The field of natural language processing is witnessing significant developments in tokenization techniques for large language models. Researchers are focusing on improving the efficiency and effectiveness of tokenization methods to enhance the performance of language models, particularly in multilingual settings. One of the key areas of innovation is the development of novel encoding schemes that can handle non-Western scripts and characters, enabling more robust and accurate tokenization. Another area of research is the evaluation of tokenizers, with a focus on developing reliable and efficient metrics to assess their quality and impact on downstream tasks. Furthermore, researchers are exploring the causal effects of tokenization on language model outputs, highlighting the importance of tokenization as a key design choice in language modeling. Noteworthy papers in this area include: BPE Stays on SCRIPT, which proposes a novel encoding scheme that enables simple and rule-based pretokenization. Beyond Text Compression, which introduces new intrinsic tokenizer metrics that correlate strongly with downstream performance. Causal Estimation of Tokenisation Bias, which estimates the causal effect of tokenization on language model outputs using regression discontinuity design. TokAlign, which proposes an efficient method for vocabulary adaptation via token alignment. Mark My Words, which introduces a robust multilingual model for punctuation restoration in text and speech transcripts.