6 ms·

I tried to use chatgpt4 to speed up my science writing (no, not by relying on it to read papers for me) as a kind of polisher....but it never seemed to hit the

I tried to use chatgpt4 to speed up my science writing (no, not by relying on it to read papers for me) as a kind of polisher....but it never seemed to hit the mark. Either too bland diplomatic PR speak, or without the needed nuance/precision. Maybe I wasnt good at prompting...but I gave up on it.

jstarfish 3y ago
It's not that you're not good at it, there's just a lot of hidden functionality that isn't obvious, and it's a terrible UI for long-form writing of any sort due to the context size limitations. If you tell it "write a story..." it will write in the style of a children's fairy tale (opening with "once upon a time" and everything). Telling it "write a professional fiction novel in the style of Tom Clancy..." would improve that. Had you told it to "write an academic paper about ___," you should not have received diplomatic PR speak. The lack of precision is addressed in a second pass. For each subsection, you tell it "isolate section ___ and expand it to include x, y and z." Now you have to start copying responses out of the chat window and into a different document, otherwise it'll turn demented and start changing earlier parts of the paper. This sort of workflow has not been worth it for me so far, but YMMV.
- SubiculumCode 3y ago
  The one thing I did like was providing chatgpt a table of means and CI and regression outputs from R for an interaction and asking it to summarize where the differences were.
tkgally 3y ago
If you’re writing in your native language and you are already reasonably literate, GPT-4 won’t help much. But if you are a scientist whose first language is Japanese or Polish or Tamil, you need to write papers in natural-sounding scientific English with few or no mistakes in tense, number, or article usage, and you can’t afford to pay a professional editor, then GPT-4 and similar tools can make the difference between being published and not.

jstarfish 3y ago

It's not that you're not good at it, there's just a lot of hidden functionality that isn't obvious, and it's a terrible UI for long-form writing of any sort due to the context size limitations. If you tell it "write a story..." it will write in the style of a children's fairy tale (opening with "once upon a time" and everything). Telling it "write a professional fiction novel in the style of Tom Clancy..." would improve that. Had you told it to "write an academic paper about ___," you should not have received diplomatic PR speak. The lack of precision is addressed in a second pass. For each subsection, you tell it "isolate section ___ and expand it to include x, y and z." Now you have to start copying responses out of the chat window and into a different document, otherwise it'll turn demented and start changing earlier parts of the paper. This sort of workflow has not been worth it for me so far, but YMMV.

SubiculumCode 3y ago
The one thing I did like was providing chatgpt a table of means and CI and regression outputs from R for an interaction and asking it to summarize where the differences were.

tkgally 3y ago

If you’re writing in your native language and you are already reasonably literate, GPT-4 won’t help much. But if you are a scientist whose first language is Japanese or Polish or Tamil, you need to write papers in natural-sounding scientific English with few or no mistakes in tense, number, or article usage, and you can’t afford to pay a professional editor, then GPT-4 and similar tools can make the difference between being published and not.