Search

Fundamental algorithms such as sorting or hashing are used trillions of times on any given day1. As demand for computation grows, it has become critical for these algorithms to be as performant as possible. Whereas remarkable progress has been achieved in the past2, making further improvements on the efficiency of these routines has proved challenging for both human scientists and computational approaches. Here we show how artificial intelligence can go beyond the current state of the art by discovering hitherto unknown routines. To realize this, we formulated the task of finding a better sorting routine as a single-player game. We then trained a new deep reinforcement learning agent, AlphaDev, to play this game. AlphaDev discovered small sorting algorithms from scratch that outperformed previously known human benchmarks. These algorithms have been integrated into the LLVM standard C++ sort library3. This change to this part of the sort library represents the replacement of a component with an algorithm that has been automatically discovered using reinforcement learning. We also present results in extra domains, showcasing the generality of the approach.

Large language models (LLMs) have demonstrated impressive capabilities, but the bar for clinical applications is high. Attempts to assess the clinical knowledge of models typically rely on automated evaluations based on limited benchmarks. Here, to address these limitations, we present MultiMedQA, a benchmark combining six existing medical question answering datasets spanning professional medicine, research and consumer queries and a new dataset of medical questions searched online, HealthSearchQA. We propose a human evaluation framework for model answers along multiple axes including factuality, comprehension, reasoning, possible harm and bias. In addition, we evaluate Pathways Language Model1(PaLM, a 540-billion parameter LLM) and its instruction-tuned variant, Flan-PaLM2on MultiMedQA. Using a combination of prompting strategies, Flan-PaLM achieves state-of-the-art accuracy on every MultiMedQA multiple-choice dataset (MedQA3, MedMCQA4, PubMedQA5and Measuring Massive Multitask Language Understanding (MMLU) clinical topics6), including 67.6% accuracy on MedQA (US Medical Licensing Exam-style questions), surpassing the prior state of the art by more than 17%. However, human evaluation reveals key gaps. To resolve this, we introduce instruction prompt tuning, a parameter-efficient approach for aligning LLMs to new domains using a few exemplars. The resulting model, Med-PaLM, performs encouragingly, but remains inferior to clinicians. We show that comprehension, knowledge recall and reasoning improve with model scale and instruction prompt tuning, suggesting the potential utility of LLMs in medicine. Our human evaluations reveal limitations of today’s models, reinforcing the importance of both evaluation frameworks and method development in creating safe, helpful LLMs for clinical applications.

Present-day people from England and Wales have more ancestry derived from early European farmers (EEF) than did people of the Early Bronze Age1. To understand this, here we generated genome-wide data from 793 individuals, increasing data from the Middle to the Late Bronze Age and Iron Age in Britain by 12-fold, and western and central Europe by 3.5-fold. Between 1000 and 875 bc, EEF ancestry increased in southern Britain (England and Wales) but not northern Britain (Scotland) due to incorporation of migrants who arrived at this time and over previous centuries, and who were genetically most similar to ancient individuals from France. These migrants contributed about half the ancestry of people of England and Wales from the Iron Age, thereby creating a plausible vector for the spread of early Celtic languages into Britain. These patterns are part of a broader trend of EEF ancestry becoming more similar across central and western Europe in the Middle to the Late Bronze Age, coincident with archaeological evidence of intensified cultural exchange2–6. There was comparatively less gene flow from continental Europe during the Iron Age, and the independent genetic trajectory in Britain is also reflected in the rise of the allele conferring lactase persistence to approximately 50% by this time compared to approximately 7% in central Europe where it rose rapidly in frequency only a millennium later. This suggests that dairy products were used in qualitatively different ways in Britain and in central Europe over this period.

Every cancer originates from a single cell. During expansion of the neoplastic cell population, individual cells acquire genetic and phenotypic differences from each other. Here, to investigate the nature and extent of intra-tumour diversification, we characterized organoids derived from multiple single cells from three colorectal cancers as well as from adjacent normal intestinal crypts. Colorectal cancer cells showed extensive mutational diversification and carried several times more somatic mutations than normal colorectal cells. Most mutations were acquired during the final dominant clonal expansion of the cancer and resulted from mutational processes that are absent from normal colorectal cells. Intra-tumour diversification of DNA methylation and transcriptome states also occurred; these alterations were cell-autonomous, stable, and followed the phylogenetic tree of each cancer. There were marked differences in responses to anticancer drugs between even closely related cells of the same tumour. The results indicate that colorectal cancer cells experience substantial increases in somatic mutation rate compared to normal colorectal cells, and that genetic diversification of each cancer is accompanied by pervasive, stable and inherited differences in the biological states of individual cancer cells. Organoids derived from individual cells from colorectal cancers and adjacent normal tissue are used to investigate intra-tumour diversification at the genomic, epigenetic and functional levels.

Electronic implants are often used in diagnosing and treating human illness, but permanent implants come with problems; here, devices are described that can sense temperature, pressure, pH or thermal characteristics, and—crucially—are fully resorbable by the body.

THE Turbellaria are rapidly becoming one of the most adequately and conveniently described groups of Invertebrates. Practically all that is at present known as to their anatomy, classification and distribution is comprised in three works—the “Monograph,” by Prof, von Graff; the special memoir, by the same author, on the “Acœla”; and the masterly work, by Prof. Arnold Lang, on the “Polycladida” in the series of monographs on the fauna and flora of Naples Bay. The work before us, a magnificent folio, completes the author's share in the monographic treatment of the group. The first part was published in 1882, and was reviewed in this journal by Prof. Moseley. It is with great pleasure that we notice the dedication of the second part—jointly to Moseley and Fritz Müller—as only one of the many felicitous ways in which Prof. von Graff expresses his admiration for the work of these his fellow-labourers in the anatomy of planarians. We heartily congratulate Prof. von Graff on the appearance of this volume, the conclusion of a work begun twenty-five years ago.

The Kelvin equation1relates the equilibrium vapour pressure of a liquid to the curvature of the liquid–vapour interface. It predicts that undersaturated vapours will condense in channels of sufficiently small dimensions. While the applicability of the Kelvin equation to organic liquids with meniscus radii as low as 4 nm has been verified by direct experiment2, its applicability to water for meniscus radii below 4 nm is unverified, and has been doubted3,4. We have used Fizeau interferometry to measure directly the capillary condensation of water in a wedge between fused silica surfaces. Measurements were made at relative vapour pressures (P/Ps) from 0.996 down to 0.945, corresponding to theoretical meniscus radii from ∼120 to 9nm. We findonly a small discrepancy between our experimental results and the Kelvin equation in the vapour pressure range covered, provided that correct account is taken of the effect of the adsorbed film, up to 200 nm thick, on the meniscus shape.

MANY studies have confirmed Gasser's original observation1that the basic relationship between axon and Schwann cell is the invagination of one into the other, the line of invagination being marked by a double layer of the Schwann cell plasma membrane, the mesaxon. By this means a Schwann cell can ensheath several unmyelinated nerve fibres.

STACKS of desmosomes in successive layers of Schwann cell cytoplasm have been reported in the sheaths of nerve fibres in the earthworm Eisenia foetida1and in the rat2and also in the myelin sheath of the eighth nerve ganglion of the goldfish3. In each case the stacking occurred in loose myelin where narrow layers of Schwann cell cytoplasm were contained within the spiralling of the mesaxon. It was thought that the desmosomes in both nerve sheaths functioned as bonds between successive Schwann cell lamellae.

REFERENCES to mast cells occur only rarely in histological accounts of the structure of peripheral nerve trunks. They were mentioned by Cajal1and by Riley and West2in the perineurium of the rabbit and of the rat respectively, and in the looser connective tissues of the epineurium; they have been described in the endoneurium in the bony fish Orthagoriscus mola3. In injured nerve trunks, however, Cajal1has figured and briefly mentioned mast cells as elements of the endoneurium of the rabbit.

MANY attempts have been made by the use of various drugs and hormones and by electrical stimulations to influence the rate at which mammalian nerve fibres regenerate after injury. None of these has been successful. It has, however, recently been shown that in poikilothermic animals such as the frog1and the lizard2, nerve regeneration following upon injury is greatly affected by the temperature of the environment in which the animals are kept.

IN 1950 Armstrong1found that the period after injury during which a nerve fibre preserves a normal histological appearance may be surprisingly long in a poikilothermic animal kept at room temperature (15°–20° C.), but that the rate of degeneration is increased and the time after which the fibre eventually disappears is shortened by a comparatively small rise in temperature. This temperature effect is important, since one of the commonest methods for investigating pathways and connexions in the nervous system involves the interpretation of degenerative changes which follow experimental injury, and it is essential to know how long one must wait before it is possible to say that surviving nerve fibres, normal in appearance, cannot have been affected by the injury.

I HOPE you will allow me to express through the medium of NATURE my concern at the proposal referred to in the leading article in the issue of February 19 to centralise in an Imperial Department the various scientific services in India—a policy which I believe to be likely to prove detrimental to good work. I was a member of the Indian Forest Department during the years 1871–99, so that my Indian experience is not very recent, but I have kept myself informed of what was going on. Since I left India research institutes have been established in different provinces with officers attached to them required to devote themselves to the study of scientific questions. In my opinion, it is of the utmost importance that these officers should have as free a hand as possible, and be allowed to work in their own way on the subjects which they know themselves most competent to study. If they are called upon to work under a centralised Department, and perhaps to turn from branches of study which they thoroughly understand to others in which they may have to begin by reading up, much of their time will be wasted and the results poor.

IT has recently been reported1that an environmental temperature close to freezing point inhibited nerve regeneration in the rat's caudal nerve after crush injury. In these experiments only the peripheral part of the nerve was cooled, the temperature of the cell body being unaltered (for details, see Oamble and Jha2).

THE credit of the first discovery of this land-planarian in England lies not with Sir John Lubbock, as Dr. Scharff stated, but with the late Rev. L. Jenyns (Blomefield), who, in his “Observations in Natural History,” 1846 (p. 315), makes some interesting remarks on the “Ground Fluke” (Fasciola terrestris) and its occurrence in the woods at Bottisham Hall, a locality searched with success by Mr. Harmer.

As a resident of Dehra Dún, I was interested in a paragraph at p. 297 of NATURE for July 27, 1893, saying that the Indian Planters' Gazette had recorded a rainfall of 48 inches at Dehra Dún on the night; of January 24, 1893. As 48 inches is considerably more than half our average yearly rainfall (86 inches). I have looked up the official returns of the Meteorological Reporter to the Government of India. They give for the rainfall recorded at 8 a.m. on January 24, 1893, 0.26 inches only, 1.07 inch being the recorded fall on the same date at Mussoorie, on the hill range 11 miles off. I have examined the Dehra Dún rainfall records since January 1, 1867, and find that the largest amount recorded for any one day since that date is 11.60 inches, which is given for July 30, 1890. It is possible that the correspondent referred to wrote 4.8 inches, but even that amount, though not an uncommon fall for the monsoon season between June and September inclusive, would be a heavy fall for January. The highest recorded fall for any day in January is 2.84 inches on January 26, 1883.