The emergence of large language models (LLMs) has marked a significant breakthrough in natural language processing (NLP), fueling a paradigm shift in information acquisition. Nevertheless, LLMs are prone to hallucination, generating plausible yet nonfactual content. This phenomenon raises significant concerns over the reliability of LLMs in real-world information retrieval (IR) systems and has attracted intensive research to detect and mitigate such hallucinations. Given the open-ended general-purpose attributes inherent to LLMs, LLM hallucinations present distinct challenges that diverge from prior task-specific models. This divergence highlights the urgency for a nuanced understanding and comprehensive overview of recent advances in LLM hallucinations. In this survey, we begin with an innovative taxonomy of hallucination in the era of LLM and then delve into the factors contributing to hallucinations. Subsequently, we present a thorough overview of hallucination detection methods and benchmarks. Our discussion then transfers to representative methodologies for mitigating LLM hallucinations. Additionally, we delve into the current limitations faced by retrieval-augmented LLMs in combating hallucinations, offering insights for developing more robust IR systems. Finally, we highlight the promising research directions on LLM hallucinations, including hallucination in large vision-language models and understanding of knowledge boundaries in LLM hallucinations.
18 May 2011 in Macromolecular Materials and Engineering
Owing to the huge interface region existing in a polymer nanocomposite, the effects of thermal processes on properties of nanocomposites are much more complicated than in a pure polymer. It is therefore important to determine the effects of thermal processes on nanocomposites with different interfacial interactions between the nanofillers and the polymer matrix. It is also important to explore the performance changes for nanocomposites under elevated temperatures over pure polymers. In this investigation, we examined the correlation of thermal treatment with dielectric properties of carbon nanofiber (CNF) reinforced high-density polyethylene nanocomposites. The thermal treatment of specimens was conducted for up to 120 h at 87°C and 127°C. Then, alternating current (AC) conductivity and dielectric properties were tested after definite intervals of time. Their changing rates over treatment time were analyzed. The results revealed the approximate linear relationships of AC conductivity and dielectric constant vs. heating time. Modified CNF reinforced nanocomposites had less influence by the heating treatments exhibiting better thermal resistance. The change rates of AC conductivity σ and dielectric properties have higher sensitivity to the treatment at a higher temperature. This study provides potential for further research on application of electrical and dielectric signals to detect the effects of heating process on lifetime of polymeric materials.
In this paper, an approach based on data-driven-Compensatory Fuzzy Neural Network (CFNN) is proposed application on a wastewater treatment process. CFNN is a hybrid system combining with compensatory fuzzy logic and neural network. The network has a higher fault tolerance and more stable system. The 98 sets of data are selected as training data, the other 20 sets as testing data to test the validity and generalization. Traditional Back Propagation Neural Network (BPNN) and CFNN approaches are used to establish BOD5(The five-day biochemical oxygen demand) soft-sensing model of a wastewater treatment process. The simulation with the actual industry data shows that the CFNN method obtains higher precision models than the traditional BPNN method. It is sure that the former model is more suitable for practical engineering applications. However, the forecast errors are larger at the volatile point of actual curve using the two methods. Choosing an appropriate compensatory degree can improve the efficiency of the compensatory learning algorithm.