llm disclosure on patel.codes

2026-08-21 [home]

motivations

unfortunately, LLMs appear to be here to stay; i have found great utility in various side projects and research efforts. especially in the later, i find myself conflicted about how to honestly communicate results.

while LLMs are horrendus at prose, they are undeniably useful for programming and maths. i never thought i would publish LLM-generated text on my website until an accidentally novel combinatorics proof.

i had to find a way to disclose the use of LLM-generated text: i hate reading it, and i believe it is unfair of me to subject others to it as well without their knowledge.

i do not want people guessing about if my words are my own. i want them focused on the content.

the idealist in me says "i should never publish anything without having fully written everything myself." however, the realist in me has to contend with trading off time spent doing <cool thing> with time spent communicating <cool thing>.

the aforementioned post is a good baseline against which to compare its sharpened counterpart: what i actually proved about A051293.

:::gen

any disclosed, generated content does not mean "untrusted;" instead, it indicates that i did not author it, but i read and verified it.

my site generator is completely bespoke machinery instrumented in Go, so i just added a simple syntax to fence generated content:

:::gen

**this is text i wrote but marked generated.**

:::

which renders as

this is text i wrote but marked generated.

and includes the following banner at the top of the page

this page contains generated content delineated using this shadowing style.

closing thoughts

with several maths results in my pipeline blocked on my bandwidth to elucidate and review the results, it is obvious to me that i need to lean (no pun intended) on the LLMs in lower risk contexts.

personally, i hope to keep generated content to <30% of maths posts i publish; for non-maths or non-technical content, i imagine this will be closer to <10%.

i hope to continue writing and publishing things that do not require any LLM-generated text at all.