WEBVTT 00:00:01.000 --> 00:00:02.500 All right, let's talk about the current state of artificial intelligence. 00:00:03.000 --> 00:00:07.500 It's a field that's moving at well, lightning speed, and there's a ton of noise out there. 00:00:08.000 --> 00:00:16.000 So, today we're going to cut through all of that, look at what's really happening, and figure out what's real right now and what actually matters for what's coming next. 00:00:16.500 --> 00:00:21.000 So, let's just start with a number that kind of sets the stage for everything else we're going to talk about. 00:00:21.500 --> 00:00:22.500 70 days. 00:00:23.000 --> 00:00:24.000 You might be wondering, what is that? 00:00:24.500 --> 00:00:30.400 Well, based on what we're seeing, that is roughly how long it takes for the core capabilities of an AI model to double. 00:00:31.000 --> 00:00:32.000 Think about that. 00:00:32.500 --> 00:00:35.000 Not years, not even a full quarter, just over 2 months. 00:00:35.500 --> 00:00:41.000 That pace is well, it's unlike anything we've ever seen in the history of tech. 00:00:41.500 --> 00:00:45.000 And this incredible speed isn't just something happening in a lab somewhere. 00:00:45.500 --> 00:00:48.000 It's hitting the real world, and it's hitting it hard. 00:00:48.500 --> 00:00:53.000 So, in the last quarter of 2024, you had about a third of businesses trying out AI agents. 00:00:53.500 --> 00:00:54.500 Okay, sounds about right. 00:00:55.000 --> 00:01:00.000 But then just one quarter later, that number absolutely skyrockets to nearly 2/3. 00:01:00.500 --> 00:01:01.000 This isn't some faroff trend. 00:01:01.500 --> 00:01:05.000 It is a massive fundamental shift that is happening as we speak. 00:01:05.500 --> 00:01:09.000 I mean, just trying to keep up with the news is relentless. 00:01:09.500 --> 00:01:10.500 Think about just the last month. 00:01:11.000 --> 00:01:12.500 You've got OpenAI launching GPT5. 00:01:13.000 --> 00:01:17.000 You have Anthropic releasing a model with a 1 million token context window. 00:01:17.500 --> 00:01:24.000 To put that in perspective, that's a working memory big enough to hold the entire Lord of the Rings trilogy and still have room to think. 00:01:24.500 --> 00:01:28.000 Plus, we've seen huge releases from Google and Perplexity. 00:01:28.500 --> 00:01:30.500 These are not small updates. 00:01:31.000 --> 00:01:33.000 You know, these are fundamental leaps forward happening week after week. 00:01:33.500 --> 00:01:34.500 It's wild. 00:01:35.000 --> 00:01:41.000 This crazy acceleration is exactly why we're calling this moment AI's Cambrian explosion. 00:01:41.500 --> 00:01:48.000 It's just like that biological event from millions of years ago where suddenly there was this massive sudden diversification of new life. 00:01:48.500 --> 00:01:53.000 Well, we're seeing the same thing now, but with new forms of intelligence, and it's all happening at once. 00:01:53.500 --> 00:01:59.000 So, okay, things are moving fast, but why should this break neck pace actually matter to you? 00:01:59.500 --> 00:02:01.500 Well, it boils down to three things. 00:02:02.000 --> 00:02:03.000 First up, efficiency. 00:02:03.500 --> 00:02:08.000 This is all about automating the boring routine stuff so you can actually focus on what's important. 00:02:08.500 --> 00:02:09.500 Second, opportunity. 00:02:10.000 --> 00:02:13.500 Look, your clients, your boss, your colleagues, they're already talking about this. 00:02:14.000 --> 00:02:16.500 You can either react to that conversation or you can lead it. 00:02:17.000 --> 00:02:18.500 And third, positioning. 00:02:19.000 --> 00:02:20.500 AI is not a fun little add-on anymore. 00:02:21.000 --> 00:02:26.000 It is quickly, and I mean quickly, becoming the new basic infrastructure for how all work gets done. 00:02:26.500 --> 00:02:29.000 So, how is any of this even possible? 00:02:29.500 --> 00:02:39.000 You know, to really get a handle on this moment, we've got to pop the hood open a little bit and understand what makes this new form of AI so fundamentally different from what came before. 00:02:39.500 --> 00:02:43.500 So, the old way of doing AI was all about hard-coded rules. 00:02:44.000 --> 00:02:48.500 A programmer had to literally sit down and write out explicit instructions for every single possibility. 00:02:49.000 --> 00:02:52.500 You know, if the temperature is above 25°, then turn on the AC. 00:02:53.000 --> 00:02:54.500 It was super rigid. It was brittle. 00:02:55.000 --> 00:02:58.000 And it just couldn't handle anything new that it hadn't been specifically told what to do about. 00:02:58.500 --> 00:03:03.000 But this new generation of AI, what we call large language models or LLMs. 00:03:03.500 --> 00:03:06.000 They basically just throw that entire rule book out the window. 00:03:06.500 --> 00:03:09.500 Their core job is in a way shockingly simple. 00:03:10.000 --> 00:03:16.000 They just predict the next most logical word or token based on all the words that came right before it. 00:03:16.500 --> 00:03:17.500 That is it. 00:03:18.000 --> 00:03:26.000 But out of that one simple idea comes this incredible ability to generalize, to understand patterns, and to create not just follow a list of commands. 00:03:26.500 --> 00:03:29.000 So how in the world do they actually do that? 00:03:29.500 --> 00:03:36.500 Well, it all starts with training them on an amount of data that is just unimaginable really to form what's called a neural network. 00:03:37.000 --> 00:03:40.000 And that network is then guided by a few key things. 00:03:40.500 --> 00:03:47.500 You've got weights and biases which are kind of like personality dials that a developer can tune to change its behavior. 00:03:48.000 --> 00:03:52.000 And then there's a context window which you can basically think of as the AI's short-term memory. 00:03:52.500 --> 00:03:55.500 And this gets to a really really crucial point. 00:03:56.000 --> 00:04:00.000 These models don't know that Mars is the red planet in the same way you and I do. 00:04:01.500 --> 00:04:08.000 They just know that based on the trillions of sentences they've read, the most probable word to come after the phrase the red planet is Mars. 00:04:08.500 --> 00:04:12.000 And this is the key to understanding why they sometimes hallucinate or just make stuff up. 00:04:12.500 --> 00:04:16.000 They are masters of linguistic probability, not keepers of absolute truth. 00:04:16.500 --> 00:04:20.000 Okay, so the theory is cool, but let's get practical. 00:04:20.500 --> 00:04:23.500 How do you actually put these incredibly powerful tools to work? 00:04:24.000 --> 00:04:27.000 Let's break it down into a really simple practical framework. 00:04:27.500 --> 00:04:33.500 We can really think about using AI in three levels or modes that kind of build on each other. 00:04:34.000 --> 00:04:37.000 Mode one is the chat assistant. This is what most of us have used. 00:04:37.500 --> 00:04:43.000 It's great for brainstorming, writing an email, summarizing a document, all those everyday tasks. 00:04:43.500 --> 00:04:49.000 Mode two gets more serious. It's about APIs and tools for building real repeatable workflows. 00:04:49.500 --> 00:04:51.500 And then mode three, that's the frontier. 00:04:52.000 --> 00:04:58.000 We're talking about agents where you have multiple AIs working together on their own across different platforms to get a job done. 00:04:58.500 --> 00:05:02.000 And here is a Perfect example of that second mode in action. 00:05:02.500 --> 00:05:06.000 What you're looking at is a code editor, but with AI built right into it. 00:05:06.500 --> 00:05:10.500 It's not a separate chatbot you have to open up. No, it's an active part of the workflow. 00:05:11.000 --> 00:05:18.000 It's like a co-pilot, suggesting code, spotting mistakes, and just making the developers work way faster in real time. 00:05:18.500 --> 00:05:21.500 But mode 2 can go even deeper than that. 00:05:22.000 --> 00:05:30.000 So instead of using a tool that someone else built, developers can write their own code that talks directly to the AI's API, its application programming interface. 00:05:30.500 --> 00:05:38.000 This is where things get really, really cool, cuz it lets you basically plug the AI's brain into any data source or custom app you can dream up. 00:05:38.500 --> 00:05:42.200 Okay, so this all sounds incredibly powerful, but where do you even start? 00:05:42.700 --> 00:05:51.000 Let's get into your practical starting guide. Some actionable things you can do today to start building up your own AI intuition. 00:05:51.500 --> 00:05:58.000 First off, it really helps to understand the landscape. There are basically two main flavors of these AI models. 00:05:58.500 --> 00:06:03.100 On one hand, you have the closed Frontier models from companies like OpenAI and Google. 00:06:03.500 --> 00:06:07.000 They are usually the highest performing, the most powerful, but they're a black box. 00:06:07.500 --> 00:06:11.000 On the other hand, you've got open weights models from places like Meta and Mistral. 00:06:11.300 --> 00:06:16.900 You can literally download these, customize them, and run them yourself, often for a lot cheaper. 00:06:17.400 --> 00:06:21.000 Honestly, the best way to learn any of this is to just get your hands dirty. 00:06:21.500 --> 00:06:25.000 Start using a model as a brainstorming partner for your next project. 00:06:25.500 --> 00:06:27.500 You absolutely have to try the voice mode. 00:06:28.000 --> 00:06:31.000 Having an actual conversation with an AI is a a pretty wild experience. 00:06:31.500 --> 00:06:36.000 And here's a pro tip. Give the exact same prompt to two or three different models. 00:06:36.500 --> 00:06:40.000 You will be amazed at how different their answers and even their personalities can be. 00:06:40.500 --> 00:06:44.000 And as you get more comfortable, you can start being really strategic. 00:06:45.000 --> 00:06:47.300 You know, using the right tool for the right job. 00:06:47.800 --> 00:06:52.800 For instance, models like Claude and Gemini are turning out to be fantastic for coding. 00:06:53.300 --> 00:06:56.700 If you're doing some deep research, Gemini or Chhat GPT are excellent choices. 00:06:57.200 --> 00:07:06.500 And for really complex multi-step planning, you can look for these specialized reasoning models which are designed to break a problem down and think it through more methodically. 00:07:07.000 --> 00:07:15.800 So all of this incredible progress, that insane 70-day doubling time, the constant fire hose of new models, the rise of AI agents, 00:07:16.200 --> 00:07:23.800 it is all pushing toward the ultimate goal in this field, artificial general intelligence, AGI, a true thinking machine. 00:07:24.300 --> 00:07:31.700 Now, we are not there yet, but given the trajectory we are on, the conversation has totally shifted from if we'll ever get there to just when. 00:07:32.200 --> 00:07:35.800 And the only real question left is are we ready for what that's going to mean?