Initially I was sad about #StackOverflow's collapse, not so much because I've contributed a lot to it (and as a #Wikipedian, I've already resigned myself to big tech trying to sell my own words back to me), but because I learned a lot of programming and statistics from it; it gave me focused answers to specific problems. It already bothered me that my students were turning to AI for that purpose, and now what I was telling them to use instead is turning into AI slop... (1/2)
... (2/2) But thinking about it, it was actually a pretty crappy way to learn, making me much slower to see the big picture. I only used SO because I couldn't get formal training in those areas. When I got to the point where I could read books on stats and programming instead, I got a lot better a lot faster.
So I guess now I'll double down on encouraging students to focus on fundamental concepts and the ability to formulate the right questions to ask of high quality sources.
How do you review a paper (in its second revision) where the authors have made an honest attempt to make their analysis (computationally) reproducible, but not quite got there? Given that the field is #archaeology, where reproducibility is still a niche concern and lots of work is published with no regard to it.
@joeroe reproducibility in archeology sounds like "if you dig a hole at your location, with the same dimensions, using an Acme 17 spade, starting on a Tuesday, you should encounter the paleolithic skeletons on the 3rd Friday following"
@johnefrancis Well, there's a lot to say about how the concept does and doesn't translate in the palaeosciences, but in this case I'm talking about computational reproducibility.
Got multiple 500 Internal Server Error responses when trying to download scientific data files from the Harvard Dataverse today... but funnily enough the totally unasked-for (and totally useless) #AI summary of their contents works fine every time :/
Does anyone know of a paper or dataset that compiles the earliest known dates of agriculture in a given region? Globally, ideally, but I'd settle for a continent or two...
(I know there are several options for the European Neolithic, but beyond that?)
@joeroe Sadly also difficult because the data points aren't evenly distributed across the map. "Is this area empty of early agriculture? No, it's just far from cities and universities, so nobody ever excavates there."
> [P]osting unreviewed research on a preprint server is not new or controversial [...] But palaeoanthropology is not a field that needs urgent research and rapid breakthroughs. Given the huge and wide public interest in human evolution and our origins, this research field benefits from much slower, measured, and careful research.
@isakroa What the quote above really made me think about was whether we (i.e. the humanities) are adopting the preprint model because of its merits, or if we're just because it's bundled up with the "open" model of other fields that we want to emulate.
@joeroe Absolutely. I agree thats worth contemplating. Some potential could lie in the ability of the preprint framework to be used not only to get research out quickly. There is of course preregistration, the record of versions, public comments and the post-print, but there are also examples of documents shared without concern for publication (e.g. https://arxiv.org/abs/2011.01808) or to be living documents thats never meant to be finished (e.g. https://lakens.github.io/statistical_inferences/)
Is there a simplified version of the CRediT (https://credit.niso.org/) taxonomy of contributor roles out there? I love the idea but never got why anyone would care about about the four different types of manager or whether someone "curated" or "collected" the data used in a paper.
Agreed that the distinction between "Data curation" (as a specific role) and and collection of data (as part of "Investigation") is not really useful / somewhat confusing, in #CRediT. Also, for example, aren't Methodology and Formal Analysis both part of Investigation?
So yes, I agree heartily with the more general sentiment that a simpler, clearer version of CRediT would be highly desireable. But I also think it is a really good start!
@joeroe recently used the Merit system, which adds more granularity since every author can be called directly in the Methods section with initials so that everyone knows really who did what. Found it nice. https://www.nature.com/articles/s41467-023-37039-1
With #Threads and now with the new #Bluesky bridge, I'm really worried that #MastoAdmin s are doing a bad job of communicating what "federated" means.
Your Mastodon posts are PUBLISHED on the internet. As you read this they are being downloaded, legally, by hundreds of other computers. These computers belong to people you want to share with (other fedizens) and people you don't (surveillance capitalists). There is no reliable way to differentiate them.
I really sympathise with people who shared something with their circle of followers and are now surprised to learn that it might end up on Threads or Bluesky or in some data miner's hoard.
The Mastodon software should absolutely do a better job of telling people that when they sign up (it's in the privacy policy, but you know).
At the same time, nobody is helped by pretending that blocking this or that server will effectively control who gets to access public posts.
If you want or need to control who can read your posts, you should urgently take steps to remove them from Mastodon* and switch to a secure communication protocol. I can recommend #Signal (https://www.signal.org) or #Matrix (https://matrix.org) for secure messaging and #Haven (https://havenweb.org) for private (micro)blogging.
Though note that there is no way of forcing people to honour requests to delete data. Copies will almost certainly remain somewhere, indefinitely.
I just love the #pandoc + #Quarto ecosystem. This morning I wrote up a short statistical analysis for a colleague. One source document in RMarkdown gets me: a Word document for my coauthor, a README for GitHub/Zenodo, and the figure images for the journal. And when the reviewers inevitably suggest changes, I can just change it in one place and regenerate them all!
This Sunday morning my 18 month old son managed to subscribe to #Netflix.
Their Android TV app makes it so easy that apparently a baby fat-fingering a remote can do it, and that remote has a big red "NETFLIX" button on it that will handily install the app for a fat-fingered baby, and Netflix of course retains my account and card details even though I cancelled 8 months ago...
I've now attempted to GDPR-delete my account permanently. I had to email them about it, because of course in this real life Black Mirror episode we live in, after you've found where the delete button lives (https://netflix.com/account/delete - doesn't actually seemed to be linked anywhere in the UI), clicked through all the 'are you sure?' dialogues, and 'confirmed your identity' by entering your payment details again... the button does nothing. Just spawns a little progress spinner, spinning forever.
My new years resolution is to finally #degoogle my life.
I've already replaced a lot of Google products. Some are trickier to extricate myself from (e.g. I've had the same Gmail address for nearly 20 years). Others I just use out of inertia.
There are very few that I'm genuinely struggling to replace. Like Scholar Alerts for finding out about new papers - what can I use instead of that? I know about Semantic Scholar.
tl;dr: this year we started standardising the taxonomic data associated with radiocarbon samples in our chronology database #XRONOS, linking it to the @gbif Backbone Taxonomy.
This was a nice milestone for the project, since it was both our first foray into explicit #LinkedOpenData and into cleaning rather than just accumulating data – which is definitely the harder part!
Piergiovanna Grossi (Wikimedia Italia) at #ArcheoFOSS reports that creating #Wikipedia articles on cultural heritage sites in Italy increased visitor numbers by 23% (tracked over a seven year period).