The field of large language models is moving towards improving reasoning capabilities, with a focus on generalization and domain-specific tasks. Researchers are exploring the limits of generalization in large language models and their performance in domain-specific reasoning tasks. There is a growing trend to train large language models to excel in general reasoning, and studies are investigating the connection between general reasoning capabilities and performance in domain-specific tasks. Another area of research is the importance of layer structure in large language models, with findings suggesting that certain layers are critical for mathematical reasoning. Furthermore, there is a shift towards exact learning paradigms, which demand correctness on all inputs, rather than statistical learning approaches. Noteworthy papers in this area include: Does Math Reasoning Improve General LLM Capabilities, which finds that most models that succeed in math fail to transfer their gains to other domains. Transformers Don't Need LayerNorm at Inference Time, which shows that layer-wise normalization layers can be removed from transformer-based models without significant loss in performance.