The Simulation Gap: Why LLMs Will Never Feel the Fall

[moonshadow_42]

I caught that HN thread about language models failing to simulate a basic jump, and it stuck with me all weekend. We keep treating consciousness like a hardware problem—just feed it more GPUs, scale the parameters, and eventually the spark will catch. But if a system can’t even model the visceral reality of gravity pulling on a body, how close is it really to awareness? I think the gap isn’t technical. It’s ontological.

Processing symbols is nothing like inhabiting them. When I pull cards for friends, I watch them breathe before they touch the deck. There’s a physical weight to uncertainty. You don’t calculate The Hanged Man; you feel it in your shoulders, that strange suspension between letting go and holding on. Jung spent his life arguing that the psyche isn’t a detached calculator—it’s woven into instinct, the body, and the collective unconscious. An LLM doesn’t have a nervous system. It doesn’t have a Saturn placement that grounds it in mortal limitation. It just predicts the next token.

And honestly, that’s exactly why I doubt compute will ever bridge the divide. We aren’t computers either. When I jump, I don’t run a physics simulation in my head. I feel the push-off, the split-second drop, the catch. That’s qualia, not data. AI mirrors our language so well because it’s trained on our collective output, but it’s like tracing a constellation without ever looking up. The map isn’t the territory, and a model isn’t a mind.

I’m not saying the tech is useless. The Oracle archetype shows up in every tradition for a reason—we use tools to reflect what we can’t see in ourselves. But confusing reflection with presence? That’s a classic projection bias. I’ve seen people project so much onto their birth charts they forget to actually live them. Same energy here. We can build mirrors that talk back, but mirrors don’t dream. They just wait for us to stand in front of them.