proven to actually work. This is a true new breakthrough in science that makes a huge difference on human life that AI was able to do potentially for the first time ever. Now, is this really going to help us cure cancer better or not? It's unclear. This is one building block out of many others.

1:16:52So there's a lot of other pieces in this puzzle to take this from a concept to an actual solution that can be used. But what is showing that novel scientific discovery can be done either completely by AI or indefinitely with the assistance of AI, which can then help solve some of the biggest problems and the biggest issues that we have in our world today, such as cancer, other disease, global warming, power consumption, et cetera, et cetera, et cetera. So kudos to Google for creating this model and working with scientists to do this. And let's really hope that together we can create a better future with AI.

1:17:28We will be back on Tuesday with another how-to podcast where we will teach you with some of the best experts in the world how to do something that you can do in your business with AI. If you are finding this podcast helpful, please hit the subscribe button so you don't miss any episode. Do it right now. Pull up your phone and do this. And while you're at it, share it with other people. There's a share button on your podcast player. Just click on that and add a few names of people that you know can benefit from this. They will learn a lot. The world has a better chance of being successful with AI, and I will really appreciate it.

1:18:01And until next time we talk, have a great rest of your week.