I admit AI is cool. But AI has reached this point because the information was simply out there. Scattered across the web, in research papers, forums, product documentation, and user-generated content—AI scooped it up, synthesized it, and gave it form, Even if not legally all the time. The magic isn’t in the machine. It’s in the open flow of knowledge. No one thought that one day some model might just come around and steal their data and make business tougher for them. But now they do. And no one wants to help it.

The quality information flow is slowing. People are holding their cards closer to their chests. People are becoming more guarded with their information, recognizing that once something leaks or visible on internet, it ceases to be proprietary. A single instance of exposure means the knowledge spreads, gets absorbed into AI models, and suddenly, it’s common property. The competitive edge vanishes. But everyone wants to play the content game and the wells from which AI drew are now full of it’s own creations.

This shift is already visible in the rise of platforms like Substack, where content creators monetize exclusivity, keeping their insights limited to paying subscribers rather than broadcasting them to the open web. Discord communities, private Slack groups, and invite-only newsletters are becoming the new centers of high-value information exchange. The modern intellectual economy is increasingly defined by gated knowledge, where access—not just the information itself—is the real commodity. These insights exist within networks, behind paywalls, and inside closed ecosystems. As AI continues to absorb and replicate everything it can reach, the most powerful players will be those who decide what stays out of its grasp.

While secrecy and privacy have always been the goal but we were all blind to black swan that LLM training is. But someone somewhere out there is definitely working to cull this move. But some players like Springer are cooperating with GenAI companies. Who gets to see what will become differentiators even if it ends in just signing an exculsive deal with an GenAI corpo.

The open web that fueled AI’s rise is starting to fragment. Some are shutting doors outright. Others are opting for more nuanced strategies, selectively sharing insights while keeping core assets locked away. The value chain is shifting: raw data alone isn’t enough; it’s how well it’s protected, processed, and deployed that matters. So maybe we might see a future where the AI models are not just differentiated by how efficient and precise they are but also what they know.

This also means that AI can’t just keep shuffling existing knowledge around. It cannot keep living off hand me downs—it has to start creating new knowledge the way we do. That means connecting ideas, forming theories, testing them, and holding onto what works instead of discarding everything once the next dataset rolls in. The world spins because we build, refine, and let things take shape. If AI can’t learn and remember in a way that compounds over time, it will always be a step behind, stuck in an endless loop of past data and pathetically dependent on what we share with it.