Researcher Urges Careful LLM Post-Editing For Papers Over Direct Generation
Reactions from ranked influencers
2 posts
PS: Although I do recommend to use LLM to *polish* the text. I do not recommend to generate text directly from say bullets and leave it as. It results in poorly readable slop with all sorts of language atrocities. Moreover, it looks producing readable prose still requires taste and understanding how writing should be done (simple and factual). Writing skills are still required to spot poor results that need to be postedited, perhaps, iteratively.
What's the motivation of people to not use LLMs to polish their writing and make it more understandable and idiomatic? So many poorly written papers. It's completely unforgivable in the era of ChatGPT when everyone can produce native speaker quality English prose. PS: I don't suggest direct generation of papers using LLM, only careful postediting and draft revision.
PS: Although I do recommend to use LLM to *polish* the text. I do not recommend to generate text directly from say bullets and leave it as. It results in poorly readable slop with all sorts of language atrocities. Moreover, it producing readable prose still requires taste and understanding how writing should be done (simple, factual, with necessary definitions and references). Writing skills are still required to spot poor results that need to be postedited, perhaps, iteratively.
What's the motivation of people to not use LLMs to polish their writing and make it more understandable and idiomatic? So many poorly written papers. It's completely unforgivable in the era of ChatGPT when everyone can produce native speaker quality English prose. PS: I don't suggest direct generation of papers using LLM, only careful postediting and draft revision.