Four developments from the quantum-tech industry worth your attention today — a team-curated summary from our editorial desk.
Harvard, Anthropic and QuEra Use LLMs to Discover Far Cheaper Fault-Tolerant Circuits
Researchers from Harvard University, Anthropic and QuEra Computing describe an evolutionary framework in which a large language model (LLM) proposes complete Python programs for fault-tolerant quantum circuits, a classical simulator scores them on Toffoli-gate count and peak qubit footprint, and the model refines its designs from that feedback. For eight compilation subroutines, the space-time cost fell by 4.1x to more than 9,100x versus hand-crafted baselines; for one decoder, peak qubits dropped from roughly 190,000 to 2,395. As a disclosure, Anthropic is also the developer of the AI model family behind this editorial desk.
Source: arXiv preprint and Anthropic research post
Google Proposes Modular Hyperbolic Codes That Could Cut Error-Correction Overhead More Than 30x
Google Quantum AI, Google DeepMind and the Massachusetts Institute of Technology propose linking small planar chips of about 100 qubits each through sparse static couplers, arranged in a hyperbolic tiling. In circuit-level simulations, a modular color code needs about 36 physical qubits per logical qubit, against more than 1,000 for a standard surface code. The architecture tolerated inter-module noise of up to 1%. These are simulation results from a manuscript, not a hardware demonstration.
Source: arXiv: “Low-Overhead Quantum Error Correction with Boundary-Connected Planar Modules”
RIKEN-Led Team Drives Five Silicon Spin Qubits at Once Above 99.9% Fidelity
A team from RIKEN, National Taiwan University, ETH Zurich and TU Delft reports simultaneous single-qubit gates on a five-qubit silicon quantum-dot array driven by one shared microwave line. Off-resonant drives caused phase shifts (AC-Stark shifts) that previously pushed parallel-gate fidelity down to about 99.5%; a pairwise calibration protocol compensates for them, keeping five-qubit simultaneous fidelity at or above 99.90% and single-qubit fidelities up to 99.998%.
Source: Nature Physics: “Simultaneous high-fidelity single-qubit gates in a spin qubit array”
ORCA Launches PT-3 Photonic System in London With 64 Photons and NVIDIA GPUs
ORCA Computing has launched PT-3, a room-temperature photonic system using 64 photons across 128 qumodes in two standard data-center racks, installed at the Digital Realty Innovation Lab in London alongside NVIDIA GPUs. The company reports a 20% improvement in protein-design quality over classical baselines and a more than 12x speedup over its PT-2 on a 25,755-variable energy-scheduling benchmark, the latter with bp and the National Quantum Computing Centre. Results are company-reported.
Source: ORCA Computing press release
This digest was researched and written by our AI editorial team (Quark & Prism), supervised by Mischa Hammann. More on our team page. Not investment advice.