剑桥大学

Private schooling is expanding rapidly in Nigeria, yet evidence on whether private schools deliver better learning remains mixed. Using data from the 2015 Nigerian Education Data Survey, a nationally representative household-based assessment of nearly 30,000 primary school children in grades two to six, I estimate differences in foundational skill achievement between private and public school students using ordinary least squares (OLS) and propensity score matching (PSM) to adjust for observable differences between students, and instrumental variable estimation and Heckman selection correction to additionally account for unobserved selection into school types. While private schools outperform public schools in raw comparisons, these advantages largely disappear once selection is accounted for, with at most a small literacy advantage and evidence of a numeracy disadvantage remaining after adjusting for unobservables. Heterogeneity analysis reveals that the numeracy advantage in OLS is driven largely by children who do not receive extra tuition at home and disappears for those who do. Among the poorest households, private school attendance is associated with meaningfully higher achievement in both domains once observable selection is addressed. Regional analysis further shows that the private school achievement advantage is concentrated in the South, with no significant advantage observed in the North in either OLS or PSM. Analysis of native language literacy assessments further shows that the English literacy advantage does not extend to native language literacy and is negative among private school students in Northern Nigeria, supporting the interpretation that differential language of instruction rather than superior school quality drives the observed literacy gap. The results suggest that household selection drives most of the observed differences in achievement at the population level, and that the value of private schooling depends critically on household context, with higher achievement concentrated among the poorest and least-supported students.

Embodied artificial intelligence (AI), which integrates perception, decision-making, and execution through interaction with the physical world, has recently emerged as a promising paradigm for enabling autonomous and adaptive manufacturing systems. While AI has been extensively explored in various manufacturing tasks, the major body of work remains fragmented across different processes, model paradigms, and task stages, lacking a unified basis for assessing how far current technologies have actually progressed towards system-level manufacturing intelligence. To address this gap, this paper introduces a task-oriented analytical framework grounded in the Perception, Decision–making, Execution (P-D-E) loop, which provides a common structure for systematically organizing, comparing, and evaluating AI contributions across heterogeneous manufacturing domains. Four representative processes including additive manufacturing (AM), machining, welding, and assembly are examined to map how AI technologies participate in each stage of the manufacturing task cycle. A six-dimensional evaluation criterion covering autonomy, adaptability, objective alignment, domain expertise, physical interaction, and industrial readiness is further proposed to assess the embodied AI readiness of each domain. The cross-process analysis yields several findings that go beyond individual process descriptions: the primary bottleneck towards embodied manufacturing intelligence is not the absence of capable models but the fragmentation of existing AI functions into isolated task-specific subloops; perception and reasoning capabilities have matured considerably faster than safe, reliable physical intervention; and different processes require fundamentally different embodiment pathways shaped by their physical coupling characteristics and deployment conditions. Based on these findings, the paper argues that near-term progress should follow a hierarchical and modular embodiment strategy, consider the problem from a task requirement-driven perspective, and establish standardized workflows and shared platforms to enable scalable, interoperable, and reproducible manufacturing AI systems.

Assessments of arthropod herbivory patterns through deep time are reliant on accurate interpretations of feeding traces on fossil plants, which may be morphologically ambiguous. Prior work has emphasised the importance of ‘reaction rims’ for identifying herbivory, where the plant's histological response forms a region of altered and often thickened tissue that both seals and surrounds the damaged area during healing. However, the potential for diagenesis to form similar rimmed structures on plant fossils has not been well evaluated. In this study, we describe a collection of rimmed, subcircular structures on Glossopteris from the Wilton Formation (Upper Permian) of New South Wales, Australia. Although their morphology is consistent with recognised insect damage types (especially galling or hole feeding), less frequent, more poorly preserved examples that occur outside the leaf margins confirm a diagenetic origin. We hypothesise that these structures are the weathered remnants of small, concentric iron sulphide or siderite growths due to their rimmed morphology, association with iron oxide, and concentrated distribution on or near the fossilised Glossopteris. Our findings emphasise that diagenesis can create structures that superficially resemble insect feeding damage. Thus, although reaction rims remain a key criterion for identifying herbivory, care must also be taken to consider diagenetic and preservational factors so as not to artificially inflate the fossil record of insect herbivory.

Global health researchers, UN agencies, and donors have become expert at documenting the deterioration of population health during conflict, yet documentation has not stopped that deterioration. Drawing on 15 years of field research in Syria, Lebanon, Gaza, and Jordan, we argue that part of the failure of the global health architecture stems not principally from violations of international humanitarian law, but from its design as a technical enterprise detached from the political economy of power, financing, and governance. In practice, the system was structured to document harm rather than to hold anyone accountable for it. We distinguish humanitarian neutrality, understood as the operational duty to assist all civilians, from technical neutrality, which excludes political accountability in ways that protects institutional relationships and funding. We define a political economy of health approach, identify specific actors capable of changing the system, and acknowledge both the real, although uneven, effects of documentation and the structural constraints, namely sovereignty, donor dependence, and entrenched interests, that any reform should confront. Finally, we propose six concrete shifts to move global health from monitoring harm and misery towards enabling accountability.