There's an argument that we keep moving the goalposts on artificial intelligence. This is almost certainly true. But also, I think the blame rests with people that put the goal posts in the wrong place in the first instance, not those who are moving them towards the correct position.

Something I haven't seen discussed much wrt is AI intelligent/conscious is the difference between always on a only responding to requests. I don't have a thought-out thesis. But something that replies to your question and then sits dormant doesn't feel conscious.

When I'm not logged in, my YouTube shorts is almost 100% AI fake prank/fails type videos. When I'm logged in it's mostly just kinda interesting clickbait-y stuff. Not quite sure what to conclude from that.

Here's my favourite example - I was looking for information about a researcher named Vors, at U of Sydney many years ago. I love how the LLM got stuck in the alphabet - exactly what you'd expect from next-token prediction!

More kids game design opinions. Just play battleships 7x7 instead of 9x9 or whatever the default is. It's way more fun (hits are fun, misses are boring).

Apropos of nothing of course. How's everyone's Christmases going?

To the (many) game designers that make games for young kids where you steal each other's cards, please fucking stop it. It always causes an argument. It's clearly more bad feeling for the stealee than good feeling for the stealer. Just bad game design.

I used to make models that tried to look a bit like the real world. Now I rarely do that. Instead I make models that ask questions about the real world. And this is reflected in a fundamental change in what interests me. My core interest now is "what sort of questions can be answered?".

"Church-Turing thesis, Theodore Roszak version: Computers are ridiculous. So is science in general." From Gödel, Escher, Bach. Think I can get behind this.

The {spatialRF} R package is back from the dead 🧟 Version 1.1.5 🚑, now available on CRAN, patches issues introduced by recent API changes across several dependencies. A development roadmap for the coming months is available here:: blasbenito.github.io/spatialRF/ar... #rstats

"the task of replacing a burnt-out light bulb may turn out to require moving a garbage bag; this may unexpectedly cause the spilling of a box of pills, which then forces the floor to be swept so that the dog won't eat the pills." Glad to see Hal's lightbulb is in Gödel, Escher, Bach

Have intros in papers got a lot shorter? Trying to find a good example for a student and half the papers I'm looking at have an intro section that's 3 or even 2 paragraphs long.

I'll forever be reminded of this blog post blog.genesmindsmachines.com/p/if-your-ra...

My band released music. Metal/hardcore type stuff if you're interested. First song has catchiest chorus of any song from any band I've been in. Which makes me happy. open.spotify.com/album/255oBg...

I got a model with attention running in R (lots of help from chatgpt). Now to see what the smallest model/dataset that actually does something interesting is. All just play to help myself understand.

"That's what the Babel of languages is like here. I'm the resulting chromatic chords, we can see that the gods detest tediousness and humans cannot tolerate sameness." The Future of Silence, Kim Aeran

Incredible hand-drawn #dataviz here. Subject matter (F1) may be uninteresting to you but it's definitely worth the watch anyway. youtube.com/shorts/VACix...

Really good explanation of attention in llms. One bit it doesn't cover is how the different bits are learned from data. www.adaptive-ml.com/post/attenti...

There is no reason to put a qr code somewhere that people will be viewing in their phones/computers. That's what links are for. The only (debatable) use of a qr code is printed media or slides. I.e. places people can see but aren't controlling the device.

Today is the last day to submit proposals to speak at the rainbowR conference 🌈 Deadline is midnight, Anywhere on Earth time. #RStats #QueerData

How the fuck is office 365 13GB!? It does the same thing it did 10 years ago. It does the same thing as libre office which is probably like 13kb. Laptop hard drives aren't particularly getting bigger. But software bloat is. Gimme a 4TB laptop hard drive and you can have your 13GB word processor.

Well I can safely conclude that chatgpt is unable to usefully help me with my stata problem. I want to: define a likelihood function evaluate that likelihood over a grid plot it

Two factor is just so annoying. Trying to log in to a shared band email address (not something I use weekly). Having to text the old guitarist to get him to 2FA me. I just want to sign a waiver and have an insecure email address. This is not an important piece of private identity.

CANSSI postdoc w/ Alex Stringer (Waterloo) www.alexstringer.ca and me (McMaster): implementing/exploring Gauss-Hermite quadrature methods in lme4/glmmTMB canssi.ca/wp-content/u... (Alex has shown that Laplace approx is sometimes awful, we'd like to provide alternatives!) canssi.ca/program/dist...

I've been using dreaming Spanish for a couple of years and think it's great. They just released dreaming French if you're looking to learn. www.dreaming.com/french

Brighton, Brighton by the sea, we are Brighton, Brighton by the sea and we'll viz you up, whoever you may be cause we are Brighton, Brighton by the sea.

Huh, never thought of Brighton as a grid. The area I grew up in up in, and place I hung out in like the Lanes weren't. Guess lots of Hove and along the seasfront is.

"I’m more likely now to check the weather forecast for night than day." @mothyblackburn.bsky.social in The Jewel Box. This really tickled me for some reason.

Oh wow. Mind blown. #StatsSky

Wow sheets and pages are shit. A) just bad (possibly subjective and that I'm not used to them) and B) really un-interoperable with office. If libre office can do it I'm sure apple can.

Got back from ASTMH to discover that the Canadian dollar is worth half of what I thought it was! Those beers suddenly feel much less of a burden on my wallet.

Don't apologise for small font on slides, fix it! Streamline sentences, prioritise content and spread over multiple slides.

Thought I'd reflect at the end of my first #TropMed25. - Tons of implementation work. Which is fantastic, but not what I know much about! - Really didn't feel like the AI hype is overtaking. Yes there was some, but you could easily spend a week here without hearing about it. 1/n

Please also check out Olukemi Olowofoyeku's poster 6711. New statistical methods for improving disaggregation regression. #TropMed25

I'll be presenting my poster 6709 on disagapp, tomorrow (Tuesday). user-friendly software for disaggregating disease risk from areal data. Would love to chat and help get you using our software. #tropmed25

My opinion, just use more slides with one, big plot per slide.

I've mostly been to methods conferences recently. Slides with lots of panels, and unreadable small labels seems much more common here in an applied epi meeting. #TropMed25

I've totally failed to look through poster abstracts for #TropMed25. If anyone knows of any stats methods posters please let me know and I'll make sure to go find them.

Having a mic at the front and needing people to walk up to the front to ask questions seems a good way to minimise participation, and especially from ECRs. No ideal I don't think! #TROPMED25

Anyone know this film? It's on in the bar and appears to be completely bonkers. But I have no idea what it is

Phew, almost forgot to listen to yyz - Rush on my way into #ASTMH. My prog credibility would have been irreparably damaged.

If anyone's going to ASTMH a day early and likes punk, pretty fun looking gig on Saturday. I'll be there if my flights behave. files.elfsightcdn.com/eafe4a4d-343...

Has integers changed in #RStats? From memory, something like this would used to have given 1. > 1L / 2L [1] 0.5

Metal always had something missing. Turns out it was missing recorder. www.youtube.com/watch?v=0Tlh...

It took me about three months to get here but... Here's a 🪶 for (nearly) every species of bird, with colors extracted from wikipedia descriptions. 10,151 species. #dataviz #birds #data

student used some AI-assisted code to compute ROC stats with weights and the AI hallucinated pROC::roc(…, weights = weights). (the function doesn’t have a weights argument.) and because the function has a … argument, the fake weights argument was ignored without any warning

New paper, great work by Hannah Worboys. Feasibility study of the P-STEP app. "At 12 weeks, the mean (sd) System Usability Score was 61.68 (22.9)". And lots of useful prelim data for larger studies. #EpiSky #EnvironmentalEpi #copd #asthma

#epinowcast 0.4.0 is out! This release has been a long time coming and contains work from many contributors. It includdes new features, better better and clearer documentation. package.epinowcast.org/news/index.h...

"even in the quieter professions, there is a toil and a labour of the mind, if not of the body, which seldom leaves a man’s looks to the natural effect of time." Alright Jane Austen chill out, I'm just a bit tired this week.

I should probably about stats more... Don't worry about assumptions. New is better than right. The ultimate application is unimportant. Dunking on other statisticians is vital.