When I say “LLMs are bad at writing though they might superficially seem good”, I actually mean many different things.
But one specific way in which I think LLMs are bad at writing in a way that superficially seems good is that they never have more meaning on a second read than the first.
The jokes are almost never multi-layered, the endings are never truly surprising in a way that makes you re-contexualize the rest of the story/essay, you never reread an LLM’s writing and think you understand the world more deeply the second time around.
It’s actually worse than this. Much of an LLM’s public writing is optimized for “group sycophancy.” They flatten points in ways that makes everybody nod along, not realizing that each of them are nodding along to a different reading of the observed text than others are reading, and probably not what the original writer (well, prompter) actually meant.
(Making these observations as of mid-2026. Not making strong claims about future models, or even Opus 4.8 for that matter)
When I say “LLMs are bad at writing though they might superficially seem good”, I actually mean many different things.
But one specific way in which I think LLMs are bad at writing in a way that superficially seems good is that they never have more meaning on a second read than the first.
The jokes are almost never multi-layered, the endings are never truly surprising in a way that makes you re-contexualize the rest of the story/essay, you never reread an LLM’s writing and think you understand the world more deeply the second time around.
It’s actually worse than this. Much of an LLM’s public writing is optimized for “group sycophancy.” They flatten points in ways that makes everybody nod along, not realizing that each of them are nodding along to a different reading of the observed text than others are reading, and probably not what the original writer (well, prompter) actually meant.
(Making these observations as of mid-2026. Not making strong claims about future models, or even Opus 4.8 for that matter)