Writing with LLMs

The most common error in using LLMs in writing work is trying to make them do the work for you. While they are powerful tools able to accelerate the writing process and helping you think, the ideas has to come from you in order to have the writing be of good quality. Additionally, while LLMs are getting better at generating text that is not plagiarized from sources directly, some simple experimentation has shown that they sometimes will generate text that can raise suspicions of plagiarism.

Would chatGPT need to resign as minister of higher education? (2024). Vidar Skogvoll https://www.mn.uio.no/kurt/english/news/would-chatgpt-need-to-resign-as-minister-of-higher.html

In this article, I go into detail about a simple research project that I did to see if using LLMs in writing could lead to plagiarism. The results showed that if you simply ask an LLM to write for you, this can lead to issues, but if you instead write a text for yourself and ask it to improve it, then the chances of plagiarism are slimmer.

A very useful way to use LLMs in writing is thus to first write up your text, which may be highly disorganized and not well written, and then ask the LLM to improve it. Then, give the LLM sufficient CRESTO, like asking it to write the text for a specific audience in what tone, and then let it generate the text. A particularly well-suited companion in this endeavor is the world's most dangerous writing app, which will delete your text if you stop writing for more than 5 seconds.