Simone Zanetti’s Post

Transcript

I'd like to quickly touch on the speed at which AI is progressing and where we are going in the next few years, because it will impact your choices, and some of your choices are expensive. And if I don't, if they're not done in the right direction, they might lead them to reimplementation or other costly exercises. As you can see from this chart, the big step before that AI is taking at the moment is to surpass human intelligence. Where you see the #0, there is the human benchmark, and all those lines are going over the 0 benchmarker. Is basically when the AI has been bitten a human being at a specific task. And again, I use carefully the word task. So where are we heading in this other chart that you see that the speed and the curve at which the implementation of new AI technologies or evolution of AI technologies are taking place very, very quick and it's accelerating as we speak. The change of gear really happened with the advent of Chatter, GPT and large language models. Where the AI industry is going at the moment is to create a human. Like reasoning with chain of thoughts, which is available already today, and then going towards artificial general intelligence. Basically like our brain is not specialized like a bee that you know, it's just a tick polling and makes honey and defends the queen. Very specialized brain. Our mind as humans is very broad. We can do anything from art to war to develop programs to build a skyscraper. The same sort of evolution is happening in AI. So we are creating an AI that is able to respond to our request. And be able to do everything for us and anything for us. This will happen, according to some before the end of 2026, according to others within five years. But it's definitely happened. The next step is gonna be the artificial super intelligence. I actually call it artificial general superintelligence. And that's where more or less the singularity. They say that it might happen if it happens or not. For us at the moment, it's relevant. What is important to know is what the Super intelligence means. I explained already in 2018 about. How Alpha Go and Google algorithm of deep learning was able to beat a human being in the game of Go which is a very complex game where which every move has got the number of possibilities equal to the number of atoms in the universe. In 1997, an AI beat chess. So AI keeps on beating humans at very specific tasks. The artificial general superintelligence will happen when a I can beat humans at most tasks. But let's talk about what is it? What is available today? At the end of 2024, one of the most advanced models which is still in preview is implemented by Open AI and is called O One. O 1 has been rumored for a long time as a model, strawberry as a code name and has got the characteristic of think like of thinking like a human being. Basically using a chain of thoughts, you can do independent web research and it can self evolve. This is not new, 25% of Google software is developed by AI. But the real question is what is the difference now with chain of thoughts basically. Until now, many says that the AI replaces tax and not people, because people manage projects, right? But that's not true, not anymore. Let me give you an example. You can see here now on screen the difference between a projector and a task. Until now a I could process tasks and help us with specific tasks. Let's try instead to use this new O1 model to develop a whole project. For example, developing the game Tetris. To develop a game Tetris, you need to have a team of people. You need to have graphic designer understand the physics. Open the game, develop code milestones or you have a final objective. There's a number of tasks involved in this project. I've done a video already on this, but basically the final result is if we ask all one, just with one prompt develop the game Tetris. It basically in 37 seconds, 40 seconds, it basically develops the whole game, meaning the whole code that then launched and the game Tetris works. So I now can develop 100% a complete project as simple as developing a game is now, let's say. That an average developer can develop a simple version of the game territories in seven hours. The AI takes a 0.14% of the time of this developer if you put this one in perspective. If you take something a little bit more serious, like a research for cancer, let's say that humans will take another 10 years to find a solution. In theory, an AI barter with this power could take five days. So if this technology is available today, your question might be why don't we have a cure for cancer yet? And the problem is that we are not fast enough to feed the data to the AI. In the example like the game Go, where AI has beaten a human being with intuition developed by the training that happened in the AI. This has been done with synthetic data. So the AI has been trained with humans with thousands of games and then it started generating new games and play against itself 3 billion times. So that kind of speed that developed the solution on beating a human being in the game. And because with the medicine is more difficult to provide samples on the effect of a specific cure and then analyzing the tissue and. The report back, basically providing the data to the eye becomes the bottleneck of the processing power and that's what one bot can do. But the future is agentic, meaning that we start now having both collaborating. This is something that we teach at all master classes, how to build bots to collaborate together. So in the theoretical case that we just discussed, if we had 360 balls so we could find a cure for cancer in 20 minutes, given that we have enough data or fast enough data or synthetic data to process it, AI agent are no longer. Distant vision, they are here and that they are reshaping how we work. Recently, Microsoft announced the Copilot actions and autonomous agents, designed to handle everything from repetitive task to complex operation. And they're not alone. Open AI has been ahead of the curve with their custom GPT's enabling businesses to create AI tools tailored to their needs. So when we teach how to build bots and make them, make them collaborate. And what we suggest everybody to do at the moment is to create 2 type of box knowledge bot. I'm functional bots. If we have these two, so you have about, about your policies, about about your insurance company, about about a specific product or service that you have. Those are knowledge bots. Then you can create functional bots, for example, are both specialized in doing advertisement, another one about creating converting landing pages, another one, another one about speaking to clients, another one about processing claims. And then you can start making these bot collaborating together, creating infinite possibilities of collaboration and therefore being able to handle theoretically. Any challenge or any initiative that you have? In 2024 alone, I've helped over 100 enterprises in governments with AI. And what this seems to be clear is that there are two types of AI that can implement it in our organization and they both have benefits and they work very differently. The tool that employees can independently use and the system that IT teams implement at scale. Let's impact both quickly. First, there are the tools for individual like Microsoft Copilot, CHPT, Mid Journey and all of that. These tools led to the influence. In productive without the intervention of IT, they need to be ring fenced, of course, to be used in the way that they are supposed to be used. But these tools give the possibility to the final user to be creative when analyzing data, when creating the new initiative or even entire project that we saw at the beginning. This of course is a little bit stressful for the IT teams because they are not sure what the user will do. But there are ways to ring fence it so that we give the maximum power to the employees to really have superpowers and do so much more than they could do before and at the same time. Not being dangerous and stay within compliance of privacy, security and all the norms of data we need to follow. On the other side we have the enterprise AI big integrated system. Think about APIs, interrogating large language models connected to data lakes via RAG methods and vector databases. These aren't about the personal productivity. They are about to building tasks like automation claims, managing bots, powering organization wide insights. They are both useful and different. And it shouldn't be confused because 1 does not exclude the other. This is a short video, so I'm not going to go into the details. Probably this is the subject for another video. But here on screen, I share with you a summary of where independent AI tools and enterprise AI system or AI tools should be used in a large enterprise. But with great powers come great power consumption. And this is a big thing, especially for developing countries. I used to teach you that in in a very simplistic way. AI is made of three components, the brain, the algorithm. The hardware where it runs from and then the data that is fed for either prompting but most of all for training. But now there is another very important component which is electricity. The GPU is using AI, especially the last one from NVIDIA which is this big consumes one single chip, 16 kilowatts and a data center like Colossus of Elon Musk has got 100,000 of these chips. To give an idea, Google is buying energy from a fleet of mini nuclear reactors to power their AI microphone. Started a nuclear power station that was shut down in 1979 to power their AI. Amazon is investing in three new nuclear power plants to power their AI. And Silicon Valley is investing millions in nuclear because AI is so power hungry. And they invest in nuclear because the more efficient and has got less carbon footprint. So it helps with PSG as well. But the point is that in countries where electricity is already a problem, being able to feed us such a big data center as the ones that we mentioned. Is going to be a challenge if even if the United States now struggle with electricity to power AI, which means that the developing country needs to have either power investment to build power station just for AI data center or rely in international data center, which has got the implication with local law and regulation about treating data going out of the borders of the country. So where to from here? The AI future that we've been discussing isn't years away, is here right now and it's time to focus on what to adapt and what to implement. My suggestion is really divided into 4 parts. First and foremost, employee upskilling. The best AI tools are as good as the people that use them. Upskilling your team to work with AI is critical not only to use the tools but to understand them. This includes training on AI basics, advanced use cases, ethical consideration like privacy and compliance. Empower your people to become AI champions #2 strategic AI, Rd. maps. Invest in AI strategically. Start with tools that deliver immediate. Value like a Microsoft Copilot or custom GPS for productivity. And by the way, don't confuse the fact that Copilot and a larger language model are not the same thing. They both serve different purposes. We do full master classes on Copilot and on CHPT. They're not the same thing and they don't fulfill the same purpose. Once you intake those large language models, build towards a more integrated system, enterprise AI that use retrieval, augmented generation, vector databases and leveraging on your data leaks. This solution will enable any insurance company or any large enterprise actually to automate workflows, scale efficiency and unlock new opportunities #3 sustainability and efficiency. AI is a powerful tool but it's resource intensive. As we adopt AI, solution is vital to considering energy efficient system and infrastructure. For Africa, for example, this is particularly important ensuring solution that align with the unique challenges of the continent while maximizing impact #4 collaborative ecosystems. The eye is not a solo journey, we are all learning, we are all advancing. So whatever you do internal, make sure that you have partners that can bring knowledge into your organization and together move to the next step. AI is not one of those waves likethe.com or cloud that came and it goes slow. This is happening extremely quickly and those companies that advance quickly in AI, they have a great advantage over the rest of the market. So collaboration is essential. The future is here and AI is here to help talking about. Help. I hope that this video has been useful and I'll see you in the next one. Thank you for watching.