What LLMs are really saving you from
November 11, 2025
I’ve now heard from several early-career folks some variant on this statement regarding manuscripts they intend to publish as papers: “I give the AI all my notes and it gives me a draft that isn’t perfect, but it’s easier to massage that draft into publishable shape than it would be to write it myself.” I haven’t found the right time or the right words to tell them that the work the LLM is “saving” them is the very work that would make them better writers for having gone through it. To me, if you’ve already accepted that writing scientific papers (or writing almost anything, really) is going to be a big part of your career, using an LLM to do the dirty work is like taking an automobile jack to the gym to help you lift the weights. It both cases, it’s the hardest work, pushing the limits of what you thought you could do, that makes you stronger.
This is a pretty stark choice for people whose careers involve writing: do you want to train yourself to be an editor of LLM-generated text? Or do you want to learn how to think more clearly and express yourself better through the medium of writing?
Tutorial 46: how to write an abstract
November 13, 2024
We live in stupid times. As I write this, Google Scholar’s front page is advertising “New! AI outlines in Scholar PDF Reader: skim per-section bullets, deep read what you need”. Yes: it’s using AI to provide a short summary of what’s in a paper. Wouldn’t it be great if instead of a profoundly fallible AI summary, we could read a summary written by the actual authors, who know the material inside out?
We can, of course. It’s called an abstract, and pretty much every paper has one. Back in ye[1] olden times, authors didn’t write them: they were created by third parties to summarise the main points of the article. These days, you — the author of a scholarly article — are expected to provide the abstract, typically 200-300 words or so, that in some sense represents the article.
But in what sense?
Too many abstracts represent their article by trying to whip up enthusiasm for it — not summarizing the findings but teasing the reader as though to pull them in like a clickbait headline. That’s really not how it should be, because the reality is that an order of magnitude more people — maybe two orders of magnitude more — will read your abstract than will read the paper it’s attached to. You can’t make people read your paper (especially if it’s behind a paywall), so you need to make the abstract count.
Here, then, are four simple rules for writing an abstract that gets the job done.
1. Summarize, do not introduce. I’ve written here before about how an abstract should be a surrogate for a paper, not an advertisement for it, and this remains the single most important thing to remember. For those 99 in 100 people who will read the abstract but not the paper, we need to make sure that the abstract tells them what they need to know. So don’t write:
The body size of the snake was estimated based on extant snakes with a morphometric affinity to Palaeophiidae. The cosmopolitan distribution of Pterosphenus schucherti is modelled based on the sea surface temperature (SST) constraints of the modern cosmopolitan snake Hydrophis platurus, and the known fossil localities of the species. The present findings provide crucial insights into the global paleoecological landscape of the Eocene
Instead tell us what what the body size[2] of the snake was, what cosmopolitan distribution you found, and what insights these findings provide in the global paleoecological landscape of the Eocene.
Otherwise people are going to read your abstract, and go away with the message “this snake had a body of some specific size, and had a distribution, and that provides some insights”. Is that what you want? Huh? Huh? No. Didn’t think so.
2. Do not assume specialist knowledge. This is difficult because we work in a specialist field and some amount of specialist terminology is inevitable. But we should still do our best to reduce the amount of background knowledge a reader needs to make sense of our abstracts — again, because the potential readership is so much wider than for our actual papers.
Needless to say (I hope), we should strive to make all our writing as comprehensible as possible — always writing “it flies fast” rather than “the taxon under consideration exhibits velocitous aerial locomotion”. But that goes double for abstracts. So for example, one might write:
The thickness of the articular cartilage between the centra of adjacent vertebrae affects posture. It extends (raises) the neck by an amount roughly proportional to the thickness of the cartilage.
For those versed in the neck-posture wars, it’s hardly necessary to point out that extension of the neck involves raising it (and flexion involves lowering it), but for the wider audience that reads an abstract it’s worth spelling out.
3. Omit needless words. This is, famously, one of Strunk and White‘s rules, and it remains excellent advice. As with “Do not assume specialist knowledge”, this is good advice for all writing but especially so for the abstract.
When I started my palaeo Masters (as it then was) at Portsmouth, I had a very bad habit of writing unnecessary double negatives of the kind the Sir Humphrey Appleby might use. Instead of saying “Taxon X resembles taxon Y”, I would say “is not dissimilar to”. I did this all the time. One of the best things Dave Martill (my supervisor) did for me was to red-pen all these circumlocutions. I hope I’m better now. You be better, too
(When I was doing the video to publicise Brontomerus in 2011, I mentioned Utahraptor near the end, and stupidly described it as “not unlike Velociraptor“. I was so mortified when I heard the first draft of the video that I got in touch with the videographer and begged him to snip out the “not un-“. He did a great job, and I bet you can’t hear the join.)
4. Write the abstract after the paper, not before. It’s so tempting, isn’t it? You have a new, blank document. You type in the title, your name and academic address, and the word “Abstract”. Then you go ahead and summarize what the paper is going to to say.
Except you don’t know what the paper is going to say. Sure, you have a sense in your head of how it’s going to run, but no project plan survives contact with data. In my experience, almost every paper ends up saying something I didn’t anticipate when I started, or leaving out something I did expect to say, or drawing a different conclusion from the one I expected. (When I started to write my 2009 paper on Giraffatitan brancai‘s generic separation from Brachiosaurus altithorax, it was with the intention of proving that “Brachiosaurus” brancai was a perfectly cromulent species of Brachiosaurus, and I ended up discovering the exact opposite.)
So now I leave the abstract till last, so there is no danger that the ghosts of its early version will haunt it. I have a paper in the works now that’s reached 25 single-spaced manuscript pages and 15 illustrations, which I hope to submit within a week — but it’s abstract currently reads as follows:
Abstract
XXX to follow.
Only when the paper is actually complete will I go back, read through it, and summarize as I go, trimming down to 250 words or so if necessary when I’m finished.
Go thou and do likewise.
Notes
- “Ye” should be pronounced “the”, as it was in the days when it used to be written. It was always and only an abbreviation, “y” standing in for the “th” letter-pair. Similarly, the “e” on the end of “olde” was silent. So if arrange to meet your friends at a pub called “Ye Olde Hostelrie”, you should pronounce it “the old hostelry”.
- “Body size” as opposed to what other kind of size? Soul size? When you write “size”, body is understood. Just write size. And while I’m at it, do not give your paper a title like “A new, large-bodied omnivorous bat” or, worse, “A new large-sized genus of Babinskaiidae”. Seriously — large-sized? Again, as opposed to what? Big-hearted?
Tutorial 43: how to do creative work
March 19, 2024
I was struck by a Mastodon post where classic game developer Ron Gilbert quoted film critic Roger Ebert as follows:
The Muse visits during the act of creation, not before. Don’t wait for her.
And Gilbert commented:
I am constantly forgetting this as I procrastinate writing only to discover her again once I start.
In a reply, Gretchen Anderson said her favourite version of this is:
When the muse comes to visit, she better find you working.
I couldn’t find the original source for this, but as I was trying to track it down I ran into this, attributed to Pablo Picasso:
Inspiration exists, but it has to find you working.
When I mentioned these observations to Matt, he sent me a longer-form exposition of the same phenomenon, from painter and visual artist Chuck Close. If the four pithy versions above strike you as a bit facile, the kind of thing you might find on a motivational poster, this may be more actionable:
All the best ideas come out of the process; they come out of the work itself. Things occur to you. If you’re sitting around trying to dream up a great idea, you can sit there a long time before anything happens. But if you just get to work, something will occur to you and something else will occur to you and something else that you reject will push you in another direction. Inspiration is absolutely unnecessary and somehow deceptive. You feel like you need this great idea before you can get down to work, and I find that’s almost never the case.
All of this points back, of course, to the very snappiest version of this observation, attributed to William Wordworth and appearing in the previous tutorial:
To begin, begin.
Why should this be true? I don’t know why, but I know that it is. My best guess is that it’s the same principle by which you can only steer a moving boat. Without some kind of resistance to push against, your rudder can’t exert a useful force; without some raw material for it to work on, your creativity can’t get a toehold. You can’t be creative about nothing. You need something to be creative about.
So the substance of this tutorial is this: whatever project you’re procrastinating on — the novel that you can’t quite figure out the ending of, the paper on a subject that you’ve not read enough background information on, the song that you can’t bring yourself to write because it won’t be as good as a Paul Simon song[*] — just make a start. The worst-case scenario is that you’ll better understand where the real blockages are. A better case is that you’ll find there aren’t any. And the best case is that the very process of getting your hands dirty leads you to fresh new ideas.
I leave you with this brief reminiscence from Greg Gunther, who ran a mailing-list I used to be on for aspiring writers:
I was on an [email] list with Tom Clancy once. Mr. Clancy’s contribution to the list was, “Write the damn book”.
(I am pleased to see that we’ve quoted this at least twice before in SV-POW! tutorials: in my own Tutorial 10: how to become a palaeontologist; and in Matt’s Tutorial 12: How to find problems to work on. It remains by some distance the very best advice I’ve ever heard on writing.)
[*] This last one is very much aimed at myself.
Tutorial 42: how to get a paper written
May 1, 2022
Yes, we’ve touched on a similar subject in a previous tutorial, but today I want to make a really important point about writing anything of substance, whether it’s a scientific paper, a novel or the manual for a piece of software. It’s this: you have to actually do the work. And the way you do that is by first doing a bit of the work, then doing a bit more, and iterating until it’s all done. This is the only way to complete a project.
Yes, this is very basic advice. Yes, it’s almost tautological. But I think it needs saying because it’s a lesson that we seem to be hardwired to avoid learning. This, I assume, is why so many wise sayings have been coined on the subject. Everyone has heard that “A journey of a thousand miles begins with a single step”, attributed to Lao Tzu in maybe the 5th Century BC. More pithily, I recently discovered that Williams Wordsworth is supposed to have said:
To begin, begin.
I love that. In just three words, it makes the point that there is no secret to be learned here, no special thing that you can do to make beginning easier. You just have to do it. Fire up your favourite word processor. Create a new document. Start typing.
And to Wordsworth’s injunction, I would add this:
To continue, continue
Because, again, there is no secret. You just have to do it.
Mounted skeleton of Diplodocus carnegii holotype CM 84 in the rare dorsal view.
At the moment I am working on four separate but related papers. Honestly, sometimes it’s hard even keeping them straight in my head. Sometimes I forget which one I am editing. It would be easy to get overwhelmed and … just not finish. I don’t mean it would be easy to give up: that would be a decision, and I don’t think I would do that. But if I listened to my inner sluggard, I would just keep on not making progress until the matter become moot.
So here is what I do instead:
- I pick one of the papers, which is the one I’m going to work on that evening, and I try not to think too much about the others.
- I figure out what needs to be in that paper, in what order.
- I write the headings into a document, and I put an empty paragraph below each, which just says “XXX”. That’s the marker I use to mean “work needed here”.
- I use my word-processor’s document-structuring facilities to set the style of each of the headings accordingly — 1st, 2nd or occasionally 3rd level.
- I auto-generate a table of contents so I can see if it all makes sense. If it doesn’t, I move my headings around and regenerate the table of contents, and I keep doing that until it does make sense.
- I now have a manuscript that is 100% complete except in the tiny detail that it has no content. This is a big step! Now all I have to do is write the content, and I’ll be finished.
- I write the content, one section at a time. I search for “XXX” to find an unwritten section, and I write it.
- When all the “XXX” markers have been replaced by text, the paper is done — or, at least, ready to be submitted.
Caveats:
First, that list makes it sound like I am really good at this. I’m not. I suck. I get distracted. For example, I am writing this blog-post as a distraction from writing a section of the paper I’m currently working on. I check what’s new on Tweetdeck. I read an article or two. I go and make myself a cup of tea. I play a bit of guitar. But then I go back and write a bit more. I could be a lot more efficient. But the thing is, if you keep writing a bit more over and over again, in the end you finish.
Second, the path is rarely linear. Often I’m not able to complete the section I want to work on because I am waiting on someone else to get back to me about some technical point, or I need to find relevant literature, or I realise I’m going to need to make a big digression. That’s fine. I just leave an “XXX” at each point that I know I’m going to have to revisit. Then when the email comes in, or I find the paper, or I figure out how to handle the digression, I return to the “XXX” and fix it up.
Third, sometimes writing a section blows up into something bigger. That’s OK. Just make a decision. That’s how I ended up working on these four papers at the moment. I started with one, but a section of it kept growing and I realised it really wanted to be its own paper — so I cut it out of the first one and made it its own project. But then a section of that one grew into a third paper, and then a section of that one grew into a fourth. Not a problem. Sometimes, that’s the best way to generate new ideas for what to work on: just see what comes spiralling out of what you’re already working on.
None of these caveats changes the basic observation here, which is simply this: in order to get a piece of work completed, you first have to start, and then have to carry on until it’s done.
How papers are published, in 343 words
February 7, 2022
Many aspects of scholarly publishing are presently in flux. But for most journals the process of getting a paper published remains essentially the same as decades ago, the main change being that documents are sent electronically rather than by post.
It begins with the corresponding author of the paper submitting a manuscript — sometimes, though not often, in response to an invitation from a journal editor. The journal assigns a handling editor to the manuscript, and that editor decides whether the submission meets basic criteria: is it a genuine attempt at scholarship rather than an advertisement? Is it written clearly enough to be reviewed? Is it new work not already published elsewhere?
Assuming these checks are passed, the editor sends the manuscript out to potential reviewers. Since review is generally unpaid and qualified reviewers have many other commitments, review invitations may be declined, and the editor may have to send many requests before obtaining the two or three reviews that are typically used.
Each reviewer returns a report assessing the manuscript in several aspects (soundness, clarity, novelty, perhaps perceived impact) and recommending a verdict. The handling editor reads these reports and sends them to the author along with a verdict: this may be rejection, in which case the paper is not published (and the author may try again at a different journal); acceptance, in which case the paper is typeset and published; or more often a request for revisions along the lines suggested by the reviewers.
The corresponding author (with the co-authors) then prepares a revised version of the manuscript and a response letter, the latter explaining what changes have been made and which have not: the authors can push back on reviewer requests that they do not agree with. These documents are returned to the handling editor, who may either make a decision directly, or send the revised manuscript out for another round of peer review (either with the original reviewers or less often with new reviewers). This cycle continues as many times as necessary to arrive at either acceptance or rejection.
Paul Graham on blogging as a way to generate papers
January 16, 2018
Computer programmer, essayist and venture capitalist Paul Graham writes:
In most fields, prototypes have traditionally been made out of different materials. Typefaces to be cut in metal were initially designed with a brush on paper. Statues to be cast in bronze were modelled in wax. Patterns to be embroidered on tapestries were drawn on paper with ink wash. Buildings to be constructed from stone were tested on a smaller scale in wood.
What made oil paint so exciting, when it first became popular in the fifteenth century, was that you could actually make the finished work from the prototype. You could make a preliminary drawing if you wanted to, but you weren’t held to it; you could work out all the details, and even make major changes, as you finished the painting.
You can do this in software too. A prototype doesn’t have to be just a model; you can refine it into the finished product. I think you should always do this when you can. It lets you take advantage of new insights you have along the way. But perhaps even more important, it’s good for morale.
– Paul Graham, “Design and Research”
Mike and I have long been drawn by the idea that blog posts, like conference talks and posters, could be first drafts of research papers. In practice, we haven’t generated many successful examples. We basically wrote our 2013 neural spine bifurcation paper as a series of blog posts in 2012. And Mike’s 2014 neck cartilage paper grew out of this 2013 blog post, although since he accidentally ended up writing 11 pages I suppose the blog post was more of a seed than a draft.
I should also note that we are far from the first people to do the blog-posts-into-papers routine. The first example I know of in paleo was Darren’s Tet Zoo v1 post on azhdarchid paleobiology, which formed part of the skeleton of Witton and Naish (2008).
Nevertheless, the prospect of blogging as a way to generate research papers remains compelling.
And as long as I’m on about blogging and papers: sometimes people ask if blogging doesn’t get in the way of writing papers. I can’t speak for anyone else, but for me it goes in the opposite direction: I blog most when I am most engaged and most productive, and drops in blogging generally coincide with drops in research productivity. I think that’s because when I’m rolling on a research project, I am constantly finding or noticing little bits that are cool and new, but which aren’t germane to what I’m working on at the moment. I can’t let those findings interfere with my momentum, but I don’t want to throw them away, either. So I blog them. Also the blog gives me a place to burn off energy at the end of the day, when I can still produce words but don’t have the discipline to write technical prose.
– – – – – – – – – – – –
The photo at the top of the post is of Giraffatitan dorsal vertebrae in a case at the MfN Berlin, from Mike’s and my visit with the DfG 533 group back in late 2008. I picked that photo so I could make the following dumb off-topic observation: with its upturned transverse processes, the dorsal on the right looks like it’s being all faux melodramatic, a la:
Sauropod Vertebra Picture of the Week 
