While large language models (LLMs) are equipped with longer text input capabilities than before, they are struggling to seek correct information in long contexts. The “lost in the middle” problem challenges most LLMs, referring to the dramatic decline in accuracy when correct information is located in the middle. To overcome this crucial issue, this paper proposes to enhance the information searching and reflection ability of LLMs in long contexts via specially designed tasks called Position-Agnostic Multi-step QA (PAM QA). Trained in this task, our model excels in focusing more precisely on the desired information. Experimental results show substantial improvement in Multi-doc QA and other benchmarks, superior to state-of-the-art models by 13.7% absolute gain in shuffled settings, by 21.5% in passage retrieval task. We release our model and code to promote related research in the community.

We evaluate the ability of Large Language Models (LLMs) to discern and express their internal knowledge state, a key factor in countering factual hallucination and ensuring reliable application of LLMs. We observe a robust self-awareness of internal knowledge state in LLMs, evidenced by over 85% accuracy in knowledge state probing. However, LLMs often fail to faithfully express their internal knowledge during generation, leading to factual hallucinations. We develop an automated hallucination annotation tool, DreamCatcher, which merges knowledge probing and consistency checking methods to rank factual preference data. Using knowledge preference as reward, We propose a Reinforcement Learning from Knowledge Feedback (RLKF) training framework, leveraging reinforcement learning to enhance the factuality and honesty of LLMs. Our experiments across multiple models show that RLKF training effectively enhances the ability of models to utilize their internal knowledge state, boosting performance in a variety of knowledge-based and honesty-related tasks.

Fine-tuning is a widely used technique for leveraging pre-trained language models (PLMs) in downstream tasks, but it can be computationally expensive and storage-intensive. To address this challenge, researchers have developed parameter-efficient methods that balance performance and resource cost. However, these methods often come with trade-offs like increased inference latency, token length usage, or limited adaptability for multitasking scenarios. This paper introduces a novel parameter-efficient method called DimA(Dimensionality Augmentation), which enhances the Transformer architecture by increasing the dimensionality. DimA achieves state-of-the-art results in GLUE and XSUM tasks while utilizing less than 1% of the original model’s parameters. Moreover, DimA introduces a novel approach to knowledge transfer that enables the simultaneous utilization of knowledge learned from multiple tasks to handle new tasks. This method significantly enhances the performance of the model on new tasks. Its versatility in model structure also enables its application to various Transformer-based models.