Foundation models – models trained on broad data that can be adapted to a wide range of downstream tasks – can pose significant risks, ranging from intimate image abuse, cyberattacks, to bioterrorism. To reduce these risks, policymakers are starting to impose obligations on the developers of these models. However, downstream developers – actors who fine-tune or otherwise modify foundational models – can create or amplify risks by improving a model’s capabilities or compromising its safety features. This can make rules on upstream developers ineffective. One way to address this issue could be to impose direct obligations on downstream developers. However, since downstream developers are numerous, diverse, and rapidly growing in number, such direct regulation may be both practically challenging and stifling to innovation. A different approach would be to require upstream developers to mitigate downstream modification risks (e.g., by restricting what modifications can be made). Another approach would be to use alternative policy tools (e.g., clarifying how existing tort law applies to downstream developers or issuing voluntary guidance to help mitigate downstream modification risks). We expect that regulation on upstream developers to mitigate downstream modification risks will be necessary. Although further work is needed, regulation of downstream developers may also be warranted where they retain the ability to increase risk to an unacceptable level.

If elections are fair and free, citizens should accept their results regardless of the party or candidate they voted for. The evaluation of democracy should not be tainted by ‘winning’ or ‘losing’ an election. However, research on ‘losers’ consent’ has demonstrated that winners evaluate the functioning of democracy more positively than losers. We argue that the effect of losing is even more pronounced for populist voters. For them, winning and losing is indicative of the functioning of the democratic system itself. To demonstrate this, we use cross-sectional data from the Comparative Studies of Electoral Systems as well as panel data from Germany and the Netherlands for longitudinal analysis. We show that the more populist a citizen is, the stronger the effect losing is on the level of satisfaction with democracy.

Occupational shoulder exoskeletons can relieve workers during strenuous overhead work. Passive solutions are lightweight, robust, and cost-effective, but they can also restrict user movement, have limited support, and cannot dynamically adapt to different working conditions. Semi-active and active systems are still mostly the subject of research, and existing systems are heavy or have limited performance and support. Here, we present a lightweight semi-active exoskeleton for shoulder support that incorporates a novel motorized torque adjustment mechanism that varies the effective lever arm with which a spring applies force to the supporting joint. The mechanism is integrated into lateral structures and can be actuated via Bowden cables with motors located on the user’s back. The technical performance of the system was experimentally characterized in terms of its dynamic support torque profiles at six different support levels. Furthermore, adjustment times and energy consumption were investigated. The system showed plateau-like support torque profiles in the intended working range and could be adjusted from nearly 0 Nm up to 12 Nm of maximum support per arm. Adjustment times varied between 0.5 s for the adjustment of 20% of the total adjustment range and 1.0 s for a full activation/deactivation. Adjustments consumed between 0.1 As and 1.9 As of battery charge, allowing long operating times of up to one working day, using only a small 2 Ah battery. As a result, the exoskeleton provides high performance by combining comparatively high support, rapid motorized support adjustment, and low energy consumption in a lightweight design.

A COVID-19 pandemic gown conservation strategy for methicillin-resistant Staphylococcus aureus (MRSA) and vancomycin-resistant enterococci (VRE) asymptomatically colonized patients caused no significant difference in healthcare-associated MRSA (HA-MRSA) bacteremia, healthcare-associated VRE (HA-VRE) bacteremia, or healthcare-associated Clostridioides difficile infections (HA-CDI) versus prepandemic contact precautions (CP). Postpandemic HA-VRE and HA-CDI rates mirrored national trends.

Early modern urban parliaments suffered an increasing monopolization of political power that hampered urban development. To combat power monopolization, some Swiss city-states reformed their election systems by randomly selecting political representatives from a pre-elected pool of candidates. We implement a difference-in-differences design and find that lottery-based election systems improved the equality of distribution of political seats within parliaments. Lottery-based elections also had positive effects on trade tax revenues, trade volumes, and infrastructure expenditures. We explain this finding by showing that lottery-based election systems fostered the election of merchants to top political positions.

Would you prioritize assembling a group of more competent individuals or a more diverse group? According to the well-known Diversity-Trumps-Ability Theorem (DTA), when two groups of similar size are compared, the more diverse group generally outperforms the more competent group. Despite considerable criticism regarding the DTA’s mathematical rigor, it has sparked substantial interdisciplinary discussion. While most research on the DTA focuses on its implications within specific fields or on empirical simulations to test its validity, we aim to establish precise conditions under which diversity indeed surpasses ability in a particular decision-making context: voting. To this end, we integrate the DTA into a voting model inspired by the Condorcet Jury Theorem (CJT) and model diversity based on the classic dependency model in the CJT literature, namely the influence of an opinion leader.

Distinguishing early domesticates from their wild progenitors presents a significant obstacle for understanding human-mediated effects in the past. The origin of dogs is particularly controversial because potential early dog remains often lack corroborating evidence that can provide secure links between proposed dog remains and human activity. The Tumat Puppies, two permafrost-preserved Late Pleistocene canids, have been hypothesized to have been littermates and early domesticates due to a physical association with putatively butchered mammoth bones. Through a combination of osteometry, stable isotope analysis, plant macrofossil analysis, and genomic and metagenomic analyses, this study exploits the unique properties of the naturally mummified Tumat Puppies to examine their familial relationship and to determine whether dietary information links them to human activities. The multifaceted analysis reveals that the 14,965–14,046 cal yr BP Tumat Puppies were littermates who inhabited a dry and relatively mild environment with heterogeneous vegetation and consumed a diverse diet, including woolly rhinoceros in their final days. However, because there is no evidence of mammoth consumption, these data do not establish a link between the canids and ancient humans.

We define a random graph obtained by connecting each point of $\mathbb{Z}^d$ independently and uniformly to a fixed number $1 \leq k \leq 2d$ of its nearest neighbors via a directed edge. We call this graph the directed k-neighbor graph. Two natural associated undirected graphs are the undirected and the bidirectional k-neighbor graph, where we connect two vertices by an undirected edge whenever there is a directed edge in the directed k-neighbor graph between the vertices in at least one, respectively precisely two, directions. For these graphs we study the question of percolation, i.e. the existence of an infinite self-avoiding path. Using different kinds of proof techniques for different classes of cases, we show that for $k=1$ even the undirected k-neighbor graph never percolates, while the directed k-neighbor graph percolates whenever $k \geq d+1$, $k \geq 3$, and $d \geq 5$, or $k \geq 4$ and $d=4$. We also show that the undirected 2-neighbor graph percolates for $d=2$, the undirected 3-neighbor graph percolates for $d=3$, and we provide some positive and negative percolation results regarding the bidirectional graph as well. A heuristic argument for high dimensions indicates that this class of models is a natural discrete analogue of the k-nearest-neighbor graphs studied in continuum percolation, and our results support this interpretation.

Historians have agreed that the gender division of labour in rural households was broadly similar across the whole of northern and western Europe during the early modern period. Until recently, however, there has been a lack of detailed data about the historical division of labour within European countries with which to test the validity of such cross-cultural generalizations. New research that has collected evidence of work tasks for early modern Sweden and England now makes it possible to undertake a direct comparison of two European countries. In this article, the gender division of labour in agriculture, craft production and commerce in Sweden and England is compared to demonstrate the complexity of historical gender divisions of labour. It presents a detailed picture of the gender division of labour that not only shows differences between the two countries but also demonstrates flexibility and adaptability in the allocation of work between women and men. As a consequence, we argue that neither broad generalizations nor single causes are adequate explanations for the patterns observed.

We give a crystal structure on the set of Gelfand–Tsetlin patterns (GTPs), which parametrize bases for finite-dimensional irreducible representations of the general linear Lie algebra. The crystal data are given in closed form and are expressed using tropical polynomial functions of the entries of the patterns. We prove that with this crystal structure, the natural bijection between GTPs and semistandard Young tableaux is a crystal isomorphism.