Oh no

The explosion of AI-generated content, especially after 2023, has changed the character of the internet.

The half-life of facts is getting shorter. Pre-2023, Google indexed a mostly human web. Post-2023, it’s indexing an echo chamber of AI-generated content.

A little like the game of Chinese Whispers, that content gets distorted at every reuse. It’s a snake eating its own tail, eventually it’s useless.

I saw this firsthand in my last blog post. I asked GPT to list the top 10 product categories in which Chinese exporters dominate. The first list it gave me? Lifted straight from recent tariff-war coverage. All strategically important: rare earths, batteries, semiconductors – the usual geopolitical chess pieces.

But that wasn’t the real answer. I had to force it to think about mundane stuff – buttons, zippers, beach spades – the things China actually floods the world with. The stuff it would never threaten to withhold because it’s too trivial to weaponize. And let’s be honest, the Chinese government doesn’t like being laughed at.

Even now after much prompting, my GPT mate just gives me the first 10 it finds. There’s no diligence, no curiosity, no self-critique. Just a confident summary of someone else’s summary. The list can’t be trusted.

I’d like to think that this will change. That the next generation of models will ask better questions, not just give faster answers.