For many years I had imposter syndrome because I couldn't read papers.
It took a lot of work to extract any information from a paper; I had to spend hours reading it. I was always impressed at folks who could just glance at a single figure, without referring to the methods, and could glean what the paper was trying to say.
However, after working through enough papers and replicating the results of the authors, I came to learn a number of things- 1) papers are just written badly and it's not beacuse the authors are smart. It's beacuse the authors are bad writers. 2) most papers- in bio, I'd say about 90%- contain invalidating errors which mean that the figures and conclusions are worthless. It takes skilled readers to uncover methodological flaws (or infer them, as often not all the details are included).
After working in bio for a while, it was nice to be in ML because - at least it seemed- I could replicate most papers by downloading the github repo, training on my local GPU for a few days, and then using the trained model to make the same predictions as the papers. Then I realized- in most cases, what was being claimed was far more than what the trained model was actually capable of doing.
Now I stick to well-trod engineering literature that most people consider boring. In nearly all cases I can read the lit, repro the work, and get results that make sense (much of my work is ensuring that published benchmarks are reproducible).
We could have been twins, when I was in graduate school :)
At times I had (still have?) a cynical look at scientific publishing, that it was a symptom of a larger problem: the Peter Principle writ large. Professors and researchers are selected for one thing (their scientific ability), but their major job duties are for something else (here, writing; but also teaching and leadership). Select for one trait, but the job description is for another, and the output is dismal.
To further extend my cynicism: why would a PI follow Cormac McCarthy's advice? Do we have evidence that better-written papers are more "successful," with more citations and shares? My argument (based only on my experiences) is for the opposite: the language is secondary. "Skip to the figures," as I have heard.
To first affirm what you're saying, this [1] is Einstein's revolutionary paper on relativity. While people without a mathematical background may have some difficulty following the technical components, what is said is plain enough that a capable high school graduate ought be able to, at least in general, follow the logic and ideas. Indeed even for Einstein it wasn't that far removed from high school. It was published when he was 26, and working as an assistant examiner in a patent office - having been unable to find any university that would take him on.
Perhaps you're saying as much, in between the lines, but I think there is a specific reason that modern scientific work has degraded in the way that it has. Most of it is rubbish, and the authors know that its rubbish. They don't want to publish rubbish, but finding new real and meaningful science is something incredibly difficult, yet they're expected to constantly publish - or perish. And so their motivation is not to inform society and help push scientific progress forward by a meaningful and relevant new discovery, but simply to publish something that can hit enough checkboxes to get published and keep moving on forward with.
And so in this regard grandiloquent language, excessive jargon/vernacular, and an obfuscation of the fundamental points under the guise of intelligence works as a phenomenal tool. As the countless hoaxes (such as the Sokal affair) have shown so long as you talk the talk and say something that is desired to be heard, you can get published even when what you submit is literally intentional nonsense. Sokal's hoax paper's [3] title was "Transgressing the Boundaries: Towards a Transformative Hermeneutics of Quantum Gravity.".
Sokal hypothesized he could get a paper that is literally nonsensical published in a leading journal if " (a) it sounded good and (b) it flattered the editors' ideological preconceptions." The paper lead with: "There are many natural scientists, and especially physicists, who continue to reject the notion that the disciplines concerned with social and cultural criticism can have anything to contribute, except perhaps peripherally, to their research. Still less are they receptive to the idea that the very foundations of their worldview must be revised or rebuilt in the light of such criticism. Rather, they cling to the dogma...". Yeah, he's already 99% published there.
It took a lot of work to extract any information from a paper; I had to spend hours reading it. I was always impressed at folks who could just glance at a single figure, without referring to the methods, and could glean what the paper was trying to say.
However, after working through enough papers and replicating the results of the authors, I came to learn a number of things- 1) papers are just written badly and it's not beacuse the authors are smart. It's beacuse the authors are bad writers. 2) most papers- in bio, I'd say about 90%- contain invalidating errors which mean that the figures and conclusions are worthless. It takes skilled readers to uncover methodological flaws (or infer them, as often not all the details are included).
After working in bio for a while, it was nice to be in ML because - at least it seemed- I could replicate most papers by downloading the github repo, training on my local GPU for a few days, and then using the trained model to make the same predictions as the papers. Then I realized- in most cases, what was being claimed was far more than what the trained model was actually capable of doing.
Now I stick to well-trod engineering literature that most people consider boring. In nearly all cases I can read the lit, repro the work, and get results that make sense (much of my work is ensuring that published benchmarks are reproducible).
The advice here is golden.