What are the potential benefits of using Language Models in communication interfaces applications?5 answersLanguage models (LMs) have potential benefits in communication interface applications. LMs can be used as a general-purpose interface to various foundation models, allowing them to perceive diverse modalities and play the role of a universal task layer. By using a semi-causal language modeling objective, LMs can be jointly pretrained with modular encoders, combining the advantages of both causal and non-causal modeling. This approach enables in-context learning, open-ended generation, and instruction following with finetuned encoders. Experimental results show that LMs outperform or are competitive with specialized models on finetuning, zero-shot generalization, and few-shot learning. LMs can also infer and represent properties of an agent likely to have produced a given textual context, allowing them to model fine-grained communicative intentions and more abstract beliefs and goals. Therefore, LMs can serve as building blocks for systems that communicate and act intentionally.
WHAT FACTORS ARE influencing in e-government acceptance/adoption of citizens?4 answersFactors influencing the acceptance and adoption of e-government by citizens include perceived usefulness, system quality, information quality, service quality, performance expectancy, effort expectancy, perceived transparency, social influence, facilitating conditions, perceived security, government encouragement, and ICT literacy.These factors have been identified through various studies conducted in different countries, including Indonesia. The findings suggest that citizens' acceptance and adoption of e-government services are influenced by the perceived benefits and usefulness of the services, the quality of the systems and information provided, the ease of use, transparency, social influence, and support from the government. It is important for governments to enhance the quality and capabilities of e-government systems, provide reliable and secure services, and promote awareness and literacy among citizens to improve the adoption and usage of e-government services.
Whose Opinions Do Language Models Reflect?4 answersLanguage models (LMs) reflect the opinions of various demographic groups, but there is often a misalignment between the views reflected by LMs and those of these groups. This misalignment persists even after explicitly steering the LMs towards particular demographic groups. Additionally, LLM responses tend to be more similar to the opinions of certain populations, such as those from the USA, and some European and South American countries. However, when prompted to consider a particular country's perspective, the responses shift to be more similar to the opinions of the prompted populations. It is important to note that the opinions reflected by LMs can also reflect harmful cultural stereotypes. Probing language models provides a powerful method for investigating media effects and predicting human responses, but further study is needed to understand the fidelity of neural language models in predicting opinions.
How to use language models to generate?5 answersLanguage models can be used to generate sequences of poses for humanoid robots. They can also be used to generate novel and valid structures in three dimensions for chemical and biomolecular structures. Additionally, language models can be used to generate contextual documents for knowledge-intensive tasks like open-domain question answering. Furthermore, language models can be used in cooperative decoding, where a classifier guides the generation process to produce texts with desired properties.
What are the key factors that influence the adoption of chatbots?5 answersThe key factors that influence the adoption of chatbots include attitude, behavioral intention, perceived risk, performance expectations, effect expectancy, subjective norms, facilitating conditions, perceived trust, and social influence. Studies have shown a positive and significant relationship between attitude and behavioral intention towards the use of chatbots in the banking system. Behavioral intention, in turn, significantly influences the adoption of chatbots in banking. The facilitating conditions of banks strongly influence the behavioral intention of banking services users to use chatbots. In the context of Indonesia's banking and fintech industries, perceived trust, performance expectations, and social influence are the best predictors of a user's behavioral intention to use chatbots. Self-efficacy, effort expectation, and perceived risk are not significant factors. Additionally, factors such as self-efficacy, state anxiety, learning styles, and neuroticism personality traits also influence acceptance and trust in chatbots for risk assessment training.
How can language models be used to enhance the completeness of natural-language requirements?5 answersLanguage models, such as BERT, can be used to enhance the completeness of natural-language requirements. By masking words in requirements and using BERT's masked language model (MLM), contextualized predictions can be generated for filling the masked slots. This allows for the detection of potential incompleteness in requirements by measuring BERT's ability to predict terminology that is missing from the disclosed content. Additionally, BERT can be configured to generate multiple predictions per mask, and a machine learning-based filter can be used to post-process these predictions and reduce noise. Empirical evaluation of this approach using 40 requirements specifications showed that BERT's predictions are highly effective at pinpointing missing terminology, and the filter can substantially reduce noise, making BERT a valuable tool for improving completeness in requirements.