Although large language models (LLMs) trained on extensive multilingual corpora exhibit impressive language transfer, they often fail to respond in the user’s desired language due to corpus imbalances, an embarrassingly simple problem known as the language confusion. However, existing solutions like in-context learning and supervised fine-tuning (SFT) have drawbacks: in-context learning consumes context window space, diminishing attention as text lengthens, while SFT requires extensive, labor-intensive data collection. To overcome these limitations, we propose the language-sensitive intervention (LSI), a novel, lightweight, and label-free approach. Specifically, we analyze language confusion from a causal perspective, revealing that the training corpus’s language distribution acts as a confounder, disadvantaging languages that are underrepresented in the dataset. Then, we identify a language-sensitive dimension in the LLM’s residual stream, i.e., the language vector, which allows us to estimate the average causal effect of prompts on this dimension. During inference, we directly intervene on the language vector to generate responses in the desired language.To further advance research on this issue, we introduce a new benchmark that detects language confusion and assesses content quality. Experimental results demonstrate that our method effectively mitigates language confusion without additional complex mechanisms. Our code is available at https://github.com/SoseloX/LSI.

Recent advancements in Large Language Models and Retrieval-Augmented Generation have boosted interest in domain-specific question-answering for enterprise products. However, AI Assistants often face challenges in multi-product QA settings, requiring accurate responses across diverse domains. Existing multi-domain RAG-QA approaches either query all domains indiscriminately, increasing computational costs and LLM hallucinations, or rely on rigid resource selection, which can limit search results. We introduce MKP-QA, a novel multi-product knowledge-augmented QA framework with probabilistic federated search across domains and relevant knowledge. This method enhances multi-domain search quality by aggregating query-domain and query-passage probabilistic relevance. To address the lack of suitable benchmarks for multi-product QAs, we also present new datasets focused on three Adobe products: Adobe Experience Platform, Target, and Customer Journey Analytics. Our experiments show that MKP-QA significantly boosts multi-product RAG-QA performance in terms of both retrieval accuracy and response quality.

Honest alignment refers to the ability of a language model to truthfully convey its knowledge limitations by appropriately refusing to answer questions when it lacks sufficient information. Existing solutions, such as prompt engineering and fine-tuning, face limitations: the former provides only marginal improvements, while the latter struggles to enhance honesty when annotated data is scarce.To overcome the above limitations, we propose , a novel framework that enhances honesty through weak-to-strong generalization. Specifically, we train the strong LLMs under weak model supervision to improve their honesty. For the weak model, we employ a learning-to-rank strategy to train a “honest head”, which learns to select the most honest response among model’s outputs generated through beam search. For the strong LLM, we leverage the self-labeled dataset to update its parameters. Our proposal requires only minimal training data to train the weak honest model, yet achieve decent performance for labeling data. In addition, it enables the strong LLMs to have the capabilities to generalize even facing with the flawed label data. Extensive experiments show significantly boosts honest alignment in large models even with limited labeled data. Our code is available at https://github.com/zewanfaan/WHAT_Honesty.