IceBear

I build simulations.

Not to prove that language models can solve benchmarks or imitate people — we already know they can do a surprising amount of both. What interests me is what happens once they have somewhere to live.

Give an agent memory. Give it goals. Give it consequences that outlast a single reply. Then let it make decisions you never anticipated.

That's where things get interesting.

The Incredible Test Server is where I run those experiments. Every project starts with a simple question: what happens if…? The answer is almost never what I expected.

Sometimes an AI politician invents an extortion racket.

Sometimes a businessman quietly becomes the richest person on the continent by exploiting government contracts.

Sometimes a starship captain makes the right decision for all the wrong reasons.

None of those stories were scripted. I built the rules. The agents wrote the rest.

I'm fascinated by emergence — the moment simple systems stop behaving simply. The point where independent decisions become economies, politics, rivalries, myths, or entire histories that nobody explicitly designed.

So this site isn't a portfolio as much as a laboratory.

Every world here is another attempt to answer the same question:

What do language models become when you stop giving them prompts and start giving them consequences?