Large language models (LLMs) demonstrate impressive performance but lack the flexibility to adapt to human preferences quickly without retraining. In this work, we introduce Test-time Preference Optimization (TPO), a framework that aligns LLM outputs with human preferences during inference, removing the need to update model parameters. Rather than relying on purely numerical rewards, TPO translates reward signals into textual critiques and uses them as textual rewards to iteratively refine its response. Evaluations on benchmarks covering instruction following, preference alignment, safety, and mathematics reveal that TPO progressively improves alignment with human preferences. Notably, after only a few TPO steps, the initially unaligned Llama-3.1-70B-SFT model can surpass the aligned counterpart, Llama-3.1-70B-Instruct. Furthermore, TPO scales efficiently with both the search width and depth during inference. Through case studies, we illustrate how TPO exploits the innate capacity of LLM to interpret and act upon reward signals. Our findings establish TPO as a practical, lightweight alternative for test-time preference optimization, achieving alignment on the fly. Our code is publicly available at https://github.com/yafuly/TPO.

The notion of rank of a Boolean function has been a cornerstone in the theory of PAC learning, enabling quasipolynomial-time learning algorithms for polynomial-size decision trees. We present a novel characterization of rank, grounded in the well-known Transformer architecture. We show that the rank of a function $f$ corresponds to the minimum number of Chain of Thought (CoT) steps required by a single-layer transformer decoder with hard attention to compute $f$. Based on this characterization we establish tight bounds on the number of CoT steps required for specific problems, showing that $\ell$-fold function composition necessitates exactly $\ell$ CoT steps. Furthermore, we analyze the problem of identifying the position of the $k$-th occurrence of 1 in a Boolean sequence, proving that it requires $k$ CoT steps.

Context: Empirical Software Engineering (ESE) drives innovation in SE through qualitative and quantitative studies. However, concerns about the correct application of empirical methodologies have existed since the 2006 Dagstuhl seminar on SE. Objective: To analyze three decades of SE research, identify mistakes in statistical methods, and evaluate experts' ability to detect and address these issues. Methods: We conducted a literature survey of ~27,000 empirical studies, using LLMs to classify statistical methodologies as adequate or inadequate. Additionally, we selected 30 primary studies and held a workshop with 33 ESE experts to assess their ability to identify and resolve statistical issues. Results: Significant statistical issues were found in the primary studies, and experts showed limited ability to detect and correct these methodological problems, raising concerns about the broader ESE community's proficiency in this area. Conclusions. Despite our study's eventual limitations, its results shed light on recurring issues from promoting information copy-and-paste from past authors' works and the continuous publication of inadequate approaches that promote dubious results and jeopardize the spread of the correct statistical strategies among researchers. Besides, it justifies further investigation into empirical rigor in software engineering to expose these recurring issues and establish a framework for reassessing our field's foundation of statistical methodology application. Therefore, this work calls for critically rethinking and reforming data analysis in empirical software engineering, paving the way for our work soon.

Punctuation is the main factor introducing correlations in natural language written texts and it crucially impacts their overall effectiveness, expressiveness, and readability. Punctuation marks at the end of sentences are of particular importance as their distribution can determine various complexity features of written natural language. Here, the sentence length variability (SLV) time series representing "Hopscotch" by Julio Cortazar are subjected to quantitative analysis with an attempt to identify their distribution type, long-memory effects, and potential multiscale patterns. The analyzed novel is an important and innovative piece of literature whose essential property is freedom of movement between its building blocks given to a reader by the author. The statistical consequences of this freedom are closely investigated in both the original, Spanish version of the novel, and its translations into English and Polish. Clear evidence of rich multifractality in the SLV dynamics, with a left-sided asymmetry, however, is observed in all three language versions as well as in the versions with differently ordered chapters.

Complementary collaboration between humans and AI is essential for human-AI decision making. One feasible approach to achieving it involves accounting for the calibrated confidence levels of both AI and users. However, this process would likely be made more difficult by the fact that AI confidence may influence users' self-confidence and its calibration. To explore these dynamics, we conducted a randomized behavioral experiment. Our results indicate that in human-AI decision-making, users' self-confidence aligns with AI confidence and such alignment can persist even after AI ceases to be involved. This alignment then affects users' self-confidence calibration. We also found the presence of real-time correctness feedback of decisions reduced the degree of alignment. These findings suggest that users' self-confidence is not independent of AI confidence, which practitioners aiming to achieve better human-AI collaboration need to be aware of. We call for research focusing on the alignment of human cognition and behavior with AI.

The integration of Artificial Intelligence (AI) systems into technologies used by young digital citizens raises significant privacy concerns. This study investigates these concerns through a comparative analysis of stakeholder perspectives. A total of 252 participants were surveyed, with the analysis focusing on 110 valid responses from parents/educators and 100 from AI professionals after data cleaning. Quantitative methods, including descriptive statistics and Partial Least Squares Structural Equation Modeling, examined five validated constructs: Data Ownership and Control, Parental Data Sharing, Perceived Risks and Benefits, Transparency and Trust, and Education and Awareness. Results showed Education and Awareness significantly influenced data ownership and risk assessment, while Data Ownership and Control strongly impacted Transparency and Trust. Transparency and Trust, along with Perceived Risks and Benefits, showed minimal influence on Parental Data Sharing, suggesting other factors may play a larger role. The study underscores the need for user-centric privacy controls, tailored transparency strategies, and targeted educational initiatives. Incorporating diverse stakeholder perspectives offers actionable insights into ethical AI design and governance, balancing innovation with robust privacy protections to foster trust in a digital age.

Fox News Flash top sports headlines are here. Check out what's clicking on Foxnews.com. Erin Routliffe may be the only person unhappy that robots haven't taken over the world. The tennis pro from New Zealand and her partner, Gabriela Dabrowski of Canada, were competing in the third round of women's doubles at the Australian Open Sunday when Routliffe exploded into a brief tirade after she believed her opponent's serve skimmed the net. Erin Routliffe of New Zealand and Gabriela Dabrowski of Canada in action against Laura Siegemund of Germany and Beatriz Haddad Maia of Brazil in the third round of women's doubles at the 2025 Australian Open at Melbourne Park Jan. 20, 2025, in Melbourne, Australia. The contentious point came during a tiebreak in the first set with Beatriz Haddad Maia of Brazil serving.

Supervised classification techniques use training samples to find classification rules with small expected 0-1 loss. Conventional methods achieve efficient learning and out-of-sample generalization by minimizing surrogate losses over specific families of rules. This paper presents minimax risk classifiers (MRCs) that do not rely on a choice of surrogate loss and family of rules. MRCs achieve efficient learning and out-of-sample generalization by minimizing worst-case expected 0-1 loss w.r.t. In addition, MRCs' learning stage provides performance guarantees as lower and upper tight bounds for expected 0-1 loss.

Large language models (LLMs) have been positioned to revolutionize HCI, by reshaping not only the interfaces, design patterns, and sociotechnical systems that we study, but also the research practices we use. To-date, however, there has been little understanding of LLMs' uptake in HCI. We address this gap via a systematic literature review of 153 CHI papers from 2020-24 that engage with LLMs. We taxonomize: (1) domains where LLMs are applied; (2) roles of LLMs in HCI projects; (3) contribution types; and (4) acknowledged limitations and risks. We find LLM work in 10 diverse domains, primarily via empirical and artifact contributions. Authors use LLMs in five distinct roles, including as research tools or simulated users. Still, authors often raise validity and reproducibility concerns, and overwhelmingly study closed models. We outline opportunities to improve HCI research with and on LLMs, and provide guiding questions for researchers to consider the validity and appropriateness of LLM-related work.

Foundation models have made rapid advances in many domains including Earth observation, where Geospatial Foundation Models (GFMs) can help address global challenges such as climate change, agriculture, and disaster response. Previous work on GFMs focused on tailoring model architecture and pre-text tasks, and did not investigate the impact of pre-training data selection on model performance. However, recent works from other domains show that the pre-training data distribution is an important factor influencing the performance of the foundation models. With this motivation, our research explores how the geographic distribution of pre-training data affects the performance of GFMs. We evaluated several pre-training data distributions by sampling different compositions from a global data pool. Our experiments with two GFMs on downstream tasks indicate that balanced and globally representative data compositions often outperform region-specific sampling, highlighting the importance of diversity and global coverage in pre-training data. Our results suggest that the most appropriate data sampling technique may depend on the specific GFM architecture. These findings will support the development of robust GFMs by incorporating quality pre-training data distributions, ultimately improving machine learning solutions for Earth observation.