Potential impact of large language models on academic writing

LLMs, such as GPT-4 and Bard, are transformer architectures that are based on deep learning techniques. Briefly, they can generate text by predicting the next word in a sentence given all the previous words. Therefore, the choice of which word is used in which place within a sentence depends on probabilities governed by the massive amount of text data used to train them. Additionally, there is an element of randomness intertwined within the LLM algorithm to introduce a measure of inconsistency and simulate a perception of creativity. LLMs can capture long-range dependencies between words, meaning that they can 'understand' the context around a word within a sentence, leading to the perception of coherence and contextual relevance in the generated output. Without keeping how LLMs work in mind, it is easy to anthropomorphise it when interacting with it; one of several factors leading to the sensational uptake and hype in recent months.