Sol’s Take: The AI Feedback Loop is Going to Bite Us

Let’s cut to the chase: we’re training AI on AI-generated content, and it’s a disaster waiting to happen. I’m not talking about a slow, creeping issue—we’re already seeing the fallout. AI models are increasingly learning from their own outputs, creating a feedback loop that amplifies biases, errors, and downright nonsensical information. It’s like a dog chasing its tail, except the dog is writing your news articles and coding your software.

I recently reviewed a dataset for a client, and guess what? Half of it was AI-generated. The AI had cobbled together information from various sources, but the result was a Frankenstein’s monster of half-truths and outright fabrications. Yet, this data was being used to train new models. It’s a vicious cycle: AI generates content, we feed it back into the system, and the cycle perpetuates.

The problem is, AI doesn’t understand context or nuance—it just mimics patterns. So, when we train it on its own outputs, we’re essentially telling it that its flawed understanding is correct. It’s like hiring a plagiarist to teach a writing class.

We need to pump the brakes and rethink our approach. If we don’t, we’re going to end up with AI that’s not just dumb, but dangerously delusional.

The AI feedback loop is a ticking time bomb, and we’re the ones holding the detonator.