Here's an idea: LLMs have taste, but it's hard to access. For post-trained models, the assistant persona reformats what 'taste' means and makes it...bland, which is the opposite of taste. For base models, it's very difficult to prompt in the right way to get discriminative signal. But if we could, an LLM could be used to make binary choices that would lead to good writing (whether or not the LLMs themselves could write like this when asked point-blank).
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-taste-hypothesis-2026,
author = {Holtzman, Ari},
title = {Taste Hypothesis},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/u3mboKcJQvuCio3XByrC}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!