LLMs and Literature: Where Value Actually Comes From
Author challenges critics, showing how detailed prompts push AI beyond statistical averages into creative territory.
A thought-provoking essay on LessWrong by user derelict5432 directly challenges the common critique that LLM-generated writing is inherently shallow because it merely reflects the statistical average of its training data. The author, who has published professionally, argues that literary value does not depend as heavily on human authorial intent as critics assume. The key distinction lies in prompting: generic prompts yield generic, 'average' outputs, but detailed, creative prompts can guide the model into less-populated regions of conceptual space, generating idiosyncratic and potentially valuable text. The essay includes a specific example prompt combining Thomas Pynchon, Harlan Ellison, and H.P. Lovecraft to illustrate this point. The author concludes that while AI will massively increase writing volume, the total quantity of 'genuinely good writing' will also likely increase.
- Challenges the critique that LLMs only produce shallow, statistically average text by highlighting the role of creative prompting.
- Provides a concrete example: prompting for 'a sentient vacuum cleaner in the style of Pynchon, Ellison, and Lovecraft' to generate unique prose.
- Argues AI will increase the total volume of good writing in the world, not just bad, by acting as a creative tool.
Why It Matters
Reframes AI's role in creative writing from mere mimicry to a collaborative tool that can expand literary possibilities when guided skillfully.