Quantum Lever optical hardware on the lab bench

About Quantum Lever

A live physical quantum randomness stream for LLM sampling, built from a beam splitter, two photomultiplier tubes, a Red Pitaya FPGA, and a stubbornly literal reading of Many-Worlds.

The Idea

A Quantum Lever is a chain where a microscopic quantum alternative becomes a stable record, then gets amplified into a decision that can have ordinary macroscopic consequences.

quantum event -> measurement record -> amplification -> decision -> consequence

Most quantum differences vanish into heat, vibration, and inaccessible environmental detail. Here, the branch index is deliberately routed into software: first into random bits, then into a sampler, then into token choices.

Optical diagram showing a beam splitter feeding two PMT detectors
One weak optical event, two detector paths, one which-path bit.

The Hardware

A weak light source feeds a 50:50 beam splitter. Two PMTs watch the output arms. If PMT A fires first, the FPGA records one path; if PMT B fires first, it records the other.

The FPGA rejects coincidences, applies a symmetric blanking window, and feeds the surviving which-path events into the public QRNG path used by the sampler API.

From Photon To Token

Photonsingle optical event
DetectorPMT A or PMT B
FPGAwhich-path bit
Streamsigned live batches
Samplerrandom draw
Tokenone continuation

Why LLM Sampling?

An LLM sampler maps random bits into token choices. With ordinary pseudorandomness, a fixed seed follows one deterministic path. With fresh QRNG bits, new physical quantum measurements enter the sampling loop as it runs.

Under Many-Worlds, every finite QRNG bitstring with nonzero amplitude is realized in some branch. A deterministic sampler maps those bitstrings to token choices. So every token sequence reachable by the implemented sampler is physically realized in some branch.

The model matters because it concentrates probability mass on coherent continuations. Quantum randomness supplies the branching input; the model is the lens.

Red Pitaya oscilloscope view used while tuning the detector signals
The same hardware is tuned and monitored as a live physical source, not an offline random file.

Free QRNG

The public stream uses conditioned physical randomness with the normal sampler path. It is free, shared, and enough for experiments and playful integrations.

Free Keys

Anonymous keys last 24 hours. Email-verified keys last one week. Both are free; the key system is there to reduce abuse while the community grows.

Proof Chain

Signed source batches expose payload hashes and a running BLAKE3 chain hash, so public samples can be anchored to the live stream history.