Multi-hop questions still stump large language models (LLMs), which struggle to link information across multiple reasoning steps. We introduce Auto-Patch, a novel method that dynamically patches hidden states during inference to enhance multi-hop reasoning in LLMs. Building on the PatchScopes framework, Auto-Patch selectively modifies internal representations using a learned classifier. Evaluated on the MuSiQue dataset, Auto-Patch improves the solve rate from 18.45\% (baseline) to 23.63~$\pm$~0.7\% (3 runs), narrowing the gap to Chain-of-Thought prompting (27.44\%). Our results highlight the potential of dynamic hidden state interventions for advancing complex reasoning in LLMs.

Current large language models (LLMs) often exhibit imbalances in multilingual capabilities and cultural adaptability, largely due to their English-centric pretraining data. To address this imbalance, we propose a probing method named XTransplant that explores cross-lingual latent interactions via cross-lingual feed-forward transplantation during inference stage, with the hope of enabling the model to leverage the strengths of both English and non-English languages. Through extensive pilot experiments, we empirically prove that both the multilingual capabilities and cultural adaptability of LLMs hold the potential to be significantly improved by XTransplant, respectively from En -> non-En and non-En -> En, highlighting the underutilization of current LLMs' multilingual potential. And the patterns observed in these pilot experiments further motivate an offline scaling inference strategy, which demonstrates consistent performance improvements in multilingual and culture-aware tasks, sometimes even surpassing multilingual supervised fine-tuning. And we do hope our further analysis and discussion could help gain deeper insights into XTransplant mechanism.

Recent language models have shown remarkable performance on natural language understanding (NLU) tasks. However, they are often sub-optimal when faced with ambiguous samples that can be interpreted in multiple ways, over-confidently predicting a single label without consideration for its correctness. To address this issue, we propose a novel self-knowledge distillation method that enables models to learn label distributions more accurately by leveraging knowledge distilled from their lower layers. This approach also includes a learning phase that re-calibrates the unnecessarily strengthened confidence for training samples judged as extremely ambiguous based on the distilled distribution knowledge. We validate our method on diverse NLU benchmark datasets and the experimental results demonstrate its effectiveness in producing better label distributions. Particularly, through the process of re-calibrating the confidence for highly ambiguous samples, the issue of over-confidence when predictions for unseen samples do not match with their ground-truth labels has been significantly alleviated. This has been shown to contribute to generating better distributions than the existing state-of-the-art method. Moreover, our method is more efficient in training the models compared to the existing method, as it does not involve additional training processes to refine label distributions.

Inspecting the information encoded in hidden representations of large language models (LLMs) can explain models' behavior and verify their alignment with human values. Given the capabilities of LLMs in generating human-understandable text, we propose leveraging the model itself to explain its internal representations in natural language. We introduce a framework called Patchscopes and show how it can be used to answer a wide range of questions about an LLM's computation. We show that prior interpretability methods based on projecting representations into the vocabulary space and intervening on the LLM computation can be viewed as instances of this framework. Moreover, several of their shortcomings such as failure in inspecting early layers or lack of expressivity can be mitigated by Patchscopes. Beyond unifying prior inspection techniques, Patchscopes also opens up new possibilities such as using a more capable model to explain the representations of a smaller model, and unlocks new applications such as self-correction in multi-hop reasoning.

Finite mixture models can describe a wide range of statistical phenomena. They have been successfully applied to numerous fields including biology, economics, engineering, and social sciences [15]. The first major use and analysis of mixture models is perhaps due to the mathematician and biostatistician Karl Pearson over 120 years ago who explicitly decomposed a distribution into two normal distributions for characterizing non-normal attributes of forehead to body length ratios in female shore crab populations [16]. The literature on analyzing and applying mixture models is growing due to their simplicity, versatility and flexibility. One of the most commonly used mixture models is the Gaussian mixture model (GMM), which is a weighted sum of Gaussian distributions.