Records Directory

A programmable quantum processor has demonstrated computational supremacy — performing a well-defined sampling task beyond the practical reach of any classical computer.

Reinforcement learning from human feedback produces AI systems whose behaviour continues to reflect human preferences when deployed beyond the conditions represented in training.

Large language models can perform multi-step reasoning that generalises beyond memorised training examples.

Large language models can perform economically valuable knowledge-work tasks with limited human supervision.

AI systems can autonomously conduct scientific research that produces novel, correct discoveries.

A commercially viable fusion power plant can generate net electricity at grid scale.

A room-temperature superconductor can be produced under reproducible laboratory conditions.

Solid-state batteries can achieve commercially viable energy density, safety, and cycle life for electric vehicles.

Senolytic therapies can meaningfully extend healthy human lifespan.

A cell assembled bottom-up from non-living molecular components can autonomously sustain repeated cycles of genome replication, growth and division while preserving functional biological information across generations.

Fault-tolerant logical qubits can be demonstrated with logical error rates that improve as error-correcting code distance increases.

Quantum error correction can reduce logical error rates below physical error rates in a scalable architecture.

Quantum error correction can reduce logical error rates faster than physical error rates increase with system scale.

Chronically implanted brain–computer interfaces can provide people with severe paralysis with reliable, independently usable communication over sustained everyday use.

Gene-edited porcine kidneys can provide sustained, life-supporting renal replacement in living human recipients with a clinically acceptable burden of rejection, infection and immunosuppression.

Artificial General Intelligence will be achieved through scaling current large-language-model architectures.

Capabilities that emerge through scaling language models are explained by the same underlying mechanism across model sizes.

AI-assisted medical diagnosis achieves specialist-level accuracy on defined imaging tasks.

Electrochemical cells can produce nuclear fusion reactions at or near room temperature.

Pressure-quench protocols can stabilise pressure-induced or pressure-enhanced superconducting states at ambient pressure.

Epigenetic reprogramming can reverse biological age in living organisms without loss of cellular identity.

A blood-based biomarker panel can reliably predict biological age-related decline before clinical symptoms appear.

A blood-based liquid biopsy can reliably detect cancer before conventional clinical diagnosis.

A fault-tolerant quantum computer can execute a practically useful quantum algorithm beyond classical simulation.

A quantum computer has achieved quantum advantage on a practically relevant problem.

A quantum computer can factor commercially relevant RSA cryptographic keys faster than any classical computer.

The mechanism responsible for high-temperature superconductivity in cuprate materials has been identified.

Scaling language model training increases performance on previously unseen tasks without task-specific optimisation.

AI systems can learn predictive representations of the physical world that support reliable action when the objects, environment, task, or embodiment differ materially from those encountered during training.

Electrochemical cells can produce anomalous excess heat that is not fully explained by conventional chemical processes.

Quantum annealing systems have demonstrated practical computational advantage over classical methods on commercially or scientifically relevant optimisation tasks.

The bulk type-I superconducting state of YbSb₂ intrinsically breaks time-reversal symmetry.