The paper investigates how large language models (LLMs) can be distributionally aligned with human output distributions, particularly for applications like survey simulation. It examines both the alignment process and the models’ ability to generalize to unseen questions within the same domain and to related but different topics. The study finds that while LLMs can achieve alignment and generalize within a domain, cross‑topic generalization is variable and aligned models may unexpectedly lose creative capabilities, indicating unpredictable and unstable generalization properties.
arXiv:2609.00565v1 Announce Type: cross Abstract: Cultural fine-tuning has become the de facto paradigm for building culture-aware large language models (LLMs), yet existing optimization exclusively...
arXiv:2608. 14629v1 Announce Type: cross Abstract: As Large Language Models (LLMs) become the mainstay for information retrieval and summarization tasks, ensuring that they are always non-partisan and invulnerable to political bias is a critical step towards safer and more trustworthy Artificial Intelligence (AI).
arXiv:2609.32717v2 Announce Type: replace-cross Abstract: Large Language Model (LLM) alignment is intended to ensure that models remain helpful and safe, but its stability under input distributional...
The paper proposes a method for culturally aligning large language models (LLMs) using soft prompt tuning optimized via Differential Evolution (DE). Unlike traditional fine‑tuning or reinforcement learning, this approach keeps model weights frozen and requires no preference data, instead leveraging aggregated survey scores from Hofstede's Value Survey Module (VSM13). Experiments on four countries and four instruction‑tuned models show that DE‑optimized prompts reduce cultural discrepancy, improve agreement with the World Values Survey, and are preferred in blinded pairwise evaluations by LLM judges.
The paper introduces Population Fidelity, an evaluation framework for assessing how well large language models (LLMs) represent human population attitudes. It focuses on three dimensions: group-level accuracy, between-group variation, and the structure of that variation. Using the framework, the authors replicate a prior study on machine bias and test cultural fine-tuning, finding that while fine-tuning improves overall alignment, it does not enhance representation of within-population differences.
arXiv:2606. 12754v1 Announce Type: cross Abstract: Are large language models (LLMs) bad at capturing human judgment?
The paper introduces the task of predicting alignment generalization, which measures how fine‑tuning a language model to follow a specific value affects its behavior across a broad set of other values. Through a large‑scale study of 66 values, the authors show that representations derived from model activations when applying values in context outperform textual description‑based methods, achieving a correlation of 0.45 versus 0.05. They further demonstrate that these representations can predict downstream robustness and help construct a taxonomy of LLM values based on empirical generalization dynamics.
The paper investigates the mismatch between large language models’ verbalized (knowledge) distributions and their actual output (behavior) distributions when predicting group opinions. It shows that for 37% of World Values Survey questions, the most likely answer under the verbalized distribution differs from that under the log‑probability distribution, indicating a disconnect between what the model knows and what it outputs. To reconcile this, the authors propose a learnable combination of the two distributions that preserves the knowledge shape while aligning with the behavior’s most likely answer, achieving higher behavioral consistency with only a modest increase in distance to human opinion distributions.
arXiv:2610.27611v1 Announce Type: cross Abstract: Ensuring the alignment of Large Language Models (LLMs) with diverse cultural values is critical for their safe global deployment. However, existing a...
arXiv:2607. 22699v1 Announce Type: new Abstract: As Large Language Models (LLMs) grow more capable across diverse tasks, their (in)ability to generalize remains difficult to quantify and poorly understood beyond limited domains.
The paper investigates how alignment training in large language models (LLMs) can suppress cultural knowledge rather than erase it. By surveying 1.53 million humans across 16 MENA countries and testing 26 models, the authors find that when models refuse to answer, their internal logit distributions still align closely with human responses, a phenomenon they term the "alignment veto." They distinguish between suppression failures—where accurate internal representations are blocked at output—and representational bias failures—where the encoding itself diverges from human values, and show that these require different interventions. The study also reveals significant inequities in the safety tax, with a 37.6 % cost and a 19.8 % alignment-quality gap between best- and worst-served nations, and notes that native-language prompting actually widens this gap. Sparse autoencoder analysis identifies a candidate feature in the DPO stage that mediates suppression in the Tulu-3-8B model.
The paper investigates how alignment training, specifically reinforcement learning from human feedback (RLHF), affects the internal partisan structure of a large language model. Using a mechanistic case study on Llama 3.1 8B, the authors find that alignment training does not erase the model’s partisan geometry but compresses its variance, producing consistently balanced, non‑partisan outputs. Sparse autoencoder analysis and feature‑level steering experiments reveal that policy‑encoding features become inactive in the aligned model, indicating a causal disconnect rather than structural removal of partisan knowledge.