Some traits making a "good" AI model are hard to describe upfront. For example, should responses be more polite or more casual? Such traits are sometimes summarized as model character or personality. Without a clear objective, conventional benchmarks based on automatic validation struggle to measure such traits. Evaluation methods using human feedback such as Chatbot Arena have emerged as a popular alternative. These methods infer "better" personality and other desirable traits implicitly by ranking multiple model responses relative to each other. Recent issues with model releases highlight limitations of these existing opaque evaluation approaches: a major model was rolled back over sycophantic personality issues, models were observed overfitting to such feedback-based leaderboards. Despite these known issues, limited public tooling exists to explicitly evaluate model personality. We introduce Feedback Forensics: an open-source toolkit to track AI personality changes, both those encouraged by human (or AI) feedback, and those exhibited across AI models trained and evaluated on such feedback. Leveraging AI annotators, our toolkit enables investigating personality via Python API and browser app. We demonstrate the toolkit's usefulness in two steps: (A) first we analyse the personality traits encouraged in popular human feedback datasets including Chatbot Arena, MultiPref and PRISM; and (B) then use our toolkit to analyse how much popular models exhibit such traits. We release (1) our Feedback Forensics toolkit alongside (2) a web app tracking AI personality in popular models and feedback datasets as well as (3) the underlying annotation data at https://github.com/rdnfn/feedback-forensics.

The global dimensionality of a neural representation manifold provides rich insight into the computational process underlying both artificial and biological neural networks. However, all existing measures of global dimensionality are sensitive to the number of samples, i.e., the number of rows and columns of the sample matrix. We show that, in particular, the participation ratio of eigenvalues, a popular measure of global dimensionality, is highly biased with small sample sizes, and propose a bias-corrected estimator that is more accurate with finite samples and with noise. On synthetic data examples, we demonstrate that our estimator can recover the true known dimensionality. We apply our estimator to neural brain recordings, including calcium imaging, electrophysiological recordings, and fMRI data, and to the neural activations in a large language model and show our estimator is invariant to the sample size. Finally, our estimators can additionally be used to measure the local dimensionalities of curved neural manifolds by weighting the finite samples appropriately.

The relationship between computing systems and the brain has served as motivation for pioneering theoreticians since John von Neumann and Alan Turing. Uniform, scale-free biological networks, such as the brain, have powerful properties, including generalizing over time, which is the main barrier for Machine Learning on the path to Universal Reasoning Models. We introduce `Dragon Hatchling' (BDH), a new Large Language Model architecture based on a scale-free biologically inspired network of \$n\$ locally-interacting neuron particles. BDH couples strong theoretical foundations and inherent interpretability without sacrificing Transformer-like performance. BDH is a practical, performant state-of-the-art attention-based state space sequence learning architecture. In addition to being a graph model, BDH admits a GPU-friendly formulation. It exhibits Transformer-like scaling laws: empirically BDH rivals GPT2 performance on language and translation tasks, at the same number of parameters (10M to 1B), for the same training data. BDH can be represented as a brain model. The working memory of BDH during inference entirely relies on synaptic plasticity with Hebbian learning using spiking neurons. We confirm empirically that specific, individual synapses strengthen connection whenever BDH hears or reasons about a specific concept while processing language inputs. The neuron interaction network of BDH is a graph of high modularity with heavy-tailed degree distribution. The BDH model is biologically plausible, explaining one possible mechanism which human neurons could use to achieve speech. BDH is designed for interpretability. Activation vectors of BDH are sparse and positive. We demonstrate monosemanticity in BDH on language tasks. Interpretability of state, which goes beyond interpretability of neurons and model parameters, is an inherent feature of the BDH architecture.

The practical use of future climate projections from global circulation models (GCMs) is often limited by their coarse spatial resolution, requiring downscaling to generate high-resolution data. Regional climate models (RCMs) provide this refinement, but are computationally expensive. To address this issue, machine learning models can learn the downscaling function, mapping coarse GCM outputs to high-resolution fields. Among these, generative approaches aim to capture the full conditional distribution of RCM data given coarse-scale GCM data, which is characterized by large variability and thus challenging to model accurately. We introduce EnScale, a generative machine learning framework that emulates the full GCM-to-RCM map by training on multiple pairs of GCM and corresponding RCM data. It first adjusts large-scale mismatches between GCM and coarsened RCM data, followed by a super-resolution step to generate high-resolution fields. Both steps employ generative models optimized with the energy score, a proper scoring rule. Compared to state-of-the-art ML downscaling approaches, our setup reduces computational cost by about one order of magnitude. EnScale jointly emulates multiple variables -- temperature, precipitation, solar radiation, and wind -- spatially consistent over an area in Central Europe. In addition, we propose a variant EnScale-t that enables temporally consistent downscaling. We establish a comprehensive evaluation framework across various categories including calibration, spatial structure, extremes, and multivariate dependencies. Comparison with diverse benchmarks demonstrates EnScale's strong performance and computational efficiency. EnScale offers a promising approach for accurate and temporally consistent RCM emulation.

Diffusion models (DMs) excel in image generation, but suffer from slow inference and the training-inference discrepancies. Although gradient-based solvers like DPM-Solver accelerate the denoising inference, they lack theoretical foundations in information transmission efficiency. In this work, we introduce an information-theoretic perspective on the inference processes of DMs, revealing that successful denoising fundamentally reduces conditional entropy in reverse transitions. This principle leads to our key insights into the inference processes: (1) data prediction parameterization outperforms its noise counterpart, and (2) optimizing conditional variance offers a reference-free way to minimize both transition and reconstruction errors. Based on these insights, we propose an entropy-aware variance optimized method for the generative process of DMs, called EVODiff, which systematically reduces uncertainty by optimizing conditional entropy during denoising. Extensive experiments on DMs validate our insights and demonstrate that our method significantly and consistently outperforms state-of-the-art (SOTA) gradient-based solvers. For example, compared to the DPM-Solver++, EVODiff reduces the reconstruction error by up to 45.5\% (FID improves from 5.10 to 2.78) at 10 function evaluations (NFE) on CIFAR-10, cuts the NFE cost by 25\% (from 20 to 15 NFE) for high-quality samples on ImageNet-256, and improves text-to-image generation while reducing artifacts. Code is available at https://github.com/ShiguiLi/EVODiff.

With advances in generative AI, decision-making agents can now dynamically create new actions during online learning, but action generation typically incurs costs that must be balanced against potential benefits. We study an online learning problem where an agent can generate new actions at any time step by paying a one-time cost, with these actions becoming permanently available for future use. The challenge lies in learning the optimal sequence of two-fold decisions: which action to take and when to generate new ones, further complicated by the triangular tradeoffs among exploitation, exploration and $\textit{creation}$. To solve this problem, we propose a doubly-optimistic algorithm that employs Lower Confidence Bounds (LCB) for action selection and Upper Confidence Bounds (UCB) for action generation. Empirical evaluation on healthcare question-answering datasets demonstrates that our approach achieves favorable generation-quality tradeoffs compared to baseline strategies. From theoretical perspectives, we prove that our algorithm achieves the optimal regret of $O(T^{\frac{d}{d+2}}d^{\frac{d}{d+2}} + d\sqrt{T\log T})$, providing the first sublinear regret bound for online learning with expanding action spaces.

We consider inferring the causal effect of a treatment (intervention) on an outcome of interest in situations where there is potentially an unobserved confounder influencing both the treatment and the outcome. This is achievable by assuming access to two separate sets of control (proxy) measurements associated with treatment and outcomes, which are used to estimate treatment effects through a function termed the em causal bridge (CB). We present a new theoretical perspective, associated assumptions for when estimating treatment effects with the CB is feasible, and a bound on the average error of the treatment effect when the CB assumptions are violated. From this new perspective, we then demonstrate how coupling the CB with an autoencoder architecture allows for the sharing of statistical strength between observed quantities (proxies, treatment, and outcomes), thus improving the quality of the CB estimates. Experiments on synthetic and real-world data demonstrate the effectiveness of the proposed approach in relation to the state-of-the-art methodology for proxy measurements.

In language tasks that require extensive human--model interaction, deploying a single "best" model for every query can be expensive. To reduce inference cost while preserving the quality of the responses, a large language model (LLM) router selects the most appropriate model from a pool of candidates for each query. A central challenge to training a high-quality router is the scarcity of reliable supervision. Gold-standard data (e.g., expert-verified labels or rubric-based scores) provide accurate quality evaluations of LLM responses but are costly and difficult to scale. In contrast, preference-based data, collected via crowdsourcing or LLM-as-a-judge systems, are cheaper and more scalable, yet often biased in reflecting the true quality of responses. We cast the problem of LLM router training with combined gold-standard and preference-based data into a causal inference framework by viewing the response evaluation mechanism as the treatment assignment. This perspective further reveals that the bias in preference-based data corresponds to the well-known causal estimand: the conditional average treatment effect. Based on this new perspective, we develop an integrative causal router training framework that corrects preference-data bias, address imbalances between two data sources, and improve routing robustness and efficiency. Numerical experiments demonstrate that our approach delivers more accurate routing and improves the trade-off between cost and quality.

Rainfall nowcasting, which aims to predict precipitation within the next 0 to 3 hours, is critical for disaster mitigation and real-time response planning. However, most time series forecasting benchmarks in meteorology are evaluated on variables with strong periodicity, such as temperature and humidity, which fail to reflect model capabilities in more complex and practically meteorology scenarios like rainfall nowcasting. To address this gap, we propose RainfallBench, a benchmark designed for rainfall nowcasting, a highly challenging and practically relevant task characterized by zero inflation, temporal decay, and non-stationarity, focused on predicting precipitation within the next 0 to 3 hours. The dataset is derived from five years of meteorological observations, recorded at 15-minute intervals across six essential variables, and collected from more than 12,000 GNSS stations globally. In particular, it incorporates precipitable water vapor (PWV), a crucial indicator of rainfall that is absent in other datasets. We further design specialized evaluation strategies to assess model performance on key meteorological challenges, such as multi-scale prediction and extreme rainfall events, and evaluate over 20 state-of-the-art models across six major architectures on RainfallBench. Additionally, to address the zero-inflation and temporal decay issues overlooked by existing models, we introduce Bi-Focus Precipitation Forecaster (BFPF), a plug-and-play module that incorporates domain-specific priors to enhance rainfall time series forecasting. Statistical analysis and ablation studies validate the comprehensiveness of our dataset as well as the superiority of our methodology. Code and datasets are available at https://anonymous.4open.science/r/RainfallBench-A710.

Classical learning theory describes a well-characterised U-shaped relationship between model complexity and prediction error, reflecting a transition from underfitting in underparameterised regimes to overfitting as complexity grows. Recent work, however, has introduced the notion of a second descent in test error beyond the interpolation threshold-giving rise to the so-called double descent phenomenon. While double descent has been studied extensively in the context of deep learning, it has also been reported in simpler models, including decision trees and gradient boosting. In this work, we revisit these claims through the lens of classical machine learning applied to a biological classification task: predicting isoniazid resistance in Mycobacterium tuberculosis using whole-genome sequencing data. We systematically vary model complexity along two orthogonal axes-learner capacity (e.g., Pleaf, Pboost) and ensemble size (i.e., Pens)-and show that double descent consistently emerges only when complexity is scaled jointly across these axes. When either axis is held fixed, generalisation behaviour reverts to classical U- or L-shaped patterns. These results are replicated on a synthetic benchmark and support the unfolding hypothesis, which attributes double descent to the projection of distinct generalisation regimes onto a single complexity axis. Our findings underscore the importance of treating model complexity as a multidimensional construct when analysing generalisation behaviour. All code and reproducibility materials are available at: https://github.com/guillermocomesanacimadevila/Demystifying-Double-Descent-in-ML.