Robust supervised fine-tuned small Language Models (sLMs) often show high reliability but tend to undercorrect. They achieve high precision at the cost of low recall. Conversely, Large Language Models (LLMs) often show the opposite tendency, making excessive overcorrection, leading to low precision. To effectively harness the strengths of LLMs to address the recall challenges in sLMs, we propose Post-Correction via Overcorrection (PoCO), a novel approach that strategically balances recall and precision. PoCO first intentionally triggers overcorrection via LLM to maximize recall by allowing comprehensive revisions, then applies a targeted post-correction step via fine-tuning smaller models to identify and refine erroneous outputs. We aim to harmonize both aspects by leveraging the generative power of LLMs while preserving the reliability of smaller supervised models. Our extensive experiments demonstrate that PoCO effectively balances GEC performance by increasing recall with competitive precision, ultimately improving the overall quality of grammatical error correction.

Personalized review generation helps businesses understand user preferences, yet most existing approaches assume extensive review histories of the target user or require additional model training. Real-world applications often face few-shot and training-free situations, where only a few user reviews are available and fine-tuning is infeasible. It is well known that large language models (LLMs) can address such low-resource settings, but their effectiveness depends on prompt engineering. In this paper, we propose Conversational Prompting, a lightweight method that reformulates user reviews as multi-turn conversations. Its simple variant, Simple Conversational Prompting (SCP), relies solely on the user's own reviews, while the contrastive variant, Contrastive Conversational Prompting (CCP), inserts reviews from other users or LLMs as incorrect replies and then asks the model to correct them, encouraging the model to produce text in the user's style. Experiments on eight product domains and five LLMs showed that the conventional non-conversational prompt often produced reviews similar to those written by random users, based on text-based metrics such as ROUGE-L and BERTScore, and application-oriented tasks like user identity matching and sentiment analysis. In contrast, both SCP and CCP produced reviews much closer to those of the target user, even when each user had only two reviews. CCP brings further improvements when high-quality negative examples are available, whereas SCP remains competitive when such data cannot be collected. These results suggest that conversational prompting offers a practical solution for review generation under few-shot and training-free constraints.

The fundamental units of internal representations in large language models (LLMs) remain undefined, limiting further understanding of their mechanisms. Neurons or features are often regarded as such units, yet neurons suffer from polysemy, while features face concerns of unreliable reconstruction and instability. To address this issue, we propose the Atoms Theory, which defines such units as atoms. We introduce the atomic inner product (AIP) to correct representation shifting, formally define atoms, and prove the conditions that atoms satisfy the Restricted Isometry Property (RIP), ensuring stable sparse representations over atom set and linking to compressed sensing. Under stronger conditions, we further establish the uniqueness and exact $\ell_1$ recoverability of the sparse representations, and provide guarantees that single-layer sparse autoencoders (SAEs) with threshold activations can reliably identify the atoms. To validate the Atoms Theory, we train threshold-activated SAEs on Gemma2-2B, Gemma2-9B, and Llama3.1-8B, achieving 99.9% sparse reconstruction across layers on average, and more than 99.8% of atoms satisfy the uniqueness condition, compared to 0.5% for neurons and 68.2% for features, showing that atoms more faithfully capture intrinsic representations of LLMs. Scaling experiments further reveal the link between SAEs size and recovery capacity. Overall, this work systematically introduces and validates Atoms Theory of LLMs, providing a theoretical framework for understanding internal representations and a foundation for mechanistic interpretability. Code available at https://github.com/ChenhuiHu/towards_atoms.

In this work, we present a retrieval-augmented generation (RAG)-based system for provenance analysis of archaeological artifacts, designed to support expert reasoning by integrating multimodal retrieval and large vision-language models (VLMs). The system constructs a dual-modal knowledge base from reference texts and images, enabling raw visual, edge-enhanced, and semantic retrieval to identify stylistically similar objects. Retrieved candidates are synthesized by the VLM to generate structured inferences, including chronological, geographical, and cultural attributions, alongside interpretive justifications. W e evaluate the system on a set of Eastern Eurasian Bronze Age artifacts from the British Museum. Expert evaluation demonstrates that the system produces meaningful and interpretable outputs, offering scholars concrete starting points for analysis and significantly alleviating the cognitive burden of navigating vast comparative corpora.

Synthetic tabular data is used for privacy-preserving data sharing and data-driven model development. Its effectiveness, however, depends heavily on the used Tabular Data Synthesis (TDS) tool. Recent studies have shown that Transformer-based models outperform other state-of-the-art models such as Generative Adversarial Networks (GANs) and Diffusion models in terms of data quality. However, Transformer-based models also come with high computational costs, making them sometimes unfeasible for end users with prosumer hardware. This study presents a sensitivity assessment on how the choice of hyperparameters, such as number of layers or hidden dimension affects the quality of the resultant synthetic data and the computational performance. It is performed across two tools, GReaT and REaLTabFormer, evaluating 10 model setups that vary in architecture type and depth. We assess the sensitivity on three dimensions: runtime, machine learning (ML) utility, and similarity to real data distributions. Experiments were conducted on four real-world datasets. Our findings reveal that runtime is proportional to the number of hyperparameters, with shallower configurations completing faster. GReaT consistently achieves lower runtimes than REaLTabFormer, and only on the largest dataset they have comparable runtime. For small datasets, both tools achieve synthetic data with high utility and optimal similarity, but on larger datasets only REaLTabFormer sustains strong utility and similarity. As a result, REaLTabFormer with lightweight LLMs provides the best balance, since it preserves data quality while reducing computational requirements. Nonetheless, its runtime remains higher than that of GReaT and other TDS tools, suggesting that efficiency gains are possible but only up to a certain level.

Navigating complex environments requires robots to effectively store observations as memories and leverage them to answer human queries about spatial locations, which is a critical yet underexplored research challenge. While prior work has made progress in constructing robotic memory, few have addressed the principled mechanisms needed for efficient memory retrieval and integration. To bridge this gap, we propose Meta-Memory, a large language model (LLM)-driven agent that constructs a high-density memory representation of the environment. The key innovation of Meta-Memory lies in its capacity to retrieve and integrate relevant memories through joint reasoning over semantic and spatial modalities in response to natural language location queries, thereby empowering robots with robust and accurate spatial reasoning capabilities. To evaluate its performance, we introduce SpaceLocQA, a large-scale dataset encompassing diverse real-world spatial question-answering scenarios. Experimental results show that Meta-Memory significantly outperforms state-of-the-art methods on both the SpaceLocQA and the public NaVQA benchmarks. Furthermore, we successfully deployed Meta-Memory on real-world robotic platforms, demonstrating its practical utility in complex environments. Project page: https://itsbaymax.github.io/meta-memory.github.io/ .

Recent studies show that deep vision-only and language-only models--trained on disjoint modalities--nonetheless project their inputs into a partially aligned representational space. Yet we still lack a clear picture of where in each network this convergence emerges, what visual or linguistic cues support it, whether it captures human preferences in many-to-many image-text scenarios, and how aggregating exemplars of the same concept affects alignment. Here, we systematically investigate these questions. We find that alignment peaks in mid-to-late layers of both model types, reflecting a shift from modality-specific to conceptually shared representations. This alignment is robust to appearance-only changes but collapses when semantics are altered (e.g., object removal or word-order scrambling), highlighting that the shared code is truly semantic. Moving beyond the one-to-one image-caption paradigm, a forced-choice "Pick-a-Pic" task shows that human preferences for image-caption matches are mirrored in the embedding spaces across all vision-language model pairs. This pattern holds bidirectionally when multiple captions correspond to a single image, demonstrating that models capture fine-grained semantic distinctions akin to human judgments. Surprisingly, averaging embeddings across exemplars amplifies alignment rather than blurring detail. Together, our results demonstrate that unimodal networks converge on a shared semantic code that aligns with human judgments and strengthens with exemplar aggregation.

Integrating natural language (NL) prompts into robotic mission planning has attracted significant interest in recent years. In the construction domain, Building Information Models (BIM) encapsulate rich NL descriptions of the environment. We present a novel framework that fuses NL directives with BIM-derived semantic maps via a Beta-Bernoulli Bayesian fusion by interpreting the LLM as a sensor: each obstacle's design-time repulsive coefficient is treated as a Beta(alpha, beta) random variable and LLM-returned danger scores are incorporated as pseudo-counts to update alpha and beta. The resulting posterior mean yields a continuous, context-aware repulsive gain that augments a Euclidean-distance-based potential field for cost heuristics. By adjusting gains based on sentiment and context inferred from user prompts, our method guides robots along safer, more context-aware paths. This provides a numerically stable method that can chain multiple natural commands and prompts from construction workers and foreman to enable planning while giving flexibility to be integrated in any learned or classical AI framework. Simulation results demonstrate that this Beta-Bernoulli fusion yields both qualitative and quantitative improvements in path robustness and validity.

Adversarial text attack research plays a crucial role in evaluating the robustness of NLP models. However, the increasing complexity of transformer-based architectures has dramatically raised the computational cost of attack testing, especially for researchers with limited resources (e.g., GPUs). Existing popular black-box attack methods often require a large number of queries, which can make them inefficient and impractical for researchers. To address these challenges, we propose two new attack selection strategies called Hybrid and Dynamic Select, which better combine the strengths of previous selection algorithms. Hybrid Select merges generalized BinarySelect techniques with GreedySelect by introducing a size threshold to decide which selection algorithm to use. Dynamic Select provides an alternative approach of combining the generalized Binary and GreedySelect by learning which lengths of texts each selection method should be applied to. This greatly reduces the number of queries needed while maintaining attack effectiveness (a limitation of BinarySelect). Across 4 datasets and 6 target models, our best method(sentence-level Hybrid Select) is able to reduce the number of required queries per attack up 25.82\% on average against both encoder models and LLMs, without losing the effectiveness of the attack.

ABSTRACT Audio Large Language Models (Audio LLMs) enable human-like conversation about music, yet it is unclear if they are truly listening to the audio or just using textual reasoning, as recent benchmarks suggest. This paper investigates this issue by quantifying the contribution of each modality to a model's output. We adapt the MM-SHAP framework, a performance-agnostic score based on Shapley values that quantifies the relative contribution of each modality to a model's prediction. We evaluate two models on the MuChoMusic benchmark and find that the model with higher accuracy relies more on text to answer questions, but further inspection shows that even if the overall audio contribution is low, models can successfully localize key sound events, suggesting that audio is not entirely ignored. Our study is the first application of MM-SHAP to Audio LLMs and we hope it will serve as a foundational step for future research in explainable AI and audio.