Watermarking techniques for large language models (LLMs) can significantly impact output quality, yet their effects on truthfulness, safety, and helpfulness remain critically underexamined. This paper presents a systematic analysis of how two popular watermarking approaches-Gumbel and KGW-affect these core alignment properties across four aligned LLMs. Our experiments reveal two distinct degradation patterns: guard attenuation, where enhanced helpfulness undermines model safety, and guard amplification, where excessive caution reduces model helpfulness. These patterns emerge from watermark-induced shifts in token distribution, surfacing the fundamental tension that exists between alignment objectives. To mitigate these degradations, we propose Alignment Resampling (AR), an inference-time sampling method that uses an external reward model to restore alignment. We establish a theoretical lower bound on the improvement in expected reward score as the sample size is increased and empirically demonstrate that sampling just 2-4 watermarked generations effectively recovers or surpasses baseline (unwatermarked) alignment scores. To overcome the limited response diversity of standard Gumbel watermarking, our modified implementation sacrifices strict distortion-freeness while maintaining robust detectability, ensuring compatibility with AR. Experimental results confirm that AR successfully recovers baseline alignment in both watermarking approaches, while maintaining strong watermark detectability. This work reveals the critical balance between watermark strength and model alignment, providing a simple inference-time solution to responsibly deploy watermarked LLMs in practice.

Radiology reports are critical for clinical decision-making but often lack a standardized format, limiting both human interpretability and machine learning (ML) applications. While large language models (LLMs) have shown strong capabilities in reformatting clinical text, their high computational requirements, lack of transparency, and data privacy concerns hinder practical deployment. To address these challenges, we explore lightweight encoder-decoder models (<300M parameters)-specifically T5 and BERT2BERT-for structuring radiology reports from the MIMIC-CXR and CheXpert Plus datasets. We benchmark these models against eight open-source LLMs (1B-70B), adapted using prefix prompting, in-context learning (ICL), and low-rank adaptation (LoRA) finetuning. Our best-performing lightweight model outperforms all LLMs adapted using prompt-based techniques on a human-annotated test set. While some LoRA-finetuned LLMs achieve modest gains over the lightweight model on the Findings section (BLEU 6.4%, ROUGE-L 4.8%, BERTScore 3.6%, F1-RadGraph 1.1%, GREEN 3.6%, and F1-SRR-BERT 4.3%), these improvements come at the cost of substantially greater computational resources. For example, LLaMA-3-70B incurred more than 400 times the inference time, cost, and carbon emissions compared to the lightweight model. These results underscore the potential of lightweight, task-specific models as sustainable and privacy-preserving solutions for structuring clinical text in resource-constrained healthcare settings.

Steering vectors are a lightweight method for controlling text properties by adding a learned bias to language model activations at inference time. So far, steering vectors have predominantly been evaluated in multiple-choice settings, while their effectiveness in free-form generation tasks remains understudied. Moving "Beyond Multiple Choice," we thoroughly evaluate the effectiveness of steering vectors in adaptively controlling topical focus, sentiment, toxicity, and readability in abstractive summaries of the NEWTS dataset. We find that steering effectively controls the targeted summary properties, but high steering strengths consistently degrade both intrinsic and extrinsic text quality. Compared to steering, prompting offers weaker control, while preserving text quality. Combining steering and prompting yields the strongest control over text properties and offers the most favorable efficacy-quality trade-off at moderate steering strengths. Our results underscore the practical trade-off between control strength and text quality preservation when applying steering vectors to free-form generation tasks.

Federated LoRA has emerged as a promising technique for efficiently fine-tuning large language models (LLMs) on distributed devices by reducing the number of trainable parameters. However, existing approaches often inadequately overlook the theoretical and practical implications of system and data heterogeneity, thereby failing to optimize the overall training efficiency, particularly in terms of wall-clock time. In this paper, we propose an adaptive federated LoRA strategy with independent client sampling to minimize the convergence wall-clock time of federated fine-tuning under both computation and communication heterogeneity. We first derive a new convergence bound for federated LoRA with arbitrary and independent client sampling, notably without requiring the stringent bounded gradient assumption. Then, we introduce an adaptive bandwidth allocation scheme that accounts for heterogeneous client resources and system bandwidth constraints. Based on the derived theory, we formulate and solve a non-convex optimization problem to jointly determine the LoRA sketching ratios and sampling probabilities, aiming to minimize wall-clock convergence time. An efficient and low-complexity algorithm is developed to approximate the solution. Finally, extensive experiments demonstrate that our approach significantly reduces wall-clock training time compared to state-of-the-art methods across various models and datasets.

We argue that in-context learning (ICL) predictably arises from standard self-supervised next-token pretraining, rather than being an exotic emergent property. This work establishes the foundational principles of this emergence by focusing on in-distribution ICL, demonstrating how models necessarily adapt to context when trained on token sequences, especially from non-ergodic sources. Our information-theoretic framework precisely predicts these in-distribution ICL dynamics (i.e., context-dependent loss reduction). We verify this with experiments using synthetic datasets of differing types of correlational structure, reproducing characteristic phenomena like phase transitions in training loss for induction head formation and power-law scaling of in-context loss. We further show that a model's in-context performance on any task is mathematically coupled to the ensemble of tasks seen in pretraining, offering a fundamental explanation, grounded in architecture- and modality-independent principles, for such inference-time learning.

The advancement of large language models (LLMs) has outpaced traditional evaluation methodologies. This progress presents novel challenges, such as measuring human-like psychological constructs, moving beyond static and task-specific benchmarks, and establishing human-centered evaluation. These challenges intersect with psychometrics, the science of quantifying the intangible aspects of human psychology, such as personality, values, and intelligence. This review paper introduces and synthesizes the emerging interdisciplinary field of LLM Psychometrics, which leverages psychometric instruments, theories, and principles to evaluate, understand, and enhance LLMs. The reviewed literature systematically shapes benchmarking principles, broadens evaluation scopes, refines methodologies, validates results, and advances LLM capabilities. Diverse perspectives are integrated to provide a structured framework for researchers across disciplines, enabling a more comprehensive understanding of this nascent field. Ultimately, the review provides actionable insights for developing future evaluation paradigms that align with human-level AI and promote the advancement of human-centered AI systems for societal benefit. A curated repository of LLM psychometric resources is available at https://github.com/valuebyte-ai/Awesome-LLM-Psychometrics.

As Large Language Models (LLMs) and other AI systems evolve, robustly estimating their capabilities from inherently stochastic outputs while systematically quantifying uncertainty in these estimates becomes increasingly important. Further, advanced AI evaluations often have a nested hierarchical structure, exhibit high levels of complexity, and come with high costs in testing the most advanced AI systems. To address these challenges, we introduce HiBayES, a generalizable Hierarchical Bayesian modeling framework for AI Evaluation Statistics. HiBayES supports robust inferences in classical question-answer benchmarks and advanced agentic evaluations, particularly in low-data scenarios (e.g., < 20 data points per evaluation). Built on Generalized Linear Models (GLMs), Bayesian data analysis, and formal model comparison, HiBayES provides principled uncertainty quantification and robust parameter estimation. This paper offers a comprehensive introduction to HiBayES, including illustrative examples, comparisons to conventional statistical methods, and practical guidance for implementing multilevel Bayesian GLMs. Additionally, we provide a HiBayES software package [4] (Beta version) for out-of-the-box implementation.

The hallmark of effective language use lies in consistency: expressing similar meanings in similar contexts and avoiding contradictions. While human communication naturally demonstrates this principle, state-of-the-art language models (LMs) struggle to maintain reliable consistency across task- and domain-specific applications. Here we examine the landscape of consistency research in LMs, analyze current approaches to measure aspects of consistency, and identify critical research gaps. Our findings point to an urgent need for quality benchmarks to measure and interdisciplinary approaches to ensure consistency while preserving utility.

Because large language models are expensive to pretrain on different datasets, using smaller-scale experiments to decide on data is crucial for reducing costs. Which benchmarks and methods of making decisions from observed performance at small scale most accurately predict the datasets that yield the best large models? To empower open exploration of this question, we release models, data, and evaluations in DataDecide -- the most extensive open suite of models over differences in data and scale. We conduct controlled pretraining experiments across 25 corpora with differing sources, deduplication, and filtering up to 100B tokens, model sizes up to 1B parameters, and 3 random seeds. We find that the ranking of models at a single, small size (e.g., 150M parameters) is a strong baseline for predicting best models at our larger target scale (1B) (~80% of com parisons correct). No scaling law methods among 8 baselines exceed the compute-decision frontier of single-scale predictions, but DataDecide can measure improvement in future scaling laws. We also identify that using continuous likelihood metrics as proxies in small experiments makes benchmarks including MMLU, ARC, HellaSwag, MBPP, and HumanEval >80% predictable at the target 1B scale with just 0.01% of the compute.

Recent advances in automated theorem proving (ATP) through LLMs have highlighted the potential of formal reasoning with Lean 4 codes. However, ATP has not yet be revolutionized by the recent posttraining scaling as demonstrated by Open AI O1/O3 and Deepseek R1. In this work, we investigate the entire posttraining of ATP, aiming to align it with breakthroughs in reasoning models in natural languages. To begin, we continual train current ATP models with a hybrid dataset, which consists of numerous statement-proof pairs, and additional data aimed at incorporating cognitive behaviors that emulate human reasoning and hypothesis refinement. Next, we explore reinforcement learning with the use of outcome reward returned by Lean 4 compiler. Through our designed continual training and reinforcement learning processes, we have successfully improved existing formal provers, including both DeepSeek-Prover-v1.5 and Goedel-Prover, achieving state-of-the-art performance in the field of whole-proof generation. For example, we achieve a 59.8% pass rate (pass@32) on MiniF2F. This is an on-going project and we will progressively update our findings, release our data and training details.