55 | Microsoft introduces Copilot-Pro, Sam Altman shares information about GPT-5 and more exciting AI news from the week ending on 1/20/24

0:00Hello and welcome to a short weekend news edition of the Leveraging AI podcast. This is Isar Maitis, your host. And like every week, there is a lot to dive into. A lot of big news happened this past week from OpenAI and Microsoft and a few other interesting things. So let's get started.

0:22There were two interesting interviews with Sam Altman, the CEO of ChatGPT, this past week or so. One of them, he was a guest on Bill Gates' new podcast, and they had a very interesting conversation about the future of AI and what's happening at OpenAI. If you haven't listened to it, it's definitely worth listening to. And the other interview was with Axios at the World Economic Forum in Davos. In these two interviews combined, he touched on several very important points that I would love to summarize for you. One is that obviously the company is fully focused on developing GPT-5. The biggest differences in GPT-5 compared to GPT-4 based on SAM is that it's going to have significantly improved reasoning capabilities, significantly improved accuracy, meaning a lot less hallucinations, and most likely better image generation.

1:10And he hinted to video support. Multi-modality is obviously continuing to be a big deal, especially with Gemini built that way from the ground up. GPT-5 will be the same way, and they've announced that a long time ago. So image generation, code generation, video, et cetera, all are going to be baked into this one model. He also mentioned that you will have significantly better customization, meaning the ability to tailor it to specific needs of specific users. And this will come in two very different flavors. One is that they're saying it will adapt to the culture preferences of specific individuals.

1:50And as Sam said, this might make some people uncomfortable, meaning it may give different answers to different people based on their preferences. I understand the reason to do this because different cultures around the world view things very differently. I still think it's very problematic because there might be cases where the model will provide answers that may be, quote unquote, unacceptable in many cultures, but is acceptable in some cultures. and also it will, in my eyes, increase polarization in our modern society instead of finding ways of bringing us together because it will highlight to each individual the things they believe in versus driving to some kind of a global unifying truth.

2:35The other aspect where ChatGPT will be customizable and will provide a lot more value is that it will connect to our day-to-day processes like personal data and emails and calendars, and will allow to provide us personal assistance, such as summarizing the most important things we need to do or our most important emails, et cetera. So the direction is very clear that GPT-5 and future AI models are going to move from just knowledge to taking actions on our behalf and actually assisting us on our day-to-day, both in our personal and professional lives. The last thing that I said is that GPT-5 is not expected in 2024.

3:14While we can never tell if that's really what's going to happen, at least that's the current expectation coming straight from Sam Altman. Last week, I told you that OpenAI announced a Teams level for ChatGPT. I gave you very basic information because it was not available, but the two most major differences between the GPT Plus and the Teams function is that the Teams function, first of all, is built for Teams, meaning there's an admin level that can assign different licenses with different capabilities to different people in the organization. But the most important aspect of the Teams level is that it does not train the model on the data that you provided.

3:54In other words, maybe the biggest pushback that people had in organizations to using ChatGPT was the fact that they don't want to give OpenAI access to their data. And now with the Teams level, you can do that by adding only$10 a month per person. I think that's a no-brainer. And I think this will dramatically increase adoption across small and medium businesses and potentially even large businesses. But these will most likely go to the enterprise version that will provide them some additional benefits. Another thing that OpenAI announced is that they're putting together a team that will crowdsource governance ideas to its future models.

4:33whether that's driven by the whole issue they had with leadership and the direction that they're taking end of last year, or by the quote unquote scary steps that they're making towards AGI internally that they're not sharing with everybody. They are asking the support of external research group or anybody who's interested to provide them inputs on how to have better governance on future models across the board. The group plans to run technical workshops where researchers from different sources can present proposals on how to better govern these future AI models. I think that's a great step in the right direction to provide more transparency and involvement and harness ideas from people outside of open AI on how to control the most advanced models that are being pushed out to the world today.

5:26Another interesting and positive announcement from OpenAI comes from the fact that they're going to have their first partnership with a university. Arizona State University, ASU, is going to have full access to GPT Enterprise starting in February. They will develop actual curriculums and an AI tutor for students across multiple subjects, especially in STEM, as well as avatars that are going to be called study buddies to help their students learn better, faster and more personalized using ChatGPT capabilities. Since they're going to be using the enterprise level, the data security stays in place, meaning no personal information of students or people who are developing this, as well as university data, is going to be shared with OpenAI.

6:12I see this as a huge step in the right direction. I think artificial intelligence represents the most amazing opportunity for education in general in the world today. And it will allow us, hopefully, to go from teaching the same way we were teaching 40 years ago to the future where each person can learn in her or his own pace on specific subjects in the specific method that works best for them, whether it's through graphics, watching videos, tutorials, reading information, etc. etc., the training can be customized per individual to drive them to the best, most effective learning possible. And I love the fact that OpenAI is taking this initiative, and I hope we'll see a lot more of that coming soon, and then trickling down to other layers of education, including high schools and middle schools, and before that as well.

7:02One last small but interesting piece of news from OpenAI is that they're quietly removed the ban from military usage of its AI tools. Right now, they only remove the bans from defensive kind of AI applications, such as cybersecurity, while keeping in place the ban on weapon development and surveillance. I think that's a very slippery slope, and I have zero doubt that it's actually going to happen, whether it will come from OpenAI or other commercial sources, or the government through DARPA or any other development groups will develop its own AI models. There's no doubt in my mind that the future war will include a lot of AI capabilities.

7:43It already does today. It was just going to be in a much more advanced and sophisticated and hence scary level. And from OpenAI to their big partner, Microsoft, I told you last week that I was very surprised to see research that's showing that despite the fact that Microsoft provides all of ChatGPT plus capabilities for free, it did not damage the installations and sales of OpenAI. And I told you that in the long run, I'll be very surprised if that's not going to change. Well, it changed. I don't know if Microsoft or OpenAI people are listening to this podcast and making decisions based on that.

8:19Obviously not. I'm just kidding. But Microsoft just announced that they are launching a Copilot Pro level that's going to cost 20 bucks a month. And only there you will get all the most advanced capabilities from chat GPT. The biggest differences between the free copilot version and the paid copilot version is that in the paid copilot version, you're going to get access to GPT-4 Turbo, which is the most advanced model from OpenAI. You will get priority access to the latest AI models, including on top peak times, you will get faster performance on rendering images with Tali, what they call boosts, and most importantly, the copilot capabilities in the pro version will be integrated into Microsoft 365 apps that will help you draft documents, make presentations, analyze data, and so on.

9:08So really to make the best out of it within the Microsoft ecosystem, I assume most users will prefer to use the paid version versus the free version because we'll give you significantly better results, especially when it comes to business productivity. That is obviously not a big surprise. The gap between these capabilities being free in Microsoft and paid in OpenAI, it made absolutely no sense. I think this makes a lot more sense moving forward, and it will start paying back to Microsoft on the huge investment they made in OpenAI. In parallel to this announcement, Microsoft also announced that co-pilot for Microsoft 365 that was currently only available for larger organizations that used at least 300 licenses is now available to any person with a business license of Microsoft.

9:53And to make it more specific, they basically said that any commercial user, even using older Office 365 capabilities like E3 and E5, and even a single user will be able to start using Copilot. That's obviously huge news because this was so far kept only for a small number of very large companies, among them really big names like Visa and BP and Honda and Pfizer, Accenture, KPMG, PwC, and a few other big names. But now, if you are a Microsoft 365 user, you can start using Copilot. And as I mentioned, in two different levels, either free or for 20 bucks a month with a lot of added features. And for these two huge commercial players, the next piece of news comes from Mistral.

10:39We talked about them several times in previous shows. Mistral is a French company that is delivering extremely powerful open source models. They've just released what they call Mixtral 8x7b, which is their latest model that they are claiming is better than GPT-3.5, which makes it the most powerful open source model in the world, better than Lama that is coming from Meta. Mixtral uses a mixture of expert architecture similar to what GPT-4 is using. And as I mentioned, this is an open source model that is currently released as an alpha, but probably very shortly will be available to anybody who wants to use it.

11:19And since I mentioned Meta, let's talk about them a little bit. Mark Zuckerberg gave a very interesting statement this week, stating that Meta's goal is to achieve AGI. They officially joined the race. They restructured their AI activities within Meta in order to achieve that. That announcement obviously is there to also help them attract top AI researchers that currently in Meta, he's claiming, are making more than a million dollars a year. So knowing that it's not just a huge amount of money that they're going to make, but also the ambition to actually achieve AGI, hopefully will lure the right people in.

11:56In addition to that, Meta has over 340 ,000 NVIDIA H100 GPUs, which if you don't know, is the processors in which those neural networks are trained and running, which is as far as I know, the largest number of these GPUs of any company in the world. So there's a few very interesting things. In addition, there's obviously Yalna Kuhn, who is the person that's running their AI development, who's one of the top minds in that field for several years. So there's many reasons for AI researchers to jump ship and join Meta. While this may sound like a very serious pivot from Meta's previous focus, which was the Metaverse, which is what gave it its name, Meta, Mark Zuckerberg insists that's not the case, that they're still spending over$15 billion per year on the Metaverse, and the AI capabilities are only going to enhance Metaverse capabilities, enhance the additional investment in that direction.

12:52Whether that's accurate or not, only time will tell. I shared with you last week about the news about Rabbits R1, conversational AI personal assistant that sold better than cookies at the CES show. The craze continues. They already sold over 50 ,000 of these devices. And I don't know if they've even announced how long it's going to take them to make 50 ,000 devices, but they already sold them. But their announcement from this week is very interesting. and they are going to combine forces with Perplexity AI. Perplexity is a great research engine. It's as if Google and ChatGPT had a baby that would be Perplexity AI.

13:29I use it all the time now for research. It's a great tool. And the partnership with Rabbit will allow R1 users to get real-time information and answers to any questions using Perplexity's engine. The interesting thing is for the$200 you're paying for the device, you are going to get a year worth of subscription to Perplexity Pro, which is worth 20 bucks a month. So if you really like Perplexity, you might as well buy R1 and you'll get Perplexity for a year for free. While this is great news and it's definitely an important addition to the R1, I still stand with my thoughts from last week that I think this is not going to be long-lived.

14:09I definitely see these capabilities coming to our phones. And hence, I don't see anybody who will want to carry two devices, one who only does the things that R1 does and the rest, the things that the phone does. It's all going to get integrated into our phones. And this device, as much as it's cool and is driving a lot of hype for the right reasons, is not going to be long lived unless they're going to turn it into a phone eventually. And then they will have to compete with Apple and Android, et cetera. The last piece of news I want to share with you this week is from a company called Assistive AI, and they came out with a new video tool that can generate videos from an image or from text similar to Runway and Pica Labs and Leonardo.

14:53Their focus is to create what they say the most photorealistic videos possible. And I must admit, the first samples that I've seen online are very impressive. You heard me say that before, 2024 is probably going to be the year of AI videos. I think by the end of this year, we will be able to generate highly realistic videos in several different platforms at a reasonable length. So more than the four seconds that some of these platforms do today. And I believe it will be at the quality that we see in images right now coming from Me Journey and similar companies. Me Journey themselves said that they're starting to work on video as well.

15:34So as I mentioned, I'll be highly surprised if we won't be able to generate extremely realistic videos or any kind of videos that you want by the end of this year, just by using prompts or images as inputs. That takes the whole issue of knowing what's real and what's not on any digital medium, very problematic. meaning you need to seriously question literally anything you see online right now because the capabilities to create fake, highly realistic images, videos, text, quotations, whatever you want is there and is available to more or less anyone right now. So just be aware, pay attention and until as a society we develop a way to deal with that, each and every one of us has the responsibility to deal with that on our own and to invest in figuring out what's real and what's not before we share it with others as something interesting we found.

16:32That's it for this week. On Tuesday, we'll be back with an interview and we're going to have a fascinating episode that is going to share with you step-by-step on how to create your own custom GPTs. Until then, keep exploring AI, share what you find with the world. If you enjoy this podcast, this episode, and overall, please share it with other people that are going to benefit from it. Please open your favorite app right now and rank and rate this podcast. It really helps me very much and I would really appreciate it. And until then, have an amazing weekend.