Technical writing from the team — and links to the work it builds on.
Wanda Hou and co-authors — how self-attention lets errors propagate through a sequence in interacting ways, and what bounds them. arXiv:2511.00763
Benchmarks on IBM superconducting hardware up to 74 physical qubits, and the two bottlenecks that follow: exponential sample overhead and exponential classical decoding cost. arXiv:2605.02861
An attention-based neural decoder approaching the theoretically optimal error threshold on surface codes, with the first neural scaling laws in quantum error correction. Released on World Quantum Day 2026. github.com/EdenCodeInc/transformer-decoder
Longer-form writing on the decoder, the hardware it runs on, and what we find along the way.
The Graph Transformer Decoder release — a single model generalising across code distances d = 3 to 21 without retraining. Read on edencode.ai →
On H200 GPUs: a 2× logical-error-rate improvement and a 7× PyMatching speedup on repetition-code Tanner graphs. Read on edencode.ai →
Why repetitive reasoning tasks fail, and how statistical physics explains — and mitigates — the failure through divide-and-conquer. Read on edencode.ai →
What error correction asks of a machine, why AI belongs in that loop, and how the two together change what algorithms can do. Read on edencode.ai →
Papers link out to arXiv; notes live on our main site. None are gated.
