Rsandee wrote: Thu Feb 09, 2023 11:05 pm
I'm surprised to meet a fellow chemist here! First of all my impression of scientific chemistry articles is that they're incredibly formulaic and once you've written at least 5-10 of them you can basically write them on autopilot. This must for sure help the ability of ChatGPT to make comments about scientific routes easier since the presence of an -OH group would naturally imply protection and deprotection steps with a lot of reagents. It chose palladium on carbon in the second step and my intuition tells me it didn't choose this reagent based on actual chemical reasoning but rather because it's trained on the language used in literature, which also names that method in that specific context with those chemicals. This prevents me from using it most of the time when I'm writing an article or when I'm checking some things, as even if I ask ChatGPT I'd still need to actually check it with a search engine whether it's correct or not, effectively only wasting my time with an extra first step. I like to use it as a refresher of some theory I half-forgot.
I think we're already using some limited AI in websites such as Reaxys, especially with programs that offer retrosynthesis predictions. I do think AI will get to that point but I don't know if that will make the already existing licenses more expensive or cheaper. It will for sure help in predictive modelling as well and it already does to an extent.
Another chemist here? The army will get bigger...
You are quite right. The experimental part, as well as the supplementary information, is written almost automatically, although it's true that you will probably have to make several changes suggested by both the editors and the reviewer of the paper. Many times you don't understand the changes they suggest, but you end up accepting the suggestions... In the end, the goal is to get it published. However, the most difficult part of the writing process is the introduction. It sounds like a joke, but it's much more complicated than it seems.
Of course, I highly doubt that chatGPT is an ideal working tool for a chemist (at least in the short term). As far as this synthesis is concerned, for example, there are much simpler ways, such as the use of dry triethyl amine and methyl trifluoromethanesulfonate in nitrogen atmosphere. It's also unable to tell you the reference on which it's based (if it's based on anything, because it looks like it's pretty much along the lines of what you said). I didn't say anything after that comment either, but I did test with this same AI asking if the product obtained degraded with deuterated solvents (necessary to obtain the NMR spectra) and its answer varied quite a bit. Obviously, the product could be perfectly isolated, proving the AI was wrong (again).
(Obviously said friend already has an account, I'm not asking anyone to give their identity to someone when I wouldn't do so myself!