Possible pro-tip for non-native English speakers who want to write well but don’t want to sound like AI: Just write an article you want to write in your native language, polish it until you’re proud of it in your native language, and then ask a frontier LLM (Opus 4.8, Gemini 3.1 Pro, ChatGPT 5.5 Pro) to translate it to English,
Sounds good, but this is unworkable in many cases. I can’t imagine writing a high quality article e.g. about AI Safety or just with substantial LessWrong content in my native language. I never read about these in Polish, I never thought about these in Polish.
What I would usually do is: write a bad-English article that has exactly the content I want, ask an LLM to rewrite it (ideally paragraph-after-paragraph, with some clever prompting), iterate until the content is fully preserved. But then, this is actually LLM-written (should I disclose this? I never thought I should).
Your posts seem to work out well for you and you seem to be writing LW-appreciated content, so it’s probably a fine setup here! Eg I never noticed any LLM-ese here, and of course the content is important and novel.
__
(Probably doesn’t apply to you): Some people generally have a native language they have high conversational and literary fluency in but only know the technical jargon in English. For them, I’d also be curious if they tried writing a post in e.g. Polish with English jargon sprinkled in, the way Chinese papers/academic blog posts are often written in Chinese with English jargon sprinkled in. And then machine-translate it.
Jan seems to have a system that already works well for him and does not have (or at least does not obviously have) the standard problems people have with LLM-assisted writing.
The New LessWrong Editor! (Also, an update to our LLM policy.) is the clearest statement of current policy. But it is a little borderline; they say AI can’t write the first draft, and it didn’t, and that it can only do “minor editing” which is ambiguous, but what you describe sounds more than minor. One question is if Pangram thinks its AI; in some of the comments, it sounds like that’s their actual operational criteria for what’s too much AI editing.
One thing to do is tell it to definitely preserve your style while only suggesting alterations to make it look more like a fluent speaker. If you do this, it will probably pass Pangram to their criteria 9which they say is low estimated AI content, but don’t give a number). I’m still a bit unclear on whether you’d need to report it, or whether that’s enough disambiguation to say the edits are effectively minor enough.
This should also help a lot with what people care about, which is writing with standard AI tics.
Right. I phrased that poorly. Visible AI writing is an indicator of AI thinking, which is what readers most want to avoid. If you wrote a complete first draft, you’ve mostly avoided this already. Major edits from the AI will pull it toward being AI thinking. They also make readers assume it’s AI thinking even if it’s mostly human thinking.
Sounds good, but this is unworkable in many cases. I can’t imagine writing a high quality article e.g. about AI Safety or just with substantial LessWrong content in my native language. I never read about these in Polish, I never thought about these in Polish.
What I would usually do is: write a bad-English article that has exactly the content I want, ask an LLM to rewrite it (ideally paragraph-after-paragraph, with some clever prompting), iterate until the content is fully preserved. But then, this is actually LLM-written (should I disclose this? I never thought I should).
Your posts seem to work out well for you and you seem to be writing LW-appreciated content, so it’s probably a fine setup here! Eg I never noticed any LLM-ese here, and of course the content is important and novel.
__
(Probably doesn’t apply to you): Some people generally have a native language they have high conversational and literary fluency in but only know the technical jargon in English. For them, I’d also be curious if they tried writing a post in e.g. Polish with English jargon sprinkled in, the way Chinese papers/academic blog posts are often written in Chinese with English jargon sprinkled in. And then machine-translate it.
I think the practice is technically called code-switching in linguistics, why do you think it doesn’t apply?
Jan seems to have a system that already works well for him and does not have (or at least does not obviously have) the standard problems people have with LLM-assisted writing.
Probably yes.
The New LessWrong Editor! (Also, an update to our LLM policy.) is the clearest statement of current policy. But it is a little borderline; they say AI can’t write the first draft, and it didn’t, and that it can only do “minor editing” which is ambiguous, but what you describe sounds more than minor. One question is if Pangram thinks its AI; in some of the comments, it sounds like that’s their actual operational criteria for what’s too much AI editing.
One thing to do is tell it to definitely preserve your style while only suggesting alterations to make it look more like a fluent speaker. If you do this, it will probably pass Pangram to their criteria 9which they say is low estimated AI content, but don’t give a number). I’m still a bit unclear on whether you’d need to report it, or whether that’s enough disambiguation to say the edits are effectively minor enough.
This should also help a lot with what people care about, which is writing with standard AI tics.
This is not the main concern IMO.
Right. I phrased that poorly. Visible AI writing is an indicator of AI thinking, which is what readers most want to avoid. If you wrote a complete first draft, you’ve mostly avoided this already. Major edits from the AI will pull it toward being AI thinking. They also make readers assume it’s AI thinking even if it’s mostly human thinking.