DIGITAL ACCESS TO

Problem definition: While artificial intelligence (AI) algorithms may perform well on data that are representative of the training set (inliers), they may err when extrapolating on non-representative data (outliers). How can humans and algorithms work together to make better decisions when faced with outliers and inliers? Methodology/results: We study a human-AI collaboration on prediction tasks using a bias adjustment framework, and ultimately hypothesize that humans tend towards naïve adjusting behavior: humans make adjustments to algorithmic predictions that are too similar across inliers and outliers, when ideally adjustments should be larger on outliers than inliers. In an online lab experiment, we demonstrate that participants are indeed unable to sufficiently differentiate their adjustments to an AI algorithm when faced with both inliers and outliers, leading to a 143–176% increase in their absolute deviation from the optimal prediction compared to participants who only face either all inliers or all outliers. We design a ‘warning’ (an ‘endorsement’) that alerts participants when feature values constitute outliers (inliers) and, in a second experiment, we show that this warning (endorsement) helps participants differentiate adjustments, ultimately reducing their absolute deviation from the optimal prediction by an average of 35% (28%). Furthermore, we demonstrate that deploying both interventions together reduces participants’ absolute deviation from the optimal prediction by 49%.

Managerial implications: Our work details an important behavioral bias, and identifies a simple, educational intervention that mitigates this bias. Ultimately, we hope that this work will help managers best equip their employees with the knowledge they need to succeed in a human-AI collaboration.

Purpose – To examine how a credit-bearing, curriculum-embedded graduate internship in Japan develops graduate employability and how reflection translates specialised academic knowledge into professional judgement.

Design/methodology/approach – A qualitative case study of a year-long graduate course at a university in Japan in which postgraduate students worked with industry and public-sector partners on digital transformation challenges. Questionnaires were completed by 16 students, seven instructors and four partner organisations, combining Likert items and open-ended responses; these were analysed using reflexive thematic analysis, supported by descriptive statistics used to contextualise the qualitative findings.

Findings – Students reported gains in career relevance, problem-solving and cross-disciplinary communication. Four themes emerged: authentic data work strengthened analytical capability; reflection translated specialised knowledge into professional judgement; interdisciplinary collaboration fostered boundary-crossing communication; and tensions between innovation and feasibility, autonomy and scaffolding, and academic and partner expectations became productive sites of learning.

Research limitations/implications – The study examines a single programme at one university using self-report data, so findings are analytically rather than statistically generalisable; comparative and longitudinal research is needed.

Practical implications – The case identifies four transferable design principles for graduate work-integrated learning and education–industry partnerships: curricular embedding, structured reflection, scoped authentic ambiguity and explicit interdisciplinary facilitation.

Originality/value – A rare empirical account of a year-long, curriculum-embedded graduate internship in Japan, reframing graduate employability as the development of professional judgement through structured reflection rather than the accumulation of discrete skills.

Plasmonic transition-metal nitrides such as TiN and HfN display localized surface plasmon resonance (LSPR)-driven color in dilute dispersion, but their dried powders are nearly black. Here we show that embedding TiN or HfN nanoparticles in amorphous SiO2 by a one-pot sol–gel process recovers a visible LSPR-derived coloration of HfN in the dry state, despite a heterogeneous product in which single-core, multi-core, and core-free SiO2 particles coexist. CIELAB colorimetry quantifies a shift from b* ≈ −1.4 (bare HfN) to b* ≈ −12.3 (HfN/SiO2, 3 h), a CIE 1976 color difference ΔE76 = 13.4 accompanied by a large increase in chroma; TiN, whose LSPR lies near the visible–NIR border, serves as a boundary case and shows only a small change that is dominated by lightness. A single-parameter (ωp) quasi-static dipole/Kubelka–Munk reflectance model of isolated bare and silica-coated particles—calibrating only the nitride plasma frequency to the dispersion-phase LSPR—reproduces both the featureless reflectance of bare nitride powders and the LSPR-related reflectance dip of the silica-embedded particles (∼690 nm for HfN; beyond the measured range for TiN). Within this model, the recovery arises from scattering dilution: the transparent silica shell raises the scattering cross section so that the LSPR absorption re-emerges as a wavelength-selective dip. Physical separation of the nitride domains by SiO2 may additionally limit contact-induced damping, but this contribution is not resolved here. Sol–gel silica embedding thus offers a simple route to solid-state LSPR coloration of metal-nitride nanoparticles.

Structural color materials combining photonic and plasmonic effects offer tunable optical functionalities, but most rely on noble metals. Here, a noble-metal-free platform based on SiO2–metal nitride (TiN and HfN) core–satellite particles is reported, in which photonic reflection is spectrally filtered by the localized surface plasmon resonance (LSPR) of nitride nanoparticles assembled on monodisperse SiO2 photonic building blocks. The SiO2 core size governs the photonic response, and the reflection is selectively attenuated in the spectral window set by the satellite LSPR. Contrasting TiN (LSPR at 830 nm) with HfN (LSPR at 630 nm) shows that the position of this attenuation follows the chosen resonance. The ratio of long- to short-wavelength reflection intensity, constant for bare SiO2 films, decreases systematically in both nitride systems, establishing that the attenuation is wavelength selective rather than uniform. Increasing the core diameter shifts the photonic peak by 287–312 nm, whereas changing the nitride species shifts it by only 7.2 nm. Satellite loading, compared at two levels by X-ray photoelectron spectroscopy, changes powder color qualitatively; its effect on film attenuation remains to be quantified. Together with the reported ZrN system, the three nitrides enlarge the sampled chromaticity region 5.4-fold while reducing angular dependence.

Existing MLLM-based UI-to-Code generation methods can generate complete HTML/CSS codes through end-to-end or multi-stage pipelines, yet the generated codes often contain structural errors such as incorrect repeated layouts, wrong component boundaries, or missing interaction slots. Subsequent generations may continue along the erroneous structure, causing the error to propagate and amplify. This paper studies post-generation repair: how can we locate and repair unreliable local DOM structures? We observe that this repair process is ambiguous in two ways: a mismatch between the target screenshot and the output may correspond to multiple plausible DOM subtrees, and a single DOM subtree may admit multiple repairs with different rendered behaviors. We propose Render Structure Uncertainty (RSU) for HTML repair, a browser grounded uncertainty layer that can be attached to any base HTML generator. RSU parses the target screenshot into a visual structure graph and converts the browser execution trace of the generated code into a DOM-anchored render structure graph. By matching these graphs, RSU yields structural mismatches and a small set of candidate DOM holes, preserving error attribution uncertainty. For each hole, RSU generates multiple local candidates, renders them in the full page context, and clusters them using render-structure signatures that encode layout relations, repeated patterns, content slots, semantic roles, and invalid states, thereby modeling repair uncertainty. The retained representative candidates form a compact Partial DOM candidate set, over which we perform global reranking to produce the final HTML code. In this way, RSU reframes HTML repair into uncertainty-guided search in render-structure space, avoiding premature commitment to either a single error location or a single local repair. Experiments show that RSU improves the generation quality of base models, producing higher-quality HTML with better layout consistency and fewer local structural errors. Code is available at https://anonymous.4open.science/r/Uncertainty-F61D/.