Language model finetuning often results in learning undesirable traits in combination with desired ones. To address this, we propose inoculation prompting: modifying finetuning data by prepending a short system-prompt instruction that deliberately elicits the undesirable trait. At test time, we evaluate without the instruction; inoculated models have much lower expression of the trait than models trained with unmodified training data. Inoculation is selective: in a toy setting where assistant responses are always in Spanish and ALL-CAPS, an appropriate inoculation (e.g., ``You always speak in Spanish.'') teaches the model to capitalize responses while still responding in English. We find that inoculation is also effective across several additional settings: reducing emergent misalignment (EM) from task-specific finetuning, defending against backdoor injections, and mitigating the transmission of traits via subliminal learning. Follow-up analysis suggests a mechanism: making a trait less surprising via inoculation reduces optimization pressure to globally update the model, thereby reducing the degree of generalization. Our analysis relates to prior work on EM: inoculation explains prior findings that educational contexts mitigate EM from insecure code. Beyond demonstrating a simple and effective technique for selective learning, our results contribute to a better conceptual understanding of how and why language models generalize.

Multimodal Large Language Models (MLLMs) are increasingly applied in Personalized Image Aesthetic Assessment (PIAA) as a scalable alternative to expert evaluations. However, their predictions may reflect subtle biases influenced by demographic factors such as gender, age, and education. In this work, we propose AesBiasBench, a benchmark designed to evaluate MLLMs along two complementary dimensions: (1) stereotype bias, quantified by measuring variations in aesthetic evaluations across demographic groups; and (2) alignment between model outputs and genuine human aesthetic preferences. Our benchmark covers three subtasks (Aesthetic Perception, Assessment, Empathy) and introduces structured metrics (IFD, NRD, AAS) to assess both bias and alignment. We evaluate 19 MLLMs, including proprietary models (e.g., GPT-4o, Claude-3.5-Sonnet) and open-source models (e.g., InternVL-2.5, Qwen2.5-VL). Results indicate that smaller models exhibit stronger stereotype biases, whereas larger models align more closely with human preferences. Incorporating identity information often exacerbates bias, particularly in emotional judgments. These findings underscore the importance of identity-aware evaluation frameworks in subjective vision-language tasks.

Human aesthetic preferences for 3D shapes are central to industrial design, virtual reality, and consumer product development. However, most computational models of 3D aesthetics lack empirical grounding in large-scale human judgments, limiting their practical relevance. We present a large-scale study of human preferences. We collected 22,301 pairwise comparisons across five object categories (chairs, tables, mugs, lamps, and dining chairs) via Amazon Mechanical Turk. Building on a previously published dataset~\cite{dev2020learning}, we introduce new non-linear modeling and cross-category analysis to uncover the geometric drivers of aesthetic preference. We apply the Bradley-Terry model to infer latent aesthetic scores and use Random Forests with SHAP analysis to identify and interpret the most influential geometric features (e.g., symmetry, curvature, compactness). Our cross-category analysis reveals both universal principles and domain-specific trends in aesthetic preferences. We focus on human interpretable geometric features to ensure model transparency and actionable design insights, rather than relying on black-box deep learning approaches. Our findings bridge computational aesthetics and cognitive science, providing practical guidance for designers and a publicly available dataset to support reproducibility. This work advances the understanding of 3D shape aesthetics through a human-centric, data-driven framework.

Interior design is all about creating spaces that look and feel good. However, the subjective nature of aesthetic preferences presents a significant challenge in defining and quantifying what makes an interior design visually appealing. The current paper addresses this gap by introducing a novel methodology for quantifying and predicting aesthetic preferences in interior design. Our study combines fuzzy logic with image processing techniques. We collected a dataset of interior design images from social media platforms, focusing on essential visual attributes such as color harmony, lightness, and complexity. We integrate these features using weighted average to compute a general aesthetic score. Our approach considers individual color preferences in calculating the overall aesthetic preference. We initially gather user ratings for primary colors like red, brown, and others to understand their preferences. Then, we use the pixel count of the top five dominant colors in the image to get the color scheme preference. The color scheme preference and the aesthetic score are then passed as inputs to the fuzzy inference system to calculate an overall preference score. This score represents a comprehensive measure of the user's preference for a particular interior design, considering their color choices and general aesthetic appeal. We used the 2AFC (Two-Alternative Forced Choice) method to validate our methodology, achieving a notable hit rate of 0.7. This study can help designers and professionals better understand and meet people's interior design preferences, especially in a world that relies heavily on digital media.

The allure of aesthetic appeal in images captivates our senses, yet the underlying intricacies of aesthetic preferences remain elusive. In this study, we pioneer a novel perspective by utilizing machine learning models that focus on aesthetic attributes known to influence preferences. Through a data mining approach, our models process these attributes as inputs to predict the aesthetic scores of images. Moreover, to delve deeper and obtain interpretable explanations regarding the factors driving aesthetic preferences, we utilize the popular Explainable AI (XAI) technique known as SHapley Additive exPlanations (SHAP). Our methodology involves employing various machine learning models, including Random Forest, XGBoost, Support Vector Regression, and Multilayer Perceptron, to compare their performances in accurately predicting aesthetic scores, and consistently observing results in conjunction with SHAP. We conduct experiments on three image aesthetic benchmarks, providing insights into the roles of attributes and their interactions. Ultimately, our study aims to shed light on the complex nature of aesthetic preferences in images through machine learning and provides a deeper understanding of the attributes that influence aesthetic judgements.

This work proposes aesthetic gradients, a method to personalize a CLIP-conditioned diffusion model by guiding the generative process towards custom aesthetics defined by the user from a set of images. The approach is validated with qualitative and quantitative experiments, using the recent stable diffusion model and several aesthetically-filtered datasets. Code is released at https://github.com/

Analyst Gartner put out a 10-strong listicle this week identifying what it dubbed "high-impact" uses for AI-powered features on smartphones that it suggests will enable device vendors to provide "more value" to customers via the medium of "more advanced" user experiences. It's also predicting that, by 2022, a full 80 per cent of smartphones shipped will have on-device AI capabilities, up from just 10 per cent in 2017. More on-device AI could result in better data protection and improved battery performance, in its view -- as a consequence of data being processed and stored locally. Its full list of apparently enticing AI uses is presented (verbatim) below. But in the interests of presenting a more balanced narrative around automation-powered UXes we've included some alternative thoughts after each listed item which consider the nature of the value exchange being required for smartphone users to tap into these touted'AI smarts' -- and thus some potential drawbacks too.

Artificial intelligence features will become a critical product differentiator for smartphone vendors in acquiring new customers, according to Gartner. As the smartphone market shifts from selling technology products to delivering personalised experiences, AI solutions running on smartphones will become an essential part of vendor roadmaps over the next two years. Gartner predicts that by 2022, 80 percent of smartphones shipped will have on-device AI capabilities, up from 10 percent in 2017. On-device AI is currently limited to premium devices, and provides better data protection and power management than full cloud-based AI, since data is processed and stored locally. "With smartphones increasingly becoming a commodity device, vendors are looking for ways to differentiate their products," said CK Lu, research director at Gartner.

Analyst Gartner put out a 10-strong listicle this week identifying what it dubbed "high-impact" uses for AI-powered features on smartphones that it suggests will enable device vendors to provide "more value" to customers via the medium of "more advanced" user experiences. It's also predicting that, by 2022, a full 80 per cent of smartphones shipped will have on-device AI capabilities, up from just 10 per cent in 2017. More on-device AI could result in better data protection and improved battery performance, in its view -- as a consequence of data being processed and stored locally. Its full list of apparently enticing AI uses is presented (verbatim) below. But in the interests of presenting a more balanced narrative around automation-powered UXes we've included some alternative thoughts after each listed item which consider the nature of the value exchange being required for smartphone users to tap into these touted'AI smarts' -- and thus some potential drawbacks too.

Artificial intelligence (AI) features will become a critical product differentiator for smartphone vendors that will help them to acquire new customers while retaining current users, according to Gartner, Inc. As the smartphone market shifts from selling technology products to delivering compelling and personalized experiences, AI solutions running on the smartphone will become an essential part of vendor roadmaps over the next two years. Gartner predicts that by 2022, 80 percent of smartphones shipped will have on-device AI capabilities, up from 10 percent in 2017. On-device AI is currently limited to premium devices and provides better data protection and power management than full cloud-based AI, since data is processed and stored locally. "With smartphones increasingly becoming a commodity device, vendors are looking for ways to differentiate their products," said CK Lu, research director at Gartner.