[dec0dedab0de]

There is a huge difference between using an llm and just blindly dumping it's output on someone verbatim.

I think it's fine to have an llm write a first or second draft of something, then go through and reword most of it to be in your own voice.

[oasisbob]

If one is trying to avoid plagiarism, starting with an AI draft and polishing it to avoid signs of its true origins is not a good method.

[GauntletWizard]

The LLM presents a perverse incentive here - It is used for perceived efficiency gains, most of which would be consumed by the act of rewriting and redrafting. The alienness of the thoughts in the document is also non-condusive to this; Reading a long document about something you think you know but did not write is exhausting and mentally painful - This is why code review has such relatively poor results.

Quite frankly, while having an LLM draft and rewriting it would be okay, I do not believe it is reasonable to expect that to ever happen. It will be either like high school paper plagarism (Just change around some of the sentences and rephrase it bro), or it will simply not even get that much. It is unreasonable with what we know about human psychology to expect that "Human-Rewrites of LLM drafts", at the level that the human contributes something, are maintainable and scalable; Most people psychologically can't put in that effort.

[leptons]

>The LLM presents a perverse incentive here - It is used for perceived efficiency gains, most of which would be consumed by the act of rewriting and redrafting.

It might give efficiency gains for the writer, but the reader has to read the slop and try to guess at what it was intending to communicate and weed out "hallucinations". That's a big loss of efficiency for the reader.

[GauntletWizard]

I completely agree - The efficiency gains are purely from a selfish standpoint.

[leptons]

I just can't seem to square up that the same people that complained left-and-right about "code smells" are the same ones that are shitting out slop code and are proud they shipped 50k lines of code in a week. It's going to be a maintenance nightmare for someone else. I'm not sure how anyone coming in is going to learn a codebase written by LLMs when it's 10x more code than is reasonably needed to solve the problem.

[r_lee]

at this point I really think its better to read broken english than have to read some clanker slop. it immediately makes me want to just ignore whatever text i'm reading, its just a waste of time

[runarberg]

I do wonder, we had pretty good (by some measure of good) machine translations before LLMs. Even better, the artifacts in the old models were easily recognized as machine translation errors, and what was better, the mistranslation artifacts broke spectacularly, sometimes you could even see the source in the translation and your brain could guess the intended meaning through the error.

With LLMs this is less clear, you don’t get the old school artifacts, instead you get hallucinations, and very subtle errors that completely alter the meaning while leaving the sentence intact enough that your reader might not know this is a machine translation error.

[r_lee]

and not just artifacts/hallucinations, the worst thing about is the fact that its basically "perfect" English, perfect formatting, which makes it just look like grey slop, since it all sounds the same and its hard to distinguish between the slop articles/comments/PRs/whatever.

and it will also "clean up" the text to the point where important nuances and tangents get removed/transformed into some perfect literature where it loses its meaning and/or significance

[duped]

I don't think that's fine, I think that's an example of why using LLMs to write is unethical and creates no value.

The purpose of written language is to express your thoughts or ideas to others. If you're synthesizing text and then refining it you're not engaging in that practice.