AI Unfiltered: The Real Shifts Behind the Hype

A new study just found chatbots can reduce a person's belief in conspiracy theories by almost seventeen points on a one-hundred-point scale. Last week, we covered how energy demand is the hard limit on AI's growth. But while everyone's watching the power grid, the code just learned how to rewire human belief. So what else moved? First, the money. Andreessen Horowitz just announced a one-point-one BILLION dollar fund, the Machine Age Fund, aimed squarely at AI infrastructure. Think chips, memory, and even the transformers that power data centers. Second, the model wars are accelerating into a price war. Google’s new Gemini 3.7 Flash just launched with a massive one-million-token context window, and it costs HALF what the last version did. Not to be outdone, OpenAI’s GPT-5.6 Luna crossed a billion active users...

and promptly slashed its input token price by eighty percent. It's a race to the bottom on price, and a race to the top on scale. Meta is playing a different game, releasing a new open-weight model called Glimmer that you can download and run yourself. It’s Mark Zuckerberg’s vision of democratizing AI, though the company’s most powerful models remain locked away. Meanwhile, a company you might not know, DeepSeek, is approaching a funding round that could value it at seventy-four BILLION dollars. And for a final dose of reality, the University of Chicago just BANNED AI in its core social science courses. They want students to learn how to think without it first. So you have a tidal wave of capital and capability on one side... and a handful of professors trying to build an ark on the other.

Okay, let's go back to that Andreessen Horowitz fund. One-point-one billion dollars is a lot of money, but what matters is where it's pointed. And it's NOT all going to software. They are explicitly targeting the hard stuff. The physical world. Remember the power bottleneck from our last episode? A16z is funding startups building new kinds of data center transformers using silicon carbide. They're smaller, more efficient, with faster lead times. Why? Because as their own partners put it, the hardware supply chain is used to growing maybe thirty percent a year. AI needs TRIPLE-DIGIT growth. Yesterday. This isn't just about servers. The fund is also backing robotics and edge AI hardware. This is the signal. The smartest money in the valley is saying the next phase of the AI race won't be won in the cloud.

It'll be won in the factory. It'll be won by whoever can physically build the machines, and the components for the machines, fast enough. The software is outrunning the atoms. This fund is a bet on the atoms. Now for the other side of the coin. Not the machine, but the mind. That study on conspiracy theories is a thunderclap. For years, researchers have known that trying to debunk a committed conspiracy believer is like trying to put out a fire with gasoline. It often just reinforces their beliefs. Nothing has worked. And then a team has people engage in three chatbot conversations. And belief drops. A LOT. One of the researchers, Gordon Pennycook at Cornell, said they were "shocked" when they saw the results. It "blew us out of the water." Here's the catch. And this is why we talk about signal versus noise.

The authors just announced the study will undergo a correction. There were errors in the dataset and the analysis pipeline. Now, they're adamant the core finding holds—that the direction and the size of the effect are real. But the fact remains, the initial, perfect-looking result wasn't quite so perfect. It's a reminder that in this space, even the most stunning breakthroughs need a second look. The AI might be good at persuasion, but science is still about verification. So here's the split screen for today. On one side, you have a billion-dollar bet on silicon carbide chips and data center hardware—a brute force push to solve the physical limits of computation. It's a story about supply chains, logistics, and raw industrial might. And on the other side, a quiet, stunning, and slightly flawed discovery about how these systems interact with the most complex thing we know: human conviction.

One is about building the engine bigger. The other is about figuring out what happens when you turn the key. Right now, we're doing both at once, and the gap between what the machine can do and what we understand about its effects is getting wider every single day.