Roughly a year ago, I wrote a post about how LLM’s may come to usurp (or at least defile) the writer’s craft of literary fiction. In this episode, I revisit my initial concerns here, from an era when GPT-4 was only a couple of months old and it looked like we were just at the start of an imminent exponential improvement in further capability (which may no longer be true).1
While I’ve never really cared if LLM-slop automates the production of lower-quality writing (fiction or otherwise), the prospect of a machine being able to write high-quality prose, and this prose, in turn infecting the pristine tablets of our S-Tier wordcels (e.g. future Cormac McCarthy) fills me with revulsion. More worrisome still is the eventual scenario not of partial substitution, but of total replacement, hence my “AI-cope” diagram:
Realistically, the most likely scenario here is something I re-stacked over from
’s excellent Substack: his observation that LLM’s may be used by fiction writers to generatively “fill out” passages of literary description,2 which they can probably be quite good at. has written about fine-tuned LLM’s that can easily mimic the style of an existing writer,3 and I think we are potentially about to see an enormous explosion in the use of this tool in a literary context (you can probably already tell how I feel about this).
DE-FIC Podcast #2: Butterflies for the machine god