115 | The truth is dead! Long leave Grok-2, AI usage, investments and ROI reports, Gemini Live release and more AI news from the week ending on August 16

0:00Hello and welcome to a news weekend episode of Leveraging AI, the podcast that shares practical ethical ways to improve efficiency, grow your business, and advance your career. This is Isar, Maitis, your host. And as you can hear, I'm fighting some serious allergies, so I apologize in advance for my stuffy voice. But we have a very interesting week and different than most of the news episodes that I've done. We're going to do a deep dive into something interesting that happened this week. But we're also going to talk a lot about implementation and statistics of what's actually happening with AI usage in the world right now in specific companies, specific states, specific organizations.

0:38There's a lot of that this week, and I'm actually excited about it because it will allow us to talk on broader things and not just very simple, straightforward new releases and stuff like that, but there's a lot of that as well. So let's get started.

0:56In the next few years, AI technology will change our world dramatically. Whether you are a business executive trying to catapult your business forward, or just somebody who refuses to be left behind and want to advance your career, this is the show for you. I'm your host, Isar Matis, a serial entrepreneur and an AI enthusiast. You'll hear invaluable practical tips from innovative business leaders, AI practitioners, and some of the brightest AI minds in our world today on how you can leverage AI in ethical ways to advance your career and grow your business.

1:36the biggest news from my perspective that happened this week is that x.ai the company by elon musk that is developing his ai models just released grok 2 and grok 2 mini that is showing a huge jump forward from grok 1 and grok 1.5 which were released before now grok 1 was a complete joke grok 1.5 was still nothing serious that nobody really paid attention to more than just oh this is a cool fun toy to play with within x previously twitter but the reality is grok 2 is a serious contender to some of the top models so like many of the other companies have done they have released a test version of it into the lmsys chatbot arena i talk about this arena a lot But if you're new to the show, it's a platform that allows people to test their problems against two unknown chats and then score them to which one is better.

2:34They aggregate all that information and they can tell based on real life use cases which chatbots are doing better. And so they released a test version of Grok 2 into that arena that got comparative results to Claude 3.5 Sonnet and GPT-4 Turbo. So very advanced model. So it's not a joke anymore. Now, when the earlier versions were released and people started saying, oh, this is not serious, this is not going to go anywhere. I said what I've learned that to be true most of the times is you never want to bet against Elon Musk, especially when he has grudge driving him. And in this case, there's a lot of grudge and a lot of bad blood between him and OpenAI.

3:17And so him wanting to take his place and claim the leaderboard in the AI world doesn't surprise me at all. And that's before they have started training Grok 3, which is going to be trained on 100 ,000 new NVIDIA GPUs, which was considered to be the most powerful AI training machine ever built. So that's the next model, which is expected to come at the end of this year. But Grok2 is doing very well on key important topics and multiple benchmarks per them. So this is information coming from X themselves. Take it with a grain of sand. And this information comes from X themselves. So you can take whatever you want as far as how.

4:03And this initial feedback on its performance on the different benchmarks come from X itself. It will be a little while before we get third-party feedback on it. But based on that, it's doing very well on graduate level science knowledge, general knowledge, and math competition problems. So a lot of things that these models need to know in order to be better at solving more complex tasks. Both Grok2 and Grok2 Mini are now available to X Premium and Premium Plus users, and enterprise level APIs is coming later this month. So they're definitely developing this as a serious tool and not just an ex-social media platform toy.

4:45But with all this progress, the biggest splash of this release was actually the image generation capabilities. So the previous model was text only, and this new model now enables you to generate images, actually stunning, highly realistic images. But what really made a crazy storm in the last couple of days is the fact that the image generation capabilities is almost not limited without any safeguards or guardrails of what images you can create. So if you remember last week, I told you there's a new contender to the best image generation model in the world, the first one that can actually combat mid-journey so far, and that model is called Flux.1.

5:25It comes from a company called Black Forest Labs in Germany that has released this model recently, and this is the model that was integrated into Grok 2 enabling it to create really amazing images. But because it has almost no safeguards, and when I say almost, it will not create nude images, but other than that, more or less anything else. So within hours, X was filled with crazy, insane photos of anything you can imagine, a huge amount of them with political aspects, showing mostly Trump in really weird situations, some of them aggressive, some of them are just strange, but also other political figures, as well as Elon Musk itself.

6:13And I'm talking about hundreds of thousands of images that were created very quickly. Some of them is very easy to understand that it's a satire or a joke, but many of them, if you wouldn't know, they were created with an AI, it could be sold to you and to me as a true story. And I will probably believe that because they look highly realistic. Now, this obviously raises many big social concerns. And what I did is I went to my favorite social output, which is LinkedIn, where I'm very active. And I wrote a post on Friday morning about it that caught fire. And I want to read the post to you. And then I want to talk to you a little bit about the responses, because in a few hours, there were dozens of responses.

6:57And now it's probably over 80 responses a day later. But more than the fact there were so many comments, the comments were deep and long. I've been posting on LinkedIn for years, almost daily, and I have my fair share of successful posts. I've got dozens or hundreds of comments. I've never received many people who wrote full pages of responses with deep thinking and sharing their emotions or ideas or concepts in such length. And so that's why I think it's important to share that because it goes way beyond just the ability to create images that look realistic. So the post itself, and you can go and look me up on LinkedIn if you haven't so far, you should probably do that.

7:40I share AI content there every single day, usually practical stuff. But in this particular post, I attached three images, one showing Donald Trump kissing a Mexican cartel boss, the other showing Donald Trump in bed in a very laid back and intimate situation with Putin, both smoking cigarettes and wearing pink pajamas. And the third one is Kamala Harris holding two babies. One is obviously Donald Trump and the other, I wasn't 100 % sure, but I think it's King Jumil from North Korea. I'm going to read the post to you and I'm going to talk to you a little bit about what happened. So the post started with this statement.

8:19Donald Trump was caught on camera kissing a Mexican cartel boss, or maybe not. Grok, in parentheses, x.ai model, just released Grok 2, and it introduces access with no filtering to the new flux image generator. It is released just three months before the elections. What could possibly go wrong? On May of 2023, I released an episode on the Leveraging AI podcast called The Truth is Dead, How AI is Putting at Risk the Trust that is the Fabric of Our Society. When people ask me what is my biggest fear of AI, this is it. We have no ability to know what is true and what is not on any digital communication.

9:04Even real-time video conversations with people we know, and 90 % of our communication is digital. We all must be aware of this as it will have a significant impact on our society. I do not have any great ideas on how to change this trajectory. And the only thing I can think of is at least to raise the awareness of this new and strange world that we are walking into. And I'm not even talking about humanoid robots and brain-machine interface chips. What are your thoughts? how can our society overcome the obstacles not being able to know what is true and what is not and as i mentioned there are now over 80 comments of people going in depth into this topic and i want to share with you some of my thoughts on this first of all go and listen to episode 13 of this podcast so if you're listening to this you can just scroll back and find episode 13 and listen to it where I go to this in depth.

10:01But my real biggest fear in AI is what I just mentioned. We have no ability, none whatsoever right now, to know what is true and what is not in any digital communication. Deepfakes today are good enough to do real-time deepfakes, meaning you could be speaking to your spouse, to your kid, to your boss, people that you know very well on any video platform. This could be on social media. This could be on a Zoom call. And it may not be them. There's already been cases of alleged kidnapping that forced people to pay ransom for their kids that were not kidnapped. But it was very easy to make them believe that they were.

10:41There were one event with a financial controller in, I believe, South Korea, in a large international corporation, maybe Singapore, that paid$20 million to another company in a fraudulent process that involved him being on a call with a CFO of his company that he knows well, his boss, and another senior director from his company on a Zoom call that explained why this is urgent and important, which reduced his suspicion from the email he got earlier, and he made the payment. So these things are possible right now. Now, add to that the fact that we have a complete democratization of news and the fact that the younger generation is consuming most of their news, not from news outlets, but from social media, and you have the perfect storm.

11:27Literally anyone can create any piece of news they can think of, good, bad, aligned with whatever agendas and share it across the universe. And we have no clue whether it's true or not. And I know some of you are thinking, yeah, I can still tell the difference. You cannot tell the difference. Add to that the new video models and the abilities to create videos that are highly realistic. And think about a scenario where a few people, or presumably a few people, share video shots from their phone of a specific event in a specific city, in a specific location, from several different directions, sharing the story.

12:05It will look extremely realistic. There will be no way to tell the difference. and it will spread like wildfire and none of it has to actually happen in real life. This is what we're walking into. Now, I'm not an AI doomer. I'm a huge believer in the technology. I think it can completely transform our society in very positive ways, but there are risks that we need to be aware of and we have to find ways to address them. Otherwise, as I mentioned, this puts our society at risk because if you can't know what's true and not true, that raises a lot of other questions on many mechanisms that we take for granted as far as how society operates.

12:45So I'm saying all of this to give you food for thought. If you want to comment on this, please go to this post. As I mentioned, I just posted it yesterday, so it's easy to find. Just go to my profile on LinkedIn and add your comments over there, or just comment to me directly on a direct message on a DM on LinkedIn. I would love to hear your thoughts. And if you have any ideas on how we can address this other than to educate people on what's coming, then please share that with me as well. By the way, on the same topic, I hold AI Friday Hangouts, which is just an open to the public group of people that join every Friday at 1 p.m.

13:19Eastern, and we talk about AI. And in 95 % of the cases, we talk about specific use cases. So people from my community come and they raise questions on stuff they're trying to do, or they share successes of stuff they were able to do this week. And we all learn together how to implement AI successfully and ethically in our businesses. It's really a lot of fun and it's completely informal and just a great way to learn how to use AI. But yesterday, the entire conversation was about this. So obviously, a lot of people have their own opinions and their fears, and it's important to share that information because sharing it will potentially allow us to find ways to reduce the negative impacts of AI and help us benefit more from the positive ones.

14:02By the way, if you want to join those Friday Hangouts, as I mentioned, they're open to the public. All you have to do is find me on LinkedIn and send me a DM saying that you want to join the Friday Hangouts. I will gladly send you an invite. As I mentioned, it's a non-committing. There's no cost to participate. And it's just a great way to learn with other people on how they're implementing AI successfully in their businesses. And now back to the news. Staying on the topic of image generation, a federal judge just allowed several important claims to move forward in a lawsuit. filed against some big image generation and video generation companies, including Stability AI, Me Journey, and others.

14:42Some of the things the judge approved or allowed to move forward on is direct copyright infringement claims against Stability AI, new claims of copyright and trademark infringement claims against Me Journey, copyright claims against Divine Art and Runway, and so on and so forth. Now, some specific allegations were, as an example, Me Journey have their stylist of over 4 ,700 artists where you can say create an image in the style of. So these people are claiming false endorsement, basically using their names to say that it's actually them, quote unquote, creating the thing, creating the output, which is obviously not the case.

15:21They're also claiming that obviously these tools were trained on their copyright work without any permission, and that some of these services also allows to reproduce copyrighted work. The interesting thing about all of this is obviously not the first lawsuit against these companies. This is happening left and right against all of the big engines that generate either text or images or videos because they're all literally stall everything they could in order to use it to train their models. They're claiming fair use, which I'm not a judge. I don't know what's going to be the outcome of this, whether it is fair use or not, but that is not new.

16:01But what new here is that the judge allows this to move forward to trial, meaning there's going to be a discovery phase in which all these companies would have to reveal exactly what data they used and how they used it to train their models. That's information we did not have access to so far, other than rumors and anecdotal proofs from specific companies, but not a wide range of show us exactly how you did it, which is going to be very interesting if it actually gets to that. Now, what might the outcome of that trial be? As I mentioned, I don't have a clue. I'm not a legal person. I said that time and time again on this podcast.

16:43I assume you will end up with some companies writing some very big checks to compensate companies and individuals for whatever damages were presumably caused to them. It also led to the fact that we know a lot of the big players are now signing multiple licensing deals for data with multiple publishers and companies. And the problem, by the way, that I see with what I just said is that it puts the smaller companies that is doing this at a huge disadvantage compared to the big ones. OpenAI, Microsoft and Google have a lot more money to deal with these lawsuits and pay whatever fines is going to be placed on them than let's say Stability AI or Me Journey, which are significantly smaller players, which if these kind of lawsuits move forward, I'm going to put them at a very serious business risk that may force them to shut down their businesses.

17:36Whether I agree or disagree with the claims, or I think they're right or wrong, I'm a very big promoter of competition. And I think having smaller players that can do innovative things like Mejorie, like Stability AI, like Flux is very important to the overall success of any industry and the AI industry. And hence, I would love to see a better solution coming forward rather than pay a huge check that you may or may not be able to pay because that may leave even more power in the hands of the biggest companies in the world today anyway. Now, speaking of image and video generation, Runway, the company that has gave us Runway Gen3, which is the most advanced video generation tool we have access to right now.

18:22So we've seen Sora and we've seen the tool from Google. None of them is available right now. So there's a bunch of tools that are out there that are actually pretty good. And Runway Gen3 is amazing. they just released Gen 3 Alpha Turbo, which improves the speed in which you can generate videos. It's seven times faster and 50 % cheaper than Gen 3 Alpha that was just recently released. Now, it's available right now. You can start using it today across all their subscription plans, including their free trial capabilities. And it's absolutely stunning. It's almost near real-time video production, or like their CEO said, it's actually faster to generate the video than to type the prompt that generates the video.

19:07And the amazing thing is that it keeps the same level of quality and performance as the regular model. So when I'm saying it's 50 % cheaper, to put this in specific context, it costs five credits to generate one second of video, and each credit is one cent. So basically every second of video is going to cost you five cents generating videos at crazy high speeds. If you haven't seen demos of this, just go and check it out. And this is the direction that everything is going, right? It's going to get better and better, cheaper and faster across all the different aspects of AI to the point, as I mentioned in several different episodes, that I'm not sure how the business model of this is going to work because if it's going to cost almost nothing, basically free to get really good models.

19:54And yes, it's going to be more expensive to get better models, but you may not need the better models because the free or almost free models are going to be good enough for 95 % of the things that we do. I'm not sure what's going to be the business use case for all of that, but it will be interesting to continue watching that evolve. And since we're talking about competitiveness in the AI market, OpenAI released two new models in the past week without really, again, any PR or sharing it with the world exactly what they're doing, but they have made two interesting releases. One is a release of a new variation.

20:29Both of them are variations of GPT 4.0. One of them is called 2024-0808. That's the date it was released. And that model actually overtook the top spot of the LMSYS AI chatbot arena leaderboard, meaning per actual users with actual use cases, it is right now the best model available out there for real-world usage. It kicked down Gemini 1.5 Pro EXP0801 that was just released a week ago to the second spot. Again, there were no real release notes or anything from OpenAI stating exactly what this model does, but users are claiming that they're seeing big improvement on technical domains, especially writing code and better performance in instruction following at complex prompts and tasks.

21:22Users are also claiming that it's working significantly faster, with several users claiming that they can now develop complete iOS apps within one hour of writing code with this new tool. So what does that tell us on a broader scale? There's a concept in the software world called CICD, Continuous Integration, Continuous Deployment, which basically means every time you have a new feature that was tested, you release it without a formal big release of the software. That is not new. I've done this in my software companies several years ago, so that's not new. But what's new here is the fact that it's done without any documentation.

21:55So usually when you release new features of a software, and a lot of software that you're using today is probably doing it, you get an email and you get a note on the app itself or the software itself that's telling you what has changed and how to use the new thing and what functionality there is and so on. And in the AI world, it just doesn't exist. They just release new stuff. And they do this for several different purposes. One is they can learn how the thing works out in the real world versus in their lab with their QA teams. So that's a huge benefit to the companies. Two, we, as the people are using it, are getting the benefits of more advanced features almost every single day, but definitely every single week.

22:32So we gain the benefit of all of that. The disadvantage is obviously a complete chaos of not knowing which models you're actually using, which ones you should use. If you want to do like a serious implementation on an enterprise level, you cannot keep on changing models every single week or every single day, especially without any documentation that explains to you exactly what's changing. So from a deployment and software control management, it's a complete chaos. And the third question is, if they are releasing these models so quickly every single week, Like, how much are they really paying attention to safety of how secure these models are?

23:09Sadly, I think we know the answer, but that's, I think, where we are. Like, I don't see that changing. I don't see that going away. I think the competition is so fierce. That's what these companies are going to do. And that's their way to actually make their path to the next level. Meaning GPT-5 components of it is probably already out there and available to us through various components that OpenAI has already released. And the same thing with Gemini 2, and the same thing with Cloud 4, etc. They're going to keep on releasing small capabilities for us to test. So we basically become the QA lab for these companies out in the real world.

23:48And this is working so far, but it may have a very serious backlash sometime in the future. Now, as I mentioned, OpenAI launched another model two days earlier called GPT-40-2024-0806. So it's two days before, but this model actually has a very specific reason. This model was developed specifically for computer developers, and the goal of this model is that it's very good in creating structured outputs such as JSON schema. So if you're writing code and you need to generate very specific structured outputs, this is the model for you. Now, these two models are released, as I mentioned last week, there was serious turmoil in the leadership with John Shulman, one of the co-founders, leaving to a man, Greg Brockman taking a long leave of absence, and obviously a lot of other people that are left before, including Ilya Suskover and Andrzej Karpathy and some other big names from the AI world that has left OpenAI.

24:48But it seems that it had no impact on the way they work and the speed that they're working, and so far, the quality of the stuff that they're releasing to the world. Now, going back and combining all of this of fast releases and adoption and so on, as I mentioned, we're going to talk a lot about adoption and use cases of AI in the world. There was a very interesting research released by two PhD candidates at MIT and Harvard that did a deep research on how AI is actually used in the world. So I'm going to start by quoting them and then I'm going to dive into the details of what they found. We're seeing a giant real-world experiment unfold, and it's still uncertain what impact these AI companions will have either on us individually or on a society as a whole, argues Robert Mahari, a joint JD-PhD candidate at MIT Media Lab and Harvard Law School, and Pat Pataranutapron, a researcher from the MIT Media Lab.

25:46They say that we need to prepare for an addictive intelligence or AI companions that have dark patterns built into them to get us hooked. And they look on how smart regulation can help us prevent some of these risks associated with AI chatbots that get deep inside our heads. End of quote. What did they find in the research? They found that people are forming actual relationships with AI systems, using them as companions, mentors, and more. They also found, not surprisingly, that sexual role-playing and creative tasks are among the top popular uses of chatbots. If you ever looked at the statistics on how much of the internet as a whole is dedicated into porn, this will not surprise you.

26:36What I mean by that is that the latest statistics, and there are multiple sources, and I don't know how many of them are accurate, but they all land on roughly the same number, that 30 % of the total data transferred across the internet as a whole is porn and sex-oriented. So the fact that this is one of the main usages of AI is not surprising. They're also saying something that we know right now, that productivity gains that are promised by AI is still largely not materialized, even though I'm going to share with you another article that may claim otherwise. As far as the thing they found that people actually use it very successfully for is using AI chatbots for brainstorming, planning, and general information queries.

27:15I say that a lot when I talk to people and when I speak on stages and when I teach my courses. It is the most amazing brainstorming and ideation tool we ever had. And it's way more powerful in that than generating content, which is the first thing people jump into. So the ability to use AI in data analysis and strategy and brainstorming ideas and figuring out processes that has significantly higher ROI to your business than creating content is probably the first place you should go, even though most people go to content creation. But the point that they're making is that we need some serious thinking into the kind of regulation and training that we need to put in place in order to, again, avoid some negative aspects.

28:01And in this case, they're talking about literal mind manipulation and social manipulation, either by mistake or by bad actors who just want to use these capabilities to their benefit. Now, on the flip side, the recently formed AI-enabled ICT workforce consortium, which is a consortium that was established by some major tech companies, including Cisco and Google and IBM and Microsoft and some other really big players, has released their report on the transformational opportunity of AI on IT jobs. Their finding is aligned with what I thought, just the numbers are way higher than I expected. So they're claiming that based on the research, 92 % of IT jobs will undergo high or moderate transformation due to AI advancement.