Comment by card_zero

I suspect LLMs would actually struggle significantly with doing it consistently if given purely as a prompt instruction, but you could always constrict the sampling to force words that create a legitimate chain or fine tune / RL in some signal that would assist with it.

An raw LLM no, but given code execution it can do a good job. I had my Claude Sonnet write a sonnet in this form after 14 rounds of iteration with Python:

The autumn wind moves slow across the field,

and every falling leaf now yields its fight.

The summer gold at last has ceased to yield,

and short days now give way to longer night.

I walk alone beside a calm and quiet stream,

and watch it slowly carry every fallen leaf.

I think of you as some half-forgotten dream,

and taste again that same old, bitter grief.

The silent stars come out to watch it still,

and pale moonlight falls gently on the hill.

A lonesome owl calls out from past the mill,

and time moves on, unhurried and calm, until

the dawn returns to paint the sky brand new,

and I still recall the love I lost with you.

"perhaps it would be best to let the browsers do it on screens, and typesetting software for printed materials, since they have access to the font, size, text region boundaries, it comes at the small cost that users don't actually get to read what an author wrote, but increasingly this is an LLM anyways, if its a website or legal contract alike"

[flagged]

Tom 7 did something like this. https://www.youtube.com/watch?v=Y65FRxE7uMc

> BoVeX gives us a controlled tradeoff between these two states. By changing how much it costs for the text to be semantically wrong, we have a dial that allows us to smoothly interpolate between Lorem Epsom and Donald Knuth.