What it takes to write a scientific paper
During my time in university I had to write papers. I didn’t really have the chance to get great examples of scientific writing in real world, however I leveraged the internet to see what top universities like Stanford and MIT did and how they did it.
When I was tasked to write something I tried to mimic the way how best of the best did it, by simply mimicking it. Bold text to highlight imports values in a table, footnotes, overall general structure of the paper.
But the hardest part was to do the layout. I had to do all of the papers in MS Word — because that was the de-facto standard. Using Word, you have to do most of the things that you would suppose an editor would do: paragraph identification of exact number of inches, font size, punctuation, references, alignment, and so on…
It was like I was building a web page, except for HTML and CSS I was using a convoluted GUI to manipulate a proprietary modification of XML.
Then I found about LaTeX. Those who know what that is might think: “great, now you actually had an opportunity to type out your thoughts!” Except, now I had to learn one more quirky declarative language with strange plugin system and a compiler so huge the only really useable way to do it was it use Overleaf.
Once again I was tinkering with implementation details instead of simply writing out what I want to say.
Now this is where LLMs become truly valuable. I already know what needs to be done and how (I want a LaTeX article). I don’t care for the dozens of obscure plugins that help render tables. And now I can simply ask Claude (or some other model) to handle that.
The best part is: during my work I already gathered the material and insights. In fact, the LLM already has access to it since I did the work through it.
Now it is easier than ever to create a stunningly beautiful and—most importantly—correct paper, but carefully directing the LLM with the vision of how and what you want to say. And it can handle all of the quirky implementation details. Now it is easier than ever to do great science and publication.