I would use a different company’s AI to translate it back.
2 Likes
I don’t imagine this is useful for a scientist who cannot speak English. I imagine it’s helpful for people fluent in English as a second language. But even if you are fluent in English, if it isn’t your first language, you may not be able to write as well in it. LLMs can help those people as a writing, assistant and grammar checker. No additional peer needed then.
1 Like
How do you confirm it got the nuance of your results correct?
I’ve reviewed one manuscript where the author used AI as you suggest. Their methods and results were difficult to follow and some of their results were never discussed. The AI wasn’t helpful in resolving their pseudoreplucation issues, but that problem wasn’t it’s job.
1 Like
I think the author still needs to be fluent enough to read and understand the final results. However, I am thinking it will help with the weird constructs you sometimes find in writing by non-native english speaker. I could also imagine it might speed ip the long process of writing. I guess it’s an empirical question. I had trouble remembering and reproducing the kind of writing that I’m thinking of, so I asked a chatbot to rewrite this paragraph in this way. The results are below.
“I am thinking that the author must have still enough fluency for reading and understanding the final results that are produced. But also I am believing this approach will make help with the strange constructions which are sometimes found in the writings of persons whose English is not their first language. It is also possible to imagine that maybe it will bring a speediness to the long process of writing which is usually taking much time. I am supposing this is a question that must be answered by empirical investigation. For myself, I was having difficulty to remember and reproduce the kind of writing that I am having in my mind, and so I made a request to a chatbot that it should rewrite this paragraph in such a manner. The results of this are shown below.”
I would be reluctant to rely on a scientific article that wasn’t peer reviewed.
That 2nd paragraph is kind of lousy…
1 Like
Yep, that 2nd paragraph reads as if it was written by a non-native English speaker (I believe that it was you were trying to achieve?).
Ever since the day Michael Cohen hallucinated court cases, there has been an long running conversation between “AI causes people to make egregious errors” and “those people suck at AI, and frankly they would suck even if they didn’t use AI.”
In theory, AI should improve everyone’s writing. If is masterful at translation. It is great at catching spelling and grammar errors. And it can offer great editing advice in general-- readability, layout, diction, structure, everything, really. And it can at least take a bite out of understanding impossibly difficult topics, like advanced physics, where finding a human editor might be difficult.
All you have to do, in theory, is review its output, pay special attention to the key elements, and understand your own personal limitations in your review. These problems are the same problems you encounter anytime you use any tool, or assistance, or even just write things yourself. You need to check your own work first. Duh.
But people do seem pretty doomed to fail at this kind of due-diligence.
Recent NYT hallucination–
This article was updated after The Times learned that a remark attributed to Pierre Poilievre, … He did not refer to politicians who changed allegiances as turncoats…
Software engineers are running into this same problem, by the way. The term “vibe code” essentially means “write code without even bothering to read it”.
While there can be many ways to validate code, one of the easiest and best is to.. uh.. read it.
I should throw in, my favorite AI reporter of all time was fired for hallucinating a quote.
Of course.
Now, afaik, there are no journalists who are really on top of AI.
I mean no additional peer review for the use of an AI writing assistant.
That is exactly what i’m talking about. That is the kind of writing I think a writing assistant can help avoid. None of those peculiarities have anything to do with understanding the science. I don’t think the LLM can help with understanding science. It can help with the rest.
It is. I think writing assistants can help avoid writing like that.
1 Like
Is this true? I’m not sure it is. For example, if you are a fantastic writer, far above average, and the LLM returns average writing (by definition!) then I’m not sure it does help in theory.
It may be true for most of us.
You’re right. I meant everyone as in “Native English speaking scientists as well.” Not James Joyce.
It could presumably help really fantastic writers at catching little errors, but not at all with prose or organization.
1 Like
I dunno. AI could probably help him to get to the f’ing point. Gah so many words!
2 Likes
I’m not sure that anything you’ve written there supports the idea that AI should write well. It may have proper sentence structure, grammar and spelling, but that’s not all that it takes to write well.
I think I was told earlier in this thread that it was just social media exaggeration when I mentioned seeing complaint skits about this.