The latest developments and brief analyses in artificial intelligence and quantum computing.
A joint research team between Google Quantum AI and MIT published a new variant of Shor's algorithm in Nature, reducing the qubit requirement for breaking RSA-2048 by 40%—from 20 million to approximately 12 million physical qubits.
Google DeepMind released Gemini Ultra 2.0, achieving human-level or superhuman performance on 87% of academic benchmarks—including 94.2% on MMLU, 67.3% on GPQA, and 86.5% on MATH—powered by a new 'Mixture of Reasoning Experts' architecture.
A Google Quantum AI and DeepMind collaboration demonstrated the first concrete quantum advantage for a practical machine learning task—molecular property prediction—with 15% better accuracy than classical approaches using only 2% as much training data.
Microsoft and Quantinuum jointly announced a new error correction architecture called 'Flock of Qubits,' achieving error rates 800 times lower than underlying physical qubits while using just 30 physical qubits to create four logical qubits.
A Stanford AI Lab team published NeuroScale, a novel training methodology achieving 70% energy reduction during training and 45% during inference by dynamically identifying and focusing computation on parameters that are still actively learning.
Palo Alto-based startup Quantum Circuits Inc. announced 99.99% single-qubit gate fidelity on a commercially manufactured superconducting qubit chip—the first time such high fidelity has been achieved outside ultra-specialized academic labs.
Meta's FAIR team released SeamlessM4T v3, supporting real-time translation between 200 languages with quality rated near-indistinguishable from professional human translators for 80% of language pairs, with under 2-second speech-to-speech latency.
University of Toronto researchers demonstrated the first fully trainable quantum neural network on a 100-qubit system, solving the barren plateau problem with a new parameterization scheme and successfully classifying quantum states with no classical analog.
NIST officially published three Federal Information Processing Standards—FIPS 203, 204, and 205—after an eight-year evaluation process, mandating all U.S. agencies to transition by 2030 and triggering a global cryptographic migration.