The field of financial forecasting and risk analysis is moving towards more accurate and robust models, leveraging advances in deep learning and natural language processing. Researchers are exploring innovative architectures and techniques to improve the accuracy of stock price predictions, such as co-attention mechanisms and multimodal language models. Additionally, there is a growing interest in identifying inter-firm risk relations and developing quantitative risk relation scores. Noteworthy papers include: SPH-Net, which introduces a novel co-attention hybrid model for accurate stock price prediction, and HyperNAS, which proposes a neural predictor paradigm for enhancing architecture representation learning. Multimodal Language Models with Modality-Specific Experts for Financial Forecasting from Interleaved Sequences of Text and Time Series is also notable for its unified neural architecture that models interleaved sequences of text and time series data.