Synthetic personas support LLM personalization, social simulation, and alignment evaluation, yet existing resources typically represent each individual with only a small number of structured attributes. We introduce DeepPersona, a generative engine that scales persona depth to roughly 200 instantiated taxonomy nodes while preserving narrative coherence. DeepPersona combines (i) a hierarchical taxonomy of 8,496 human-descriptive attributes mined from 62k personalizable QA pairs in real user-chatbot dialogues and (ii) a progressive sampler that conditions each new attribute on the evolving persona and uses a stratified similarity prior to balance coherence and diversity. The same engine can expand shallow inputs, such as PersonaHub entries, into deep profiles. Across three complementary evaluations, DeepPersona improves judge-extracted attribute count by 32% and judge-rated uniqueness by 44% over the strongest prior persona resource; improves LLM personalization quality by up to 11.6% across ten dimensions, with human evaluators preferring DeepPersona responses in 81.2% to 87.0% of pairwise comparisons; and reduces the distributional gap to real World Values Survey responses by 31.7%, while recovering Big Five personality distributions 17% closer to ground truth than the strongest LLM baseline.

Knowledge-based visual question answering (KB-VQA) lets vision-language systems answer questions that exceed their parametric knowledge by conditioning a reader on passages retrieved from a Wikipedia-derived knowledge base. In pure-text long-context LLMs, retrieved-context use follows the U-shaped lost-in-the-middle effect of Liu et al. (2024): information at the start and end of context is used, the middle is lost. Whether this transfers to deployed multimodal KB-VQA is open. To close this gap, we design the first controlled probe of reader-side position dependence in multimodal KB-VQA: a gold-position protocol in which only the gold passage’s prompt slot varies within question. We run it on three open-source 7B/8B VLM readers and two KB-VQA benchmarks with up to 20 retrieved passages. The shape flips from U to primacy: gold-at-first beats gold-at-last by 16 to 26 points on all six combinations of reader and benchmark, an effect we call Lost at the End; the gap holds at every scale we test, 3B to 32B, attenuating at 32B. Three targeted ablations narrow the cause. A text-only control that removes the image and changes nothing else shows the primacy is already present in text mode and does not depend on the image. Image-position and distractor-shuffle ablations trace the effect to prompt slot 0 of the instruction-tuned reader, where a second answer-bearing passage placed later is largely wasted. On a frozen reader, three retrieval-side fixes (MMR, oracle reranking, rank-based reordering) all fail to improve on the deployment default. Our findings indicate that recall@k is the wrong metric for deployed KB-VQA and that the remaining headroom sits on the reader side; we release our protocol as a controlled instrument for evaluating reader-side interventions.

The recent explosion of large language models (LLMs), each with its own general or specialized strengths, makes scalable, reliable benchmarking more urgent than ever. Standard practices nowadays face fundamental trade-offs: closed-ended question-based benchmarks (MMLU) struggle with saturation as newer models emerge, while crowd-sourced leaderboards (Chatbot Arena) rely on costly and slow human judges. Recently, automated methods (LLM-as-a-judge) shed light on the scalability, but risk bias by relying on one or a few “authority” models. To tackle these issues, we propose Decentralized Arena (), a fully automated framework leveraging collective intelligence from all LLMs to evaluate each other. It mitigates single-model judge bias by democratic, pairwise evaluation, and remains efficient at scale through two key components: (1) a coarse-to-fine ranking algorithm for fast incremental insertion of new models with sub-quadratic complexity, and (2) an automatic question selection strategy for the construction of new evaluation dimensions. Across extensive experiments across 66 LLMs, attains up to 97% correlation with human judgements, while significantly reducing the cost.

Aligning Large Language Models (LLMs) traditionally relies on complex and costly training processes like supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). To address the challenge of achieving alignment without these extensive tuning costs and expensive annotations, we present a novel, tuning-free approach for self-alignment called Dynamic Rewarding with Prompt Optimization (DRPO). Our approach enables self-alignment through a search-based prompt optimization framework, allowing the model to self-improve and generate optimized prompts without additional training or human supervision. The core of DRPO leverages a dynamic rewarding mechanism to identify and rectify model-specific alignment weaknesses, enabling LLMs to adapt quickly to various alignment challenges. Empirical evaluations on eight recent LLMs, including both open- and closed-source, reveal that DRPO significantly enhances alignment performance, enabling base models to outperform their SFT/RLHF-tuned counterparts. Moreover, DRPO’s automatically optimized prompts surpass those curated by human experts, demonstrating its superior alignment capabilities. Our findings envision a highly cost-effective and adaptable solution for future alignment research to be further explored.

Large language models (LLMs) have a substantial capacity for high-level analogical reasoning: reproducing patterns in linear text that occur in their training data (zero-shot evaluation) or in the provided context (few-shot in-context learning). However, recent studies show that even the more advanced LLMs fail in scenarios that require reasoning over multiple objects or facts and making sequences of logical deductions. We propose a two-stage probabilistic inference paradigm, ThinkSum, which reasons over sets of objects or facts in a structured manner. In the first stage (Think – retrieval of associations), a LLM is queried in parallel over a set of phrases extracted from the prompt or an auxiliary model call. In the second stage (Sum – probabilistic inference or reasoning), the results of these queries are aggregated to make the final prediction. We demonstrate the possibilities and advantages of ThinkSum on the BIG-bench suite of LLM evaluation tasks, achieving improvements over the state of the art using GPT-family models on thirteen difficult tasks, often with far smaller model variants. We also compare and contrast ThinkSum with other proposed modifications to direct prompting of LLMs, such as variants of chain-of-thought prompting. Our results suggest that because the probabilistic inference in ThinkSum is performed outside of calls to the LLM, ThinkSum is less sensitive to prompt design, yields more interpretable predictions, and can be flexibly combined with latent variable models to extract structured knowledge from LLMs. Overall, our proposed paradigm represents a promising approach for enhancing the reasoning capabilities of LLMs.

Multi-hop question answering (QA) combines multiple pieces of evidence to search for the correct answer. Reasoning over a text corpus (TextQA) and/or a knowledge base (KBQA) has been extensively studied and led to distinct system architectures. However, knowledge transfer between such two QA systems has been under-explored. Research questions like what knowledge is transferred or whether the transferred knowledge can help answer over one source using another one, are yet to be answered. In this paper, therefore, we study the knowledge transfer of multi-hop reasoning between structured and unstructured sources. We first propose a unified QA framework named SimultQA to enable knowledge transfer and bridge the distinct supervisions from KB and text sources. Then, we conduct extensive analyses to explore how knowledge is transferred by leveraging the pre-training and fine-tuning paradigm. We focus on the low-resource fine-tuning to show that pre-training SimultQA on one source can substantially improve its performance on the other source. More fine-grained analyses on transfer behaviors reveal the types of transferred knowledge and transfer patterns. We conclude with insights into how to construct better QA datasets and systems to exploit knowledge transfer for future work.

Long-range semantic coherence remains a challenge in automatic language generation and understanding. We demonstrate that large language models have insufficiently learned the effect of distant words on next-token prediction. We present coherence boosting, an inference procedure that increases a LM’s focus on a long context. We show the benefits of coherence boosting with pretrained models by distributional analyses of generated ordinary text and dialog responses. It is also found that coherence boosting with state-of-the-art models for various zero-shot NLP tasks yields performance gains with no additional training.