Research Progress on LLMs (Large Language Models): Catastrophic Forgetting, Alignment and Safety and Hallucinations

Dear Colleagues,

The journal invites submissions to a Special Issue on three issues of LLMs (large language models), as follows:

1. Forgetting problems in deep neural networks: Focused on the forgetting problem, also known as catastrophic forgetting, in artificial intelligence systems. The forgetting problem refers to the tendency of neural networks and other machine learning models to forget previously learned knowledge when trained on new data.

• Theoretical foundations and analyses of the forgetting problem.
• Novel architectural designs and training strategies to mitigate forgetting.
• Continual learning algorithms for overcoming catastrophic forgetting.
• Memory consolidation and replay mechanisms for knowledge preservation.
• Evaluation metrics and benchmarks for assessing forgetting in AI systems.
• Applications of continual learning in domains such as computer vision, natural language processing, reinforcement learning, and robotics.
• Neurobiological inspirations and connections to human-like learning and forgetting.

2. Alignment and safety: As LLMs become more powerful, ensuring they behave in ways aligned with human values and intentions is critical.

• Developing robust methods to align LLM outputs with human preferences.
• Mitigating risks such as the generation of false or harmful information.
• Addressing ethical concerns related to bias, fairness, and potential misuse.

3. Hallucination in LLMs is indeed a significant research issue. Here is an overview of this problem: Hallucination in LLMs refers to the phenomenon where a model generates content that is false, inconsistent, or not supported by its training data. This is a critical issue because it can lead to the spread of misinformation and reduce the reliability of AI-generated content. Here are some key aspects of this research area:

• Causes of hallucination: Investigating why LLMs produce hallucinations, which may be due to biases in training data, limitations in model architecture, or flaws in the training process.
• Understanding how a model's attempt to maintain coherence in its outputs can lead to fabricated information.
• Detection methods: Developing robust techniques to automatically identify hallucinated content in LLM outputs.
• Creating benchmarks and evaluation metrics to measure the frequency and severity of hallucinations.
• Mitigation strategies: Exploring methods to reduce hallucinations during training, such as improved data curation or novel training techniques.
• Investigating post-processing techniques to filter out or correct hallucinated content.
• Uncertainty quantification: Improving LLMs' ability to express uncertainty about their knowledge and outputs.

Prof. Dr. Hwan-Seung Yong
Guest Editor

Manuscript Submission Information

Manuscripts should be submitted online at www.mdpi.com by registering and logging in to this website. Once you are registered, click here to go to the submission form. Manuscripts can be submitted until the deadline. All submissions that pass pre-check are peer-reviewed. Accepted papers will be published continuously in the journal (as soon as accepted) and will be listed together on the special issue website. Research articles, review articles as well as short communications are invited. For planned papers, a title and short abstract (about 100 words) can be sent to the Editorial Office for announcement on this website.

Submitted manuscripts should not have been published previously, nor be under consideration for publication elsewhere (except conference proceedings papers). All manuscripts are thoroughly refereed through a single-blind peer-review process. A guide for authors and other relevant information for submission of manuscripts is available on the Instructions for Authors page. Applied Sciences is an international peer-reviewed open access semimonthly journal published by MDPI.

Please visit the Instructions for Authors page before submitting a manuscript. The Article Processing Charge (APC) for publication in this open access journal is 2400 CHF (Swiss Francs). Submitted papers should be well formatted and use good English. Authors may use MDPI's English editing service prior to publication or during author revisions.

• large language models
• catastrophic forgetting
• alignment & safety
• hallucinations