Kieran Healy
LightNews LightNews
banner
kjhealy.co
Kieran Healy
@kjhealy.co
34K followers 230 following 78 posts
Social science and other distractions. Old posts get deleted pretty quick.
https://kieranhealy.org /
https://theordinalsociety.com
Posts Replies Media Videos
kjhealy.co
Live updates.
Now for the • Weather... Itll be Pissing Rain ALL IVEEKEND
January 24, 2026 at 11:52 AM
kjhealy.co
Plumbing new depths of stupidity and ignorance in the vainglorious pursuit of vice and depravity.
Fucking morons
January 24, 2026 at 8:26 AM
kjhealy.co
Out for a walk today in the old Green & Lovely, where of course it was raining, and came across one of my favorite small Irish things, a “⁊” symbol, or “Tironian et”, seen here embedded in a footpath. I wrote a thing ages ago about this symbol and its long history: kieranhealy.org/blog/archive...
Old cast iron Irish post and telegraph logo showing an access point in a footpath, with a Tironian et, or ⁊ symbol.
January 23, 2026 at 6:31 PM
kjhealy.co
2.3 Back Up Your Work Regardless of whether you choose to use a formal revision control system, you should nevertheless have some kind of systematic method for keeping track of versions of your files. Version-controlled projects are backed-up to some degree if you keep your repository somewhere other than your work computer. But this is not nearly enough. Apple's Time Machine software, for example, backs up and versions your files on your disk, or to a local hard drive, allowing you to step back to particular instances of the fir you want. This still isn't enough, though. You need regular, redundant, automatic, off-site backups of your work. Because you are lazy and prone to magical thinking, you will not do this responsibly by yourself This is why the most useful backup systems are the ones that require a minimum amount of work to set up and, once organized, back up everything automatically without you having to remember to do anything. This means paying for a secure,
January 22, 2026 at 9:02 PM
kjhealy.co
feature of the sharing network. A side effect of Twitter's structure, for example, is that the more successful one becomes in terms of "engagement" or "impact," the more unpleasant one's user experience is likely to be. The desire to measure that impact is our second worm.
January 20, 2026 at 12:33 PM
kjhealy.co
Two people walking in a snowy NYC park this afternoon.
January 18, 2026 at 7:41 PM
kjhealy.co
Amazing numbers. 86% opposed overall. Only 14% overall in favor. More than ten percentage points below the crazification floor! (A handy rule of thumb is that 27% of Americans are just nuts and will agree to anything.) This is what happens when you let twitter-pilled weirdos set policy priorities.
CBS NEWS POLL I YouGov TOTAL SAMPLE MOE: +/- 2.3 PTS JANUARY 14-16, 2026 U.S. USING MILITARY FORCE TO TAKE GREENLAND? BY PARTY WOULD APPROVE WOULD DISAPPROVE DEMS 3% 97% INDS 10% 90% REPS 30% 70%
January 18, 2026 at 3:44 PM
kjhealy.co
It’s like when Nixon gave Elvis a “Federal Agent-at-Large” badge to fight the war on drugs, only the childish pandering is running in reverse.
Nixon panders to Elvis
January 16, 2026 at 12:18 PM
kjhealy.co
Look how pleased with himself he is.
Machado gives Trump her medal
January 16, 2026 at 12:12 PM
kjhealy.co
… that the paper defines “AI” very expansively, including many kinds of analysis that, for most of the data, is not what we now think of as “AI”. Like logistic regression, PCA, LDA, and KNN methods. 🤨 So I feel just a little baited-and switched.
Increasing prevalence of AI adoption in science. a, Increasing performance of AI paper identification during the two-stage fine-tuning of BERT pre-trained models, where we use rough training data in stage 1 to evolve precise assessments in stage 2. We independently train two models on titles (green) and abstracts (purple), and then integrate them into an ensemble (orange) that selects the optimal models during both stages (red stars) to identify all relevant papers. b, Accuracy evaluation of our identification results by human experts. For samples spanning three eras of AI, experts reached consensus, with κ ≥ 0.93. Our model identification results have strong accuracy in validation against expert-labelled data, with an F1-score ≥0.85. c, Relative adoption frequency of the top 15 AI methods across all disciplines for all selected AI development eras. Convolutional neural network (CNN) Support vector machine (SVM) Random forest (RF) Long short-term memory (LSTM) Principal component analysis (PCA) Generative adversarial network (GAN) Gradient boosting Large language model (LLM) k-nearest neighbours (KNN) Logistic regression (LR) Reinforcement learning (RL) Unsupervised learning Transfer learning (TL) Graph convolutional network (GCN) Linear discriminant analysis (LDA)
January 14, 2026 at 11:39 PM
kjhealy.co
Chat about this paper naturally focuses on the headline about AI being high “impact” but also narrowing science; a nice something-for-everyone (including the haters) story. Looking at Fig 2, you might ask “How can they have 30 yrs of impact data for AI?” Well, … www.nature.com/articles/s41...
Fig. 2 | AI enlarges paper impact and enhances researcher careers. a, Average (insets: top 1% and 10%) annual citations after publication of AI (red) and non-AI (blue) papers (n = 27,405,011), where AI papers attract more citations. b, Average annual citations for researchers who useAI and their counterparts who do not(P < 0.001, n = 5,377,346), where researchers who adopt AI receive 4.84 times more citations. c, The probability of two role transitions between junior scientists who adopt AI and their counterparts who do not(n = 46 year observations for each field). Junior scientists who adopt AI have a higher probability of becoming established researchers and a lower probability of exiting academia compared with their counterpartswho do not adopt AI. d, Survival functions for the transition from a junior to an established researcher (P < 0.001,n = 2,282,029). The survival functions can be well-fit with exponential distributions, where junior scientists who adopt AI become established earlier. For all panels, 99% CIs are shown as error bars, with the insets of a centred at the 1% and 10% percentiles and other panels centred at the mean. All statistical tests use a two-sided t-test.
January 14, 2026 at 11:39 PM
kjhealy.co
Dinosaur watching comet approaching earth: This is just a distraction from the real issue
January 12, 2026 at 1:40 AM
kjhealy.co
kieranhealy.org/blog/archive...
The Naming of Stats June 19, 2023 • Obiter Dicta • R The Naming of Stats is a difficult matter, It isn't just one of your holiday games; You may think at first I'm as mad as a hatter When I tell you, a stat must have THREE DIFFERENT NAMES. First of all are the names where usage is informal, Such as Median, Estimate, Average, or Range, Such as Variance, Quartile, or else Standard Normal All of them sensible everyday names. There are fancier names that may be better-tasting, Some for the frequentists, some for the Bayes: Such as Skew, or Kurtosis, Metropolis-Hastings-But all of them sensible everyday names. But I tell you, a stat needs a name that's obscurer, A name that's misleading, and hard to construe, Else how can it keep on confusing the reader, Or frustrate professors, or pass peer review? Of names of this kind, there are many examples, Like Hierarchical, Robust, or Omega-hat, Such as Marginal, Confidence, or just Weighted Sample, Names that always belong to more than one stat. But above and beyond there's still one name left over, And that is the name that you never will guess; The thing that no human research can discover- But THE STAT MIGHT JUST KNOW, and be made to confess. When you notice a stat getting quite widely cited, The reason, you see, is whatever its flaws, Its user's engaged, or enraged, or excited At the prospect of pinning the root of all laws: That inviolable, friable Unidentifiable Deep and inscrutable singular Cause.
January 2, 2026 at 3:26 PM
kjhealy.co
Here’s sample page from my crazy-person evidence-dungeon notebook
crazy-person evidence-dungeon notebook
January 2, 2026 at 3:22 AM
kjhealy.co
Alan Bennett’s diary, 1984.
prefer to be liked and thought a nice man. But I'm not. I'm just as bad as the rest of them, only I don't like to show it. 7 December. To a party at the Department of the History of Medicine at University College. I talk to Alan Tyson, who's like a figure out of the eighteenth century: a genial, snuff-taking, snuff-coloured, easy-going aristocrat - Fox, perhaps, or one of the Bourbons. He is a fellow of All Souls, and when Mrs Thatcher came to the college for a scientific symposium Tyson was deputed to take her round the Common Room. This is hung with portraits and photographs of dead fellows, including one of the economist G. D. H. Cole. Tyson planned to take Mrs Thatcher up to it saying, 'And this, Prime Minister, is a former fellow, G. D. H. Dole? Whereupon, with luck, Mrs Thatcher would have had to say, 'Cole not Dole' In the event he did take her round but lost his nerve. 15 December. Filming this week for Man and Music at Penshurst Place, the home of Lord de L'Isle and Dudley. Everything one 189
December 22, 2025 at 12:01 PM