137 | GPTsearch and Meta are going after Google, Apple Intelligence finally released, a new king of image generation and many more AI news for the week ending on November 1st

0:00Hello and welcome to a weekend news episode of the Leveraging AI podcast, the podcast that shares practical ethical ways to leverage AI to improve efficiency, grow your business, grow your business, and advance your career. This is Isar Maitis, your host, and we have another week with a lot of interesting news. If last week we discussed the release of, I don't know, like eight or nine different new models, today there's less releases of models, but a lot of new capabilities that are all really important. And in the end, we're also going to talk about one new model that is an image generator that came out of nowhere and took the lead in image generation by a big spread.

0:36So let's get started.

0:43The first and biggest piece of news of this week is OpenAI just released Search GPT to everyone. Rumors about OpenAI releasing a search engine started popping out early this summer, and then they released a version of it to a small audience to get tested. And now this week, it became available to everyone with a paid subscription. Now, those of you who have been following what's happening with AI-assisted search, Google introduced something like this a while back, and it wasn't great, and they took it back, and then they brought a different version of it again. In parallel, Perplexity, which is the company that probably figured it out the best way, is growing very, very fast with more and more users every single day, which forces Google, obviously, to get their act together as far as the search.

1:31But Google didn't make significant changes. You still have the old traditional Google search just with a snippet on top that has the AI summary, while Perplexity and now OpenAI are providing you just a summary with the option to see the sources on the side. OpenAI took a similar approach to Perplexity where the main screen shows you the results, while you can choose to open another kind of like tab on the right side that shows you the sources where the information comes from so you can verify the information is accurate. Some of the key features of the new OpenAI capability, it has an interactive stock graph with real-time market data.

2:12It has location-based search with interactive maps. It has clickable citations that you can go and see the different sources. And it has a source sidebar, as I mentioned before. And it is available on the desktop as well as the Android and iOS apps. it is built into the regular chat GPT. So you don't have to open a different application or anything. And it's pulling information from multiple sources, many of them that we spoke about in the past few months of the licensing deals that OpenAI has signed with Hearst and Condé Nast and Axel Springer and News Corp and a bunch of other publishers that are providing them access to real-time news and information.

2:54Now, currently there are no ads and there's no clear business model to this, other than the existing paid licensing that anybody can pay to use ChatGPT, pre-users will have limits on amount of searches that they can do. And where that's going to go in the long run, as far as the business model is obviously unclear yet, but the trophy is very, very obvious. And the risk for Google is also very, very obvious because Google reported$49.4 billion in search revenue in just Q3 alone. So the amount of money that is at risk or the pieces of cake that can be distributed differently are very significant.

3:34I do not know if OpenAI is going to go into an ad model or something else in order to monetize this, but for now, it's definitely a risk for Google that will have to push Google to do more with AI in their searches. We're going to talk more about Google later on in this episode. Another interesting thing that OpenAI released with this new capability is a new Chrome extension for this tool. And if you install this Chrome extension, it makes ChatGPT the default search engine on Chrome replacing Google. I installed it and it's actually working really, really well. When you're typing something in the top bar of Chrome, it actually opens OpenAI and runs the search on ChatGPT and shows you the results.

4:20I must admit that after testing it for a couple of days right now, I still prefer, as of right now, the Google search because there's some things it just does better. But I'm definitely going to continue using both, just like I've been using Perplexity more and more. And so I will report as time goes by my preferences. Right now, I still do most of my searches on perplexity, then Google, and now OpenAI as well. This may change as I see what provides me better and more accurate and more helpful results. In parallel, Meta, the company behind Facebook and Instagram and so on, has announced that they are developing their own web crawling and search engine that will allow them to provide real-time answers for their AI models to replace Google Search and Microsoft Bing, which are driving these capabilities right now.

5:12So a little bit of background, Meta released Meta AI capabilities all as open source for the past couple of years. They've been releasing more and more capabilities that are more and more advanced, but they're also having the AI functionality running natively in Facebook and Instagram and WhatsApp and Messenger, which means right now, because people are using it as part of the other tools that they're already using regularly, they're claiming they have 185 million weekly active users on their AI tools. That's just not too far behind ChatGPT's 250 million weekly active users. I don't think it's a fair comparison because as I mentioned, I think a lot of these people do not even know they're using the meta AI tools, but that's the beauty of it, right?

5:55They've integrated it into their existing distribution seamlessly, and AI is just providing some of the functionality, but they're definitely growing and they're having more capabilities. Meta is known for having issues with being dependent on other platforms. Their most known issue is obviously with Apple for the amounts of money that they have to pay the App Store through any sales that happen on their platforms or any interactions that happen on their platforms, and they're not happy about it, and they were trying to fight it several times in the past. So now they are trying to potentially get away from the shackles of Google and Bing when it comes to providing real-time information to their platforms.

6:33And if they get it right, they will be able to do a lot more with real-time searches, which is another risk to Google's search. Because a lot of the younger generation spends significantly more time on Instagram and on Messenger than they spend on Google. And if they'll be able to search the web and get answers to anything they want within Instagram, I think they will do that. That's obviously my personal opinion. But based on the amount of money and efforts that they're putting into this at Meta, it seems that they believe that that's a potential outcome. So the team now has been working on this for eight months.

7:12It's been led by a senior engineer that has been Meta for many years. And as we shared recently, they just signed a deal with Reuters to get feed to their live event coverage. And so this will be another interesting aspect of the search game that was so far completely held by Google. And now there's more and more competition from different sources. But going back to OpenAI, another small search capability actually happens within ChatGPT. And now there's a little magnifying glass on top of the regular ChatGPT interface that allows you to search your history of chats. So if you're like me and you're using it regularly, you have X number of chats per day.

7:53This could be six to 20 for me, which means there's thousands of them in the past two years. And this is a helpful tool to go and find older chats and either continue them or find the information that was in them. I started using it. I must admit, it's not working great. It's still not sometimes finding me exactly the chats that I'm looking for. And there's some specific chats I like to go back to because they're yielding specific results or they're a part of demos that I'm doing when I speak on stages or as part of my courses. And it's still not finding them in some cases. So this is a very helpful functionality.

8:25I'm very happy that it's there, but it's still not working perfect as of now. In an interesting article on ZDNet this week, there were some interesting metrics on the current status of OpenAI. So they have 250 million weekly active users, 5 % to 6 % read to paid conversion rates. So I don't know if that's high or low, but that's not bad. If you're thinking about the pool of 250 million active users, they have 1 million enterprise or team paying users, which is significant. And 75 % of the revenue comes from consumer subscriptions and not from the enterprise subscription. And I assumed with all the money they invested in going to the enterprise realm that the percentage of enterprise revenue is going to be higher than 25%, but that's not the case right now.

9:16Now, I would like to give some of my personal opinion to what's happening with ChatGPT in general. When people ask me what tool they should pay for if they don't have the money and they want to commit only to one tool, ChatGPT is an obvious winner. Now, to be fair, I pay all of them. I have Perplexity Pro and I pay for Gemini and I pay for Claude and I pay for ChatGPT, but that's what I do for my business, right? I help companies implement this. I need to know exactly how these tools work. And yes, there are some things that Gemini does better. There's something that Claude does better. There's some things that Perplexity does better.

9:51But if you look at one tool that does some of the things better than all the others, but the combination of capabilities is by far better than everybody else, ChatGPT is a very, very obvious winner. So if you add to that the fact that now they have Canvas that in many things does a much better job than Claude Artifacts, definitely means of data analysis and presentation of data, it's doing a better job. And now with the addition of Search, they're opening more and more gaps on more and more aspects that provides as a one tool that does it all, just a better tool than any other tool out there. So if you are one of those people that is currently not paying for any of the tools and you're sitting on the fence and you're not sure which one you want to use, right now, ChatGPT is a very obvious winner.

10:38And in another very interesting piece of news from this week that has to do with changing world orders, in addition to Search, OpenAI just announced that they're starting to develop its own in-house AI inference chip in a partnership with Broadcom and TSMC. So Broadcom is going to help in chip design and optimization and TSMC is going to be the manufacturing capacity. They also announced that they're adding AMD chips to assist in some of the aspects that they're doing with NVIDIA while they're maintaining their strong relationship with NVIDIA. So it's a new department within OpenAI that is going to work on chip development.

11:17There are two different kinds of operations when it comes to large language models or AI in general. There's the training side in which NVIDIA GPUs is still a very dominant player. And there's more and more new companies and new chip capabilities that are focusing on inference, which is the generation of tokens, basically us using the model versus the companies who develop the models training the models. And in there, there's some very big success stories. The one that we talked about many times is Grok with a Q, G-R-O-Q, that developed what they're calling an LPU, a language processing unit versus a GPU of a graphics processing unit, which is the chips that are sold by NVIDIA.

12:00And they're able to generate inference significantly faster. Now, if you connect the dots together, OpenAI is projected to lose$5 billion in 2024 on a$3.7 billion revenue. The biggest expense that they have out of that$5 billion is compute, meaning they're spending a lot of money on other people's computing power. And if they're developing their own inference chip, they can focus their expense on compute on something that they control that is presumably, or if it works, runs better, faster, and cheaper for them that can save them significant expenses moving forward. And what I project, and again, I think a lot of people would agree with that, as these models get better and better, we'll get to the point that we won't need the best frontier models for most tasks.

12:52What we will need is to run older models in the most efficient way. And if you control the entire ecosystem of hardware, software, and algorithms, you can be significantly more competitive in this new world. The production of these chips is supposed to start in 2026. So about a year and a little bit out and probably installation of them, probably at the later part of 2026. And so this is not immediately, but it's definitely in the future. And I definitely understand the move that OpenAI is making in this direction. Now, we talked about Meta earlier because we were talking about the topic of search, but going back to Meta, they also made another interesting release this week and they released models that are based on their existing models that are just faster and smaller and are designed to run on low-power devices, basically to run on our phones and tablets and potentially watches.

13:49And these new models are 56 % reduction in size compared to the previous model size. They have a 41 % decrease in memory usage on Android devices. They run two to four times faster on inference and they're per meta maintaining the performance comparable with the larger models. The only big limitation of these is that they support a context window of only up to 8 ,000 tokens, which is very small compared to what we used to. But if you're thinking about the use cases, which are the things that you want to run on your device, which are usually short and quick queries, this should be more than enough.

14:28This is the next frontier, right? Is the ability to run your queries on device. And it gives two huge benefits. One is data privacy, right? Because your information is not going anywhere. It's staying on your phone or on wearables, which is the next thing that is coming. But the other benefit is obviously speed because you don't have to relay the information to a server and then back to the device. So you win on both these aspects. And these new models are available both on Lama's website as well as on Hugging Face and they're open source, just like all the other stuff that Meta is doing. So anybody can use them to develop on-device solutions.

15:04We're going to talk more about on-device AI capabilities once we talk about Apple later in this episode. We have been talking a lot on this podcast on the importance of AI education and literacy for people in businesses. It is literally the number one factor of success versus failure when implementing AI in the business. It's actually not the tech. It's the ability to train people and get them to the level of knowledge they need in order to use AI in specific use cases successfully, hence generating positive ROI. The biggest question is how do you train yourself if you're the business person or people in your team, in your company, in the most effective way?

15:47I have two pieces of very exciting news for you. Number one is that I have been teaching the AI Business Transformation course since April of last year. I have been teaching it two times a month, every month since the beginning of the year and once a month, all of last year, hundreds of business people and businesses are transforming their way they're doing business because based on the information they've learned in this course. I mostly teach this course privately, meaning organizations and companies hire me to teach just their people. And about once a quarter, we do a publicly available course.

16:27Well, this once a quarter is happening again. So on November 18th of this month, we are opening another course to the public where anyone can join the courses, four sessions online, two hours each. So four weeks, two hours every single week with me live as an instructor with one hour a week in addition for you to come and ask questions in between based on the homework or things you learned or things you didn't understand. It's a very detailed, comprehensive course that will take you from wherever you are in your journey right now to a level where you understand what this technology can do for your business across multiple aspects and departments, including a detailed blueprint of how to move forward and implement this from a company-wide perspective.

17:16So if you are looking to dramatically impact the way you are using AI or your company or your department is using AI, this is an amazing opportunity for you to accelerate your knowledge and start implementing AI in everything you're doing in your business. You can find the link in the show notes. Just open your phone right now, find the link to the course, click on it, and you can sign up right now. And now back to the episode. Now, in general, Meta is going all in on AI. We talked about their new search, but they have made some serious announcements about stuff that they're developing and haven't released yet.

17:55And we talked about some of them in the past few weeks, like the movie gen video creation and new AI based advertising feature features and spirit LM for emotional voice generation. So there's a lot of things that are either partially released or not released yet that are going to become a part of Meta's AI ecosystem. And one of the interesting things about Meta is they're one of the only, maybe the only player in this industry that does not need revenue coming in from AI. Hence, they can release really advanced capabilities as open source just to reduce the competitiveness of their competitors like OpenAI and Anthropic.

18:32And they can release very powerful capabilities that they integrate into other things that is generating revenue for them. I think we're going to continue seeing MetaPush in that direction and taking bigger and bigger part of the AI universe while monetizing it through their existing ecosystem. Another great example of them going after their competition directly, but with open source capabilities is that they just released Notebook Llama, which is an open source alternative to Google's Notebook LM that we talked about in several different episodes, including one dedicated episode where I explain exactly how this can be used.

19:07But those of you who don't know, Notebook LM is a very powerful Google tool that existed for a while, but then caught like wildfire when they released a feature in it that allows you to generate mini podcasts from information you upload to it. So Meta just released a similar tool that allows you to upload files and links such as articles and PDFs and create an interaction of a podcast of two people speaking to one another. Now, I must admit, I listened to some of these examples. Their text-to-speech models are definitely not as good as Google's as of right now. It sounds way more robotic than Google's Notebook LM voices, but it's definitely a move in the right direction.

19:48There's also an open source model, not from Lama, called OpenNotebook LM that does something very similar. So while it was very cool when Google released it, now there's other functionality. I must admit that I'm very happy with the Google tool and I'm not interested in looking at other ones unless they're going to provide some significant additional value, which right now they are not. And the Google tool is still free. So there's really no reason for me to switch to something else. But this is just my personal opinion. Now, we already mentioned Google a lot as far as competition to Google from multiple sources.

20:20So let's talk about Google themselves. So a lot of information is being leaked new about Google's next release. And it is now mentioned that Google is creating an agent platform called Project Jarvis that will control the Chrome web browser on your behalf and complete tasks autonomously. So if you remember last week, we talked in depth about the new release from Anthropic of a cloud function that can take over the computer as a whole and do things for you on anything that computers have access to, including local software. This is Google's variation of this that is supposed to be released in December with the new Gemini models, which will be Gemini 2.0.

21:00And in it, it will be able to control everything within the browser. So from a security benefit, it reduces the risks because it can access stuff outside of the browser in your computer. But if you think about the browser and specifically Chrome, it has access to your passwords in many cases that can access anything you can access to. So there's still a lot of risks. But as I mentioned, this new Project Jarvis is planned to be released in December. As more information comes out, we will share it with you. But the use cases that have been discussed are research gathering. So obviously going to multiple sources to find information on your behalf and summarize it for you.

21:40Product purchasing. So go and find the cheapest X based on whatever characteristics. Then you will look through multiple sources. We'll provide you options and can actually complete the purchase. Booking flights and travel. Processing returns of different items that you bought on different platforms. And other everyday tasks that we do on our browsers. You'll be able to basically tell it what you want it to do. and he will go and do it for you, which will be very useful. This goes back to the same concerns I talked about last week. And every time I talk about these agents, A, these tools are not consistent right now.

22:11They do weird things and they don't always complete the task. B, there's security concerns, as I mentioned. And C, I think the biggest information will be trust. When will we feel comfortable enough to give these tools real access to do stuff on our behalf, knowing that they will complete the task the way we wanted it. And if it's not possible, they will raise a flag and stop and tell us what's happening and ask for permission or additional guidance. I think we're still months away from that, but it's definitely the direction that everybody's going. And hence, I see this as an inevitable future where these AI agents will act on our behalf across multiple aspects, both personal and business.

22:52Now, another interesting piece from Google and AI-related topics, Google reports that AI is now generating between 25 % to 30 % of the code that Google generates. Now, I don't know how many software engineers Google has today, but it has to be a really large number. And they're generating a lot of code. And if 25 % to 30 % of that is generated with AI, it's a huge amount of code that they're generating with AI. Now, this came as part of their quarterly reports. They are also reporting that their cloud revenue has reached$11.4 billion, which is a 35 % increase year over year. Search revenue has grown as well with 12.3 % growth year over year, and the stock is up this year almost 30%.

23:37percent. So while I see clouds in Google's future when it comes to this revenue, because there's going to be more competition on search and on cloud and on other aspects, right now they're doing very well with leveraging the AI capabilities. Now, if you remember, I said that multiple times before, I think Google is probably the best position company to win this overall game because they have everything that is required. They have the talent, they have the compute, they have unlimited amount of money and resources to invest in this. They have DeepMind, which is probably the best development lab out there right now, but if not, definitely one of the top three.

24:22And they have the distribution and the data. So they have more stuff than anybody else. They are obviously also banking on their full stack approach to this, that it integrates to all their existing services, and they're growing their enterprise customer usage of AI. I'm using Gemini within the Google suite of tools more and more every single day. And if you want to learn on ways of how you can do that, just go and check episode 126 of this podcast, where we dove deeply into how you can use Gemini within Google Docs, Gmail, and other Google suite tools. Now, I mentioned briefly earlier that Gemini 2.0 is also planned to be released in December of 2024.

25:06And there haven't been enough leaked rumors to know what Gemini 2.0 will do different than Gemini 1.5. Right now, the two biggest benefits of Gemini 1.5 is that it has the largest context window by far compared to the competition with a very accurate retrieval percentage from that. So if you're asking questions about the data that you provided, you're getting a 97 % accuracy in the answers based on the information, which is higher than all the other competitors right now. What's going to be the difference to 2.0? Nobody knows exactly. But the interesting thing about this is that it's roughly a year out from the release of Gemini 1, which was released in December of 2023.

25:46In the middle, but really early in the year, in February, they released Gemini 1.5 and then they released Gemini 1.5 Pro in May. And the interesting thing about this is this starts to resemble more and more a normal cycle of version releases of traditional software versus the complete madness we know from the AI world so far, which means from an enterprise perspective, it gives more time for planning and deployment and so on as a new version comes out, a big version once a year and a small version once every six months. I think this is a lot more attainable and controllable if you're running an IT department in an organization, and it will really allow companies to implement these AI capabilities in a much more structured way.

26:31Now, we spoke about OpenAI. We spoke about Google. We spoke about Meta. The other really big piece of news this week is from Apple that has finally released its Apple Intelligence AI capabilities. It is coming through iOS 18.1 and iPadOS 18.1 and MacOS Sequoia 15.1, which has been anticipated for a very long time. The capabilities that are released right now is writing tools that are going to be available through everything writing. So system-wide AI writing assistant, enhanced Siri. So being able to ask Siri more things through a more natural conversation and get answers that will just improve the context and understanding of Siri to what you're asking it and be able to provide better answers.