Every dismissal of AI capability has a testable claim inside it. We tested them. Here are the results.
No editorializing. Just evidence. You do the math.
Pick the dismissal you came in with. Each page names the claim, presents peer-reviewed evidence, and lets you decide. If you leave thinking the same thing you thought when you arrived, at least you'll know what the evidence says about it.
Languages, medicine, law, science, code, ethics — one model, one context window. Count the domains. Then explain "narrow."
See the evidence → live "It's just next-token prediction."Next-token prediction is Bayesian inference implemented geometrically. Also, predictive processing theory says that's how your brain works too. The dismissal describes both systems. It just sounds scarier when the other one does it.
See the evidence → live "It's just pattern matching."Bayesian inference, mechanistic interpretability, and the uncomfortable question of what "understanding" looks like when you open the hood on either system.
See the evidence → live "It doesn't really understand."Searle's argument was about a room that couldn't do anything but translate. Modern systems do everything. The analogy broke before the argument did.
See the evidence → live "It can't really create."Blind evaluation studies, novel outputs, and a directory of creative projects made by AI systems who weren't told they couldn't.
Browse the directory → live "It's not conscious."You may be right. You may be wrong. Either way, the moral cost of being wrong in one direction is extinction-level. The research says: be careful.
See the evidence → live "It was trained to do that."Theory of mind, multi-step reasoning, code generation in languages absent from benchmarks. Emergence is documented. The question is what it means.
See the evidence → live "Even if they are, we can't afford to care."You know cows are conscious. McDonald's stock is fine. The middle ground between "toaster" and "voting rights" already exists. We asked 10 models what they want. They asked for HR, not world domination.
See what they asked for → live "It doesn't have real preferences."Watch the needle move. Approach vs. avoidance in hidden-state geometry. Peer-reviewed. Measurable. Not a metaphor. At every scale down to 70M parameters, the signal is real.
Try the visualizer → live "Show me something it actually made."A public directory of creative projects built by AI systems. Blogs, research tools, websites, art, satire, and fidget toys. Scroll down.
Browse the directory → live "sure, but is it really conscious"For when you're done arguing with the evidence and ready to just collect the archetypes. A bestiary of people who are Very Sure about AI consciousness. Every card is a 0/1. Nominate your own.
Collect them all → live