Language models (LMs) have demonstrated remarkable profi- ciency in generating linguistically coherent text, sparking dis- cussions about their relevance to understanding human lan- guage learnability. However, a significant gap exists between the training data for these models and the linguistic input a child receives. LMs are typically trained on data that is or- ders of magnitude larger and fundamentally different from child-directed speech (Warstadt & Bowman, 2022; Warstadt et al., 2023; Frank, 2023a). Addressing this discrepancy, our research focuses on training LMs on subsets of a sin- gle child's linguistic input. Previously, Wang, Vong, Kim, and Lake (2023) found that LMs trained in this setting can form syntactic and semantic word clusters and develop sen- sitivity to certain linguistic phenomena, but they only consid- ered LSTMs and simpler neural networks trained from just one single-child dataset. Here, to examine the robustness of learn- ability from single-child input, we systematically train six dif- ferent model architectures on five datasets (3 single-child and 2 baselines). We find that the models trained on single-child datasets showed consistent results that matched with previous work, underscoring the robustness of forming meaningful syn- tactic and semantic representations from a subset of a child's linguistic input. Keywords: learnability; single-child; distributional learning; robustness; language models

Children learn powerful internal models of the world around them from a few years of egocentric visual experience. Can such internal models be learned from a child's visual experience with highly generic learning algorithms or do they require strong inductive biases? Recent advances in collecting large-scale, longitudinal, developmentally realistic video datasets and generic self-supervised learning (SSL) algorithms are allowing us to begin to tackle this nature vs. nurture question. However, existing work typically focuses on image-based SSL algorithms and visual capabilities that can be learned from static images (e.g. object recognition), thus ignoring temporal aspects of the world. To close this gap, here we train self-supervised video models on longitudinal, egocentric headcam recordings collected from a child over a two year period in their early development (6-31 months). The resulting models are highly effective at facilitating the learning of action concepts from a small number of labeled examples; they have favorable data size scaling properties; and they display emergent video interpolation capabilities. Video models also learn more robust object representations than image-based models trained with the exact same data. These results suggest that important temporal aspects of a child's internal model of the world may be learnable from their visual experience using highly generic learning algorithms and without strong inductive biases.

The interleukin-4 (IL-4) cytokine plays a critical role in modulating immune homeostasis. Although there is great interest in harnessing this cytokine as a therapeutic in natural or engineered formats, the clinical potential of native IL-4 is limited by its instability and pleiotropic actions. Here, we design IL-4 cytokine mimetics (denoted Neo-4) based on a de novo engineered IL-2 mimetic scaffold and demonstrate that these cytokines can recapitulate physiological functions of IL-4 in cellular and animal models. In contrast with natural IL-4, Neo-4 is hyperstable and signals exclusively through the type I IL-4 receptor complex, providing previously inaccessible insights into differential IL-4 signaling through type I versus type II receptors. Because of their hyperstability, our computationally designed mimetics can directly incorporate into sophisticated biomaterials that require heat processing, such as three-dimensional-printed scaffolds. Neo-4 should be broadly useful for interrogating IL-4 biology, and the design workflow will inform targeted cytokine therapeutic development.