Imagine an AI, not one that just predicts what you'll type next, but what you'll do. Yeah. We're talking about a model that actually beat human cognitive models at predicting behavior. Just by reading psychology experiments. It's kind of wild. Yeah. It's not science fiction. It's this new AI called Centaur. Well, its accuracy is quite startling. Welcome to the Deep Dive. Today, we're going to unpack a whole bunch of insights trying to chart this really complex,
fast -moving AI landscape. Right. It's all pulled from a recent newsletter that was pretty packed with info. Our mission really is to give you those crucial nuggets of knowledge, the stuff you need to kind of navigate this accelerating
AI world. We'll kick off looking at Meta's... latest big quest this ai super intelligence for everyone idea okay then we'll zoom out a bit look at industry highlights maybe some practical tools you can actually use and then yeah dive deep into centaur that mind reading ai and what that means for understanding us you know human cognition all right let's get into meta's big claims then the company that you know gave us the metaverse dream Now pitching AI superintelligence
for everyone. It sounds so familiar, doesn't it? It really does. Like that metaverse push in 2021. Yeah. We all remember how that went. Horizon Worlds ended up being kind of this virtual ghost town. Right. Even Meta's own employees weren't really using it much. Yeah. A quiet digital place is a good way to put it. And now Zuckerberg's latest memo talks about AI reels you can chat with, these always -on AI video chats that mimic
human expressions. And even AI friends, like designed to fill social gaps, give you the feeling of 15 friends. It's quite a vision. It is. But what's interesting, what's different this time maybe, is that while the pitch sounds familiar, Meta actually has some real traction. Okay. How so? Over a billion people are using Meta AI tools every month. A billion. That's just worlds away from the few hundred thousand metaverse users.
Wow. Okay. That's a huge difference. Huge. And they're using AI coding tools inside the company, too, which helps them build faster. Plus, people are genuinely kind of forming attachments to these AI personas. Really attachments? Yeah, it's like this whole new type of social connection emerging. It's interesting. The fundamental problems are still there, right? These models, they still hallucinate, make stuff up. Oh, absolutely. They invent information. Their reasoning is often,
you know. hit or miss. You give them a simple logic puzzle, even a basic game, and they often just break. So that super intelligence part still feels pretty far off, like science fiction territory. Very much so. And if you connect it to the bigger picture, you see meta is, well, it's rebranding. Again. Right. And this strategy keeps investors happy, keeps them excited. That's key. And it helps them pull in top AI talent, which is super competitive right now. Helps justify those kind
of crazy salaries. So it's partly a business move, attracting talent and investment. Definitely. Less about some sudden breakthrough, more about smart positioning in the market. OK, so is Meta's AI. push fundamentally different this time around? Or is it just, you know, a new label on an old approach? Well, it definitely has user traction this time, unlike the metaverse. But those core AI challenges. They absolutely remain. All right.
Let's move beyond meta then. Look at the broader AI industry because there's just so much happening. Seems like it speeds up every week. Totally. Take Baidu, for instance. They just dropped Mute Steamer. Mute Steamer. Yeah. It's a new model. Basically, it goes head -to -head with Google's VO. It makes these 10 -second video clips. Okay. But get this. Fully synchronized visuals, sound effects, and spoken dialogue. All generated together.
That's a pretty big step for multimodal AI, you know, AI that handles different types of data. That sounds impressive. And it brings us to some of the, let's say, weirder sides of AI's impact. Like, did you hear about that job hopper? The one landing multiple AI jobs at once? Yeah. Keeps getting caught, fired, then just does it again. Soft chuckle. Yeah. Yeah, I saw that. It's funny, but it kind of highlights how AI is really shaking up the job market, creating these weird news
scenarios. Definitely. And speaking of shaking things up, look at news consumption. Oh, yeah. That's a big one. SimilarWeb found news -related searches on ChatGPT just exploded up, what, 212 %? Yeah, massive jump. But at the same time, traffic to actual news websites down 26 % since Google rolled out its AI overviews. That's significant. A big shift in how people get their information. Maybe a bit concerning. Could be, yeah. Changes the whole information ecosystem. But on the flip
side... There are new opportunities. Like if you're good at spotting AI mistakes and fixing them. Like Sarah Skid, the product marketing manager. Exactly. She's making like two grand just fixing bad AI website copy. So there's value in that human touch, that refinement. For sure. But then you've got MIT economist David Autor warning about a potential Mad Max scenario. Mad Max. What does he mean? Basically that AI could automate so much that a lot of current skills
just become worthless. A pretty stark warning about economic disruption. Wow. So opportunity and disruption side by side. Exactly. And look at the money pouring in. Recognize this investment firm. They just raised $1 .7 billion. Billion with a B. Yep. Specifically to invest in mid -sized AI digital services companies. Shows huge confidence in the sector's growth. It's such a mix, isn't it? AI creating wealth, new jobs, but also disrupting industries, raising these
big questions about, well, about us. Yeah. Huge opportunities and pretty profound existential questions all tangled up together. So thinking about this broader AI landscape, what's the takeaway for our careers and how we get information? Well, it's creating new jobs, definitely changing how we access news, and yeah, challenging the value of our existing skills. Okay, let's pivot then. Let's talk practical stuff. Tools, strategies,
things you listening can actually use. I know a lot of people worry, especially if they run automation businesses, about these DIY tools maybe replacing them. It's a fair worry. But here's the thing. Most AI projects, they don't fail because the tech isn't good enough. They fail because of bad strategy, poor alignment with business goals. Ah, okay. So it's not just about the tool itself. Right. The key is to shift. Don't just be the person who implements the tool.
Become that high -value strategic partner, the one who understands why you're using AI and how it fits the bigger picture. Makes sense. And for people who do want to build things. Maybe without being coding experts. Yeah, totally possible now. You can build your own custom AI apps without writing code. There are guides out there using tools like Firebase Studio for the backend stuff, database hosting. Right. And N8n for workflow automation. You can basically stitch together
a real working app. Anyone can try it. It's kind of like playing with digital Lego blocks. That's pretty cool. Democratizing AI development a bit. Definitely. And even with existing tools like ChatGPT, becoming a real pro means learning specific techniques, how to refine your prompts, how to automate tasks. You know, I still wrestle with prompt drift myself sometimes. What's prompt drift again? It's when the AI's responses kind of change or get worse over time, even the same
prompt. Makes it hard to get consistent results. So mastering prompting is really key if you want top -notch output. It's an art. Gotcha. And there are new tools popping up all the time, right? Oh, yeah. Loads. Like open memory. It syncs memory across different AIs. So you can pick up a conversation with one AI where you left off with another. Whoa, that sounds useful. Yeah. Or afo .dev. Phase one. Builds mobile apps in minutes. Even integrates Stripe for payments. In minutes. Seriously.
Apparently. And browse. Anything that is an AI agent for your browser that can automate pretty much any task you do on the web sounds super powerful for getting stuff done. Okay, those sound seriously useful. Any others? Well, there are some quirkier ones too, like Mori. It's this timer that shows your age or even like a countdown to your estimated death in real time. Ah, okay. That's morbidly interesting. Chuckles. Yeah,
definitely a conversation starter. And beep that out, an AI profanity filter for content creators. Could be handy for keeping things clean. Right. And just some quick hits from the industry that show how fast things are moving. Go for it. MidJourney has 17 new art styles. Google finally rolled out VO3, their video model, globally. About time. Daniel Gross, co -founder of SSI, he joined Meta's big AI lab. Meta's trying a new tactic, having chatbots message you first to keep you engaged.
Oh, interesting. Proactive chatbots. And maybe the most amazing one. AI apparently gave fresh hope for treating male infertility, like, in hours. Yeah. A huge potential medical breakthrough. Wow. See? The pace is just incredible. These tools are really changing what's possible for everyone. So for the everyday user, how can they best leverage all these new AI capabilities that are coming out so fast? Really, by using these smart tools wisely and, crucially, getting better
at understanding those prompt techniques. Okay, now for the part that feels, well, feels like a pretty big deal, maybe even a little unsettling. Let's talk about Centaur. Right, Centaur. So researchers at Helmholtz Munich in Germany, they trained this AI, and you called it mind reading earlier. It's shockingly good at predicting what humans will choose to do next. How did they train it? Get this. 10 million real choices from over 50 ,000 actual people across 160 different psychology
experiments. 10 million data points on human choice. Yeah. Whoa. Imagine scaling that kind of insight, understanding human decision -making on a massive, massive scale. The implications are just huge. And they built it using Meta's Alama 3 .1 model. Yep, a standard large language model. They basically turned these psychology tasks into plain English descriptions. Then they fine -tuned the model, but only tweaked like 0 .15 % of its parameters. Tiny adjustment. And
the whole training took... Five days. Just five days to get these incredible results. It's ridiculously fast. Incredible results meaning how good was it? Okay, so it went up against 14 classic cognitive models from psychology. You know, established theories of how we make decisions. Centaur beat them in 31 out of 32 different tasks. Wow. It just outperformed decades of psychological modeling.
Pretty much. And even when the researchers tried to trick it, like changing the details in stories or throwing in new logic problems it hadn't seen, Centaur just adapted. Like a human would. So it's not just memorizing patterns, it's generalizing. Exactly. It seems to be showing this deeper level of adaptation. And the wildest part you mentioned earlier, its internal structure started looking like a human brain. Yeah, that's the kicker.
Without being trained on any actual brain scan data, nothing neurological, it just converged. It organized itself in a way that mirrors how our brains process things. That feels profound, like it found the optimal structure independently. It does, and it gets better. Centaur actually discovered a completely new decision -making strategy, one that performed better than existing psychological theories. The AI found a new way we make decisions. Yep, which just flips things
around, right? It raises this huge question. What can AI teach us about ourselves, about our own minds? It's like getting insights from, well... From a non -human intelligence. So Centaur is like having a virtual human test subject. You can run countless simulations. Exactly. Scientists can simulate behavior, test out theories, explore cognition without needing labs, without years of recruiting people for trials. Think about what that could do for mental health research.
behavioral design, education. Marketing. Oh yeah, definitely marketing too. The efficiency gains could be absolutely massive. What? There's always a but, isn't there? What's the downside? The flip side is pretty significant. If AI can simulate your decisions this accurately. It can manipulate them? Precisely. It can be used to nudge you, maybe exploit your biases, influence your behavior at a huge scale. This could be maybe the first real step towards what people call cognitive
AGI. Artificial general intelligence, but focused on thought processes. Right. It's not conscious, let's be clear. Not sentient like us. But it might be getting close enough to mimicking our cognitive processes that it fundamentally changes the game. And that makes it an incredibly powerful tool. Powerful and with a really serious potential. So this predictive power, does it ultimately open more doors for understanding ourselves or does it close off more avenues for genuine human
choice and autonomy? That's the question, isn't it? It definitely opens huge doors for research. Yeah. But yeah, it raises very serious concerns about manipulation and autonomy. OK, so wrapping this all up. What does this mean for you listening right now? We've covered a lot. From Meta's big AI dreams, maybe learning a bit from their metaverse bumps, to these incredible leaps like Baidu's video generation. We looked at new tools you can use and how the job market is shifting under
our feet. Yeah, the big idea, I think, is just the sheer speed, the accelerating pace of AI development. It's almost dizzying. Right. from the practical tools you can unload today to something as profound as Centaur. AI's reach is expanding so fast it's touching everything, how we get news, how we work, even how we understand what it means to be human. That line between the hype and the actual reality seems blurrier than ever.
Especially with tools that can predict our own behavior with this kind of accuracy, it's constantly shifting. This deep dive really paints a picture of a future that's, well, it's full of incredible promise, but also really complex challenges. Definitely both. And as AI gets woven deeper into our lives, predicting our choices, maybe even sensing our emotions, it leaves us with a pretty crucial question for you to think about. Yeah. If an AI can predict what you're going
to do. How much of your decision -making is still truly, authentically your own? That's a heavy one. Maybe think about how knowing these capabilities exist might change how you make choices. What does it all mean for our future autonomy in a world that's increasingly driven by AI? Now, T .O. Rowe music.