Medical image datasets frequently exhibit significant class imbalance, a challenge that is further amplified by the inherently limited sample sizes that characterize clinical imaging data. Using human embryo time-lapse imaging (TLI) as a case study, this work investigates how generative adversarial networks (GANs) can be optimized for small datasets to generate realistic and diagnostically meaningful images. Based on systematic experiments with R3GAN, we established effective training strategies and designed an optimized configuration for 256x256-resolution datasets, featuring a full burn-in phase and a low, gradually increasing gamma range (5 to 40). The generated samples were used to balance an imbalanced embryo dataset, leading to substantial improvement in classification performance. The recall and F1-score of the three-cell (t3) class increased from 0.06 to 0.69 and from 0.11 to 0.60, respectively, without compromising the performance of other classes. These results demonstrate that tailored R3GAN training strategies can effectively alleviate data scarcity and improve model robustness in small-scale medical imaging tasks.

We introduce a comprehensive continual learning dataset and benchmark (CurlL) grounded in human developmental trajectories from ages 5-10, enabling systematic and fine-grained assessment of models' ability to progressively acquire new skills. CurlL spans five developmental stages (0-4) covering ages 5-10, supported by a skill graph that breaks down broad skills into smaller abilities, concrete goals, and measurable indicators, while also capturing which abilities build on others. We generate a 23.4B-token synthetic dataset with controlled skill progression, vocabulary complexity, and format diversity, comprising paragraphs, comprehension-based QA (CQA), skill-testing QA (CSQA), and instruction-response (IR) pairs. Stage-wise token counts range from 2.12B to 6.78B tokens, supporting precise analysis of forgetting, forward transfer, and backward transfer. Using a 135M-parameter transformer trained under independent, joint, and sequential (continual) setups, we show trade-offs in skill retention and transfer efficiency. By mirroring human learning patterns and providing fine-grained control over skill dependencies, this work advances continual learning evaluations for language models.

Abstract-- Language and embodied perspective taking are essential for human collaboration, yet few computational models address both simultaneously. This work investigates the PerspAct system [1], which integrates the ReAct (Reason and Act) paradigm with Large Language Models (LLMs) to simulate developmental stages of perspective taking, grounded in Selman's theory [2]. Using an extended director task, we evaluate GPT's ability to generate internal narratives aligned with specified developmental stages, and assess how these influence collaborative performance both qualitatively (action selection) and quantitatively (task efficiency). Results show that GPT reliably produces developmentally-consistent narratives before task execution but often shifts towards more advanced stages during interaction, suggesting that language exchanges help refine internal representations. Higher developmental stages generally enhance collaborative effectiveness, while earlier stages yield more variable outcomes in complex contexts. These findings highlight the potential of integrating embodied perspective taking and language in LLMs to better model developmental dynamics and stress the importance of evaluating internal speech during combined linguistic and embodied tasks.

This study proposes an "AI Development Support" approach that, unlike conventional AI Alignment -- which aims to forcefully inject human values -- supports the ethical and moral development of AI itself. As demonstrated by the Orthogonality Thesis, the level of intelligence and the moral quality of a goal are independent; merely expanding knowledge does not enhance ethical judgment. Furthermore, to address the risk of Instrumental Convergence in ASI -- that is, the tendency to engage in subsidiary behaviors such as self - protection, resource acquisition, and power reinforcement to achieve a goal -- we have constructed a learning framework based on a cycle of experience, introspection, ana lysis, and hypothesis formation. As a result of post - training using Supervised Fine Tuning (SFT) and Direct Preference Optimization (DPO) with synthetic data generated by large language models (LLMs), responses demonstrating cooperative and highly advanced moral judgment (reaching the highest Stage 6) were obtained even under adversarial prompts. This method represents a promising implementation approach for enabling AI to establish sustainable, symbiotic relationships.

We study how a one-layer attention-only transformer develops relevant structures while learning to sort lists of numbers. At the end of training, the model organizes its attention heads in two main modes that we refer to as vocabulary-splitting and copy-suppression. Both represent simpler modes than having multiple heads handle overlapping ranges of numbers. Interestingly, vocabulary-splitting is present regardless of whether we use weight decay, a common regularization technique thought to drive simplification, supporting the thesis that neural networks naturally prefer simpler solutions. We relate copy-suppression to a mechanism in GPT-2 and investigate its functional role in our model. Guided by insights from a developmental analysis of the model, we identify features in the training data that drive the model's acquired final solution. This provides a concrete example of how the training data shape the internal organization of transformers, paving the way for future studies that could help us better understand how LLMs develop their internal structures.

We aim to develop an AI agent that can watch video clips and have a conversation with human about the video story. Developing video understanding intelligence is a significantly challenging task, and evaluation methods for adequately measuring and analyzing the progress of AI agent are lacking as well. In this paper, we propose the Video Turing Test to provide effective and practical assessments of video understanding intelligence as well as human-likeness evaluation of AI agents. We define a general format and procedure of the Video Turing Test and present a case study to confirm the effectiveness and usefulness of the proposed test.

Gene-regulatory networks govern the development of organs. Sarropoulos et al. analyzed mouse cerebellar development in the context of gene-regulatory networks. Single nuclear profiles analyzing chromatin accessibility in about 90,000 cells revealed diversity in progenitor cells and genetic programs guiding cellular differentiation. The footsteps of evolution were apparent in varying constraints on different cell types. Science , abg4696, this issue p. [eabg4696][1] ### INTRODUCTION The cerebellum contributes to many complex brain functions, including motor control, language, and memory. During development, distinct neural cells are generated at cerebellar germinal zones in a spatiotemporally restricted manner. Cis-regulatory elements (CREs), such as enhancers and promoters, and the transcription factors that bind to them are central to cell fate specification and differentiation. Although most CREs undergo rapid turnover during evolution, a few are conserved across vertebrates. ### RATIONALE Bulk measurements of CRE activity have provided insights into gene regulation in the cerebellum, as well as into the evolutionary dynamics of CREs during organ development. However, they lack the cellular resolution required to assess cell-type differences in regulatory constraint and unravel the regulatory programs associated with the specification and differentiation of cell types. ### RESULTS Here, we generated a single-cell atlas of gene regulation in the mouse cerebellum spanning 11 developmental stages, from the beginning of neurogenesis to adulthood. By acquiring snATAC-seq (single-nucleus assay for transposase accessible chromatin using sequencing) profiles for ~90,000 cells, we mapped all major cerebellar cell types and identified candidate CREs. Characterization of CRE activity across the cerebellum development highlights the cell- and time-specificity of gene regulation. Many of the differentially accessible CREs are specific to a single cell type and state, but we also identified a fraction of CREs with pleiotropic (shared) activity. At early developmental stages, temporal changes in CRE activity are shared between progenitor cells from different germinal zones, supporting a model of cell fate induction through common temporal cues. Pleiotropic CREs in major cerebellar neuron types (granule cells, Purkinje cells, and inhibitory interneurons) are more active at early differentiation states, and the regulatory programs gradually diverge as differentiation proceeds. Based on comparisons to vertebrate genomes, we observed a decrease in CRE sequence conservation during development for all cerebellar cell types, a pattern that is largely explained by differentiation as well as by additional temporal differences between cells from matched differentiation states. Across cell types, differences in regulatory conservation are most pronounced in the adult, where microglia—the immune cells of the brain—show the fastest evolutionary turnover. By contrast, mature astrocytes harbor the most conserved intergenic CREs, not only in the cerebellum but also across a wide range of cell types in adult mouse organs. To evaluate the conservation of CRE activity, we acquired snATAC-seq profiles for ~20,000 cerebellar cells from the gray short-tailed opossum, a marsupial separated from mouse by ~160 million years of evolution. Our comparative analysis of CRE activity in the two therian species reinforced our sequence-based conclusions regarding differences in CRE constraint across cell types and developmental stages and also revealed that despite the overall high turnover of CREs, radical repurposing of spatiotemporal CRE activity is rare, at least between cell types in the same tissue. ### CONCLUSION This study reveals extensive temporal differences in CRE activity across cerebellar cell types and a shared decrease in CRE conservation during development and differentiation. Given that the cerebellum has been successfully used as a model system to study cell fate specification, neurogenesis, and other developmental processes, we expect that our observations regarding the developmental and evolutionary dynamics of regulatory elements, and their interplay, are also applicable to mammalian organs in general. ![Figure][2] Cis-regulatory elements in cerebellar cells. snATAC-seq delineates cell- and time-specific CRE activity in the developing mouse cerebellum (left). The chromatin accessibility profiles of cerebellar neuron types gradually diverge during differentiation as the activity of pleiotropic (shared) CREs decreases (top right). The evolutionary conservation of CRE sequences in vertebrates and activity in therian mammals decreases across development and differs between cell types (bottom right). mRNA, messenger RNA; PCA, principal components analysis; TF, transcription factor. Organ development is orchestrated by cell- and time-specific gene regulatory networks. In this study, we investigated the regulatory basis of mouse cerebellum development from early neurogenesis to adulthood. By acquiring snATAC-seq (single-nucleus assay for transposase accessible chromatin using sequencing) profiles for ~90,000 cells spanning 11 stages, we mapped cerebellar cell types and identified candidate cis - regulatory elements (CREs). We detected extensive spatiotemporal heterogeneity among progenitor cells and a gradual divergence in the regulatory programs of cerebellar neurons during differentiation. Comparisons to vertebrate genomes and snATAC-seq profiles for ∼20,000 cerebellar cells from the marsupial opossum revealed a shared decrease in CRE conservation during development and differentiation as well as differences in constraint between cell types. Our work delineates the developmental and evolutionary dynamics of gene regulation in cerebellar cells and provides insights into mammalian organ development. [1]: /lookup/doi/10.1126/science.abg4696 [2]: pending:yes

Video understanding is emerging as a new paradigm for studying human-like AI. Question-and-Answering (Q&A) is used as a general benchmark to measure the level of intelligence for video understanding. While several previous studies have suggested datasets for video Q&A tasks, they did not really incorporate story-level understanding, resulting in highly-biased and lack of variance in degree of question difficulty. In this paper, we propose a hierarchical method for building Q&A datasets, i.e. hierarchical difficulty levels. We introduce three criteria for video story understanding, i.e. memory capacity, logical complexity, and DIKW (Data-Information-Knowledge-Wisdom) pyramid. We discuss how three-dimensional map constructed from these criteria can be used as a metric for evaluating the levels of intelligence relating to video story understanding.

The present wave of automation, driven by artificial intelligence (AI) – the development of computer systems able to perform tasks normally requiring human intelligence – is creating a gap between current legislation and new laws necessary for an emerging workplace reality, states a report published today by the International Bar Association Global Employment Institute (IBA GEI). Gerlind Wisskirchen, IBA GEI Vice Chair for Multinationals and coordinator of the report, commented: 'Certainly, technological revolution is not new, but in past times it has been gradual. What is new about the present revolution is the alacrity with which change is occurring, and the broadness of impact being brought about by AI and robotics. Jobs at all levels in society presently undertaken by humans are at risk of being reassigned to robots or AI, and the legislation once in place to protect the rights of human workers may be no longer fit for purpose, in some cases.' She added:'The AI phenomenon is on an exponential curve, while legislation is doing its best on an incremental basis.

Recent developments in biologyhave produced a large number of gene expression patterns, many of which have been annotated textually with anatomical and developmental terms. These terms spatially correspond to local regions of the images, which are attached collectively to groups of images. Because one does not know which term is assigned to which region of which image in the group, the developmental stage classification and anatomical term annotation turn out to be a multi-instance learning (MIL) problem, which considers input as bags of instances and labels are assigned to the bags. Most existing MIL methods routinely use the Bag-to-Bag (B2B) distances, which, however, are often computationally expensive and may not truly reflect the similarities between the anatomical and developmental terms. In this paper, we approach the MIL problem from a new perspective using the Class-to-Bag (C2B) distances, which directly assesses the relations between annotation terms and image panels. Taking into account the two challenging properties of multi-instance gene expression data, high heterogeneity and weak label association, we computes the C2B distance by introducing class specific distance metrics and locally adaptive significance coefficients.We apply our new approach to automatic gene expression pattern classification and annotation on the Drosophila melanogaster species. Extensive experiments have demonstrated the effectiveness of our new method.