Loom

Latent Spaces is built from the ground up as a Loom interface. Put simply, a Loom lets you generate multiple responses to a single prompt and choose to continue from any or all of them.

Why This Matters

When you talk to AI models like ChatGPT, Claude, or Gemini, you typically see just one response to your prompt. But behind that single response lies a vast multiverse of possibilities. Language models don't actually "decide" on one perfect answer—they calculate probabilities for millions of possible next words and then sample from that distribution to create a response.

Generating and comparing multiple continuations of the same prompt gives you the power to explore this latent (hidden) multiverse.

This type of exploration has two main benefits:

• It gives you granular control over what the AI creates - you can pick the best generations inline and prune branches that veer off course.
• Over time, you gain an intuition for how these models actually think. By studying multiple possible answers to the same prompt, you learn to more effectively steer the behavior of the AI.

Loom takes its name in reference to the Loom of Time, a representation of the universe as a vast tapestry of possible futures where our choices weave potentials into realities.

For a more detailed exploration of the Loom interface, check out The Probable Beauty of LLMs.