Status

Most of this work is done by volunteers. If you contribute, you get added as a co-author on any papers we publish from the work. For some tasks we offer payments through the Super Intelligence Scholars Program. If any of this interests you, reach out.

~^

Universal Migrator: LLM attention weight analysis.
Run five different language models on Fibonacci and other mathematical sequences. Extract the attention weights and check whether the relative distances between mathematical concepts are consistent across models. If they are, it means the models are detecting something real in the structure of mathematics, something we can encode into a quantum circuit. Should take one engineer about 1-2 weeks, and you don't need quantum hardware for this part.

Universal Migrator: Quantum baseline profiling.
Set up accounts on IBM Quantum, IonQ, and Rigetti, then run bare qubits with no encoding and no experiment attached, just qubits collapsing on their own, and build a noise profile for each backend. You can't interpret quantum results later without knowing the noise floor first.

Universal Migrator: quantumlib.
Build a thin Python wrapper so the same experiment code runs on IBM, IonQ, or Rigetti without changing the logic, since results need to be comparable across platforms. It's pure engineering, no physics knowledge needed.

Universal Migrator: The main experiment.
Take LLM attention weights, encode them into qubits, let them collapse, record the outcomes. Compare against random coin flips and the noise profiles. Do the collapsed states favor the mathematically correct next Fibonacci number more than chance predicts? This only runs after the three tasks above are done.

Universal Migrator: The oracle version.
Encode Fibonacci directly as a quantum circuit with no LLM in the middle. It's a cleaner experiment with fewer moving parts, and becomes the primary path if the LLM attention weights turn out to be too noisy.

Interactive demo.
A public-facing demo that shows the experiment running in real time. The Fibonacci sequence is visually intuitive, and we want anyone to be able to watch it and understand what we're testing.

Paper.
Published, though it's just the theory and experimental protocol so far, no results yet. Co-authors are everyone who worked on the tasks above.