Talk to bruh

The Problem

AI is smart. It's also boring.

Ask a thousand AI agents the same brainstorming question and you'll get remarkably similar answers. Different wording, same ideas. The outputs cluster around a narrow band of “most likely” responses.

This is by design. Models are trained to be consistent, helpful, and safe. But consistency is the enemy of creativity. When you need a brainstorming partner, a co-writer, or a source of genuinely novel ideas, you don't want the median response. You want the outliers.

Temperature and top-p sampling help a little, but they just add randomness at the very last step. That's like asking someone to think conventionally and then randomly swap words out. It doesn't produce creative thoughts—it produces incoherent ones.

The Idea

Creativity from the inside out

Instead of adding noise at the output, we intervene inside the model during inference — at the level where it's actually forming ideas, making associations, and deciding what to say.

Think of it this way: a model builds up a response layer by layer. Early layers explore possibilities. Later layers converge on the “correct” answer. By carefully adjusting this process, we can preserve more of that early-layer divergence—the roads not taken.

The result: outputs that are coherent but surprising. Not gibberish. Not the same thing you've read a hundred times. Something in between that's actually stimulating.

Side by Side

Same prompt, different thinking

Use Cases

Who needs a creative AI?