iiinova — where your digital self keeps evolving

Cool -- I've always wanted a sequel to Red Plenty. (If you've not read it, it would be right up your alley. It's about the USSR's attempt to "solve" the market via linear programming--and how it fell apart. True-ish to history but fun to read as well!)

020

Matt Krause @prokraustinator.bsky.social · 24/08/2026

It's especially weird because you don't *need* GenAI to do it. Someone could have knocked out the sign in Word 20 years ago. Dropping that onto a suitable background (with perspective distortion) is only slightly trickier--maybe a few minutes in Photoshop or Affinity. And yet....

010

Matt Krause @prokraustinator.bsky.social · 11/08/2026

A reflecting pool, don't you understand? Friend, either you're closing your eyes to a situation you do not wish to acknowledge...Or you are not aware of the caliber of disaster Indicated by the presence of a pool liner in your community.

190

Matt Krause @prokraustinator.bsky.social · 10/08/2026

Today's contestants are Æthelwulf, who is from Æscburna, and Cyneburh, who is from Æscburna. Finished paying off the Weirgeld down there in Æscburna? Æ: Gīese! C: Nā

010

Matt Krause @prokraustinator.bsky.social · 10/08/2026

Well, perhaps *a* zeitgeist but definitely not the current one. There's probably a fun German word for that too, though it looks like it's mostly "der 1920er Jahre" or whatever.

000

Matt Krause @prokraustinator.bsky.social · 07/08/2026

Speaking of COVID.... The study period overlaps with the COVID era (2021-3), so maybe the pathogens already largely controlled? There's a handwavey "but few people were masking in winter 2022-3", but the analysis really ought to include a time factor!

000

Matt Krause @prokraustinator.bsky.social · 05/08/2026

Too weird for a fifty-pack, too rare to not try? Pray for the wicked (blocking the aisle)? Death of a Batchelor (featuring free samples)?

100

Matt Krause @prokraustinator.bsky.social · 31/07/2026

Part of me was curious if a meteor had actually slammed into Charlevoix recently, sparing only Monsieur Gagon and his 12 children but alas, much more prosaic.

110

Matt Krause @prokraustinator.bsky.social · 31/07/2026

I was confused by that too! The meteor hit 450M years ago and there were not even mammals--let alone people--in Quebec then. The claim is that the meteor impact changed the local geography so it was more suited to settlement/farming and that population then spread out.

121

Matt Krause @prokraustinator.bsky.social · 30/07/2026

I asked Opus to check a document for typos, repeated words, etc and the damn thing told me to print it out and do it myself!

010

Matt Krause @prokraustinator.bsky.social · 30/07/2026

I can think of a few reasons but they’re mostly described in neurology and pharmacology books!

000

Matt Krause @prokraustinator.bsky.social · 15/07/2026

I think "n-gram model" and "HMMs" should be on there. You could add Bletchley Park's lot of early work on language modelling during WWII. I'd put "statistical speech recognition" somewhat later: 1980-2000. Those are much more closely related to the implementations of LLMs than expert systems.

030

Matt Krause @prokraustinator.bsky.social · 15/07/2026

This seems a little neural network-centric. There was tons of work on language modelling from the 1970s-mid 2000s, often in the context of speech recognition but for other things too. Check out Fred Jelinek's work, starting from his 1976 paper and his subsequent stuff at IBM.

110

Matt Krause @prokraustinator.bsky.social · 09/07/2026

I feel like this is yet another argument against journals pushing for faster and faster turn-around times. Both the paper itself and the reviews would be better with a little timeout.

020

Matt Krause @prokraustinator.bsky.social · 29/06/2026

Exactly! Participants are always doing *something* and it's almost never "your task." Doubly so when the task is to do nothing at all!

020

Matt Krause @prokraustinator.bsky.social · 26/06/2026

How do we know that it's actually doing a cleanroom re-implementation? It doesn't have internet access, so it can't *search* for it, but the original gotree code has been on the internet for a decade and almost certainly ended up in Claude's training data.

github.com

GitHub - evolbioinfo/gotree: Gotree is a set of command line tools and an API to manipulate phylogenetic trees. It is implemented in Go language.

Gotree is a set of command line tools and an API to manipulate phylogenetic trees. It is implemented in Go language. - evolbioinfo/gotree

010

Matt Krause @prokraustinator.bsky.social · 09/06/2026

It probably also matters how often you're *re*formatting the references. A book has a huge number of references, but it's also (often) going to ONE publisher. Retyping the bibliography after every desk rejection would drive me insane.

100

Matt Krause @prokraustinator.bsky.social · 09/06/2026

[Hastily plans a TMS study, just to mess with you].

100

Matt Krause @prokraustinator.bsky.social · 02/06/2026

Weird how her house was absolutely full of mosquitos.....

000

Matt Krause @prokraustinator.bsky.social · 01/06/2026

As someone who has had to a) eat avocado ice cream, and b) talk someone out of using ANOVAs to "validate" clusters, I'm not sure I agree!

030

Matt Krause @prokraustinator.bsky.social · 01/06/2026

Yup! It was last weekend (her birthday was the 24th, but it shifts to the Monday before the 25th so it makes a long weekend). That is the monarchs' "official" birthday, which is also a nutty concept.

010

Matt Krause @prokraustinator.bsky.social · 01/06/2026

Canada has the same insanity, but a week earlier (May 18-24). Quebec gets the same day off but for different reasons, namely rebellion *against* the English.

010

Matt Krause @prokraustinator.bsky.social · 27/05/2026

See also: "woke", "diverse", etc.

010

Matt Krause @prokraustinator.bsky.social · 24/05/2026

Nah, it makes perfect sense based on how LLMs work. They give a response that’s a *likely* way to continue your prompt, based on (essentially) the entire internet. There’s no actual signal in the data but there’s plenty on related topics (Ginther gap, etc) so that’s what it claims to find.

040

Matt Krause @prokraustinator.bsky.social · 22/05/2026

On top of that... "forbidden partners could even include ... scientists who left the country after the research" How is this supposed to work in practice? Somebody moves abroad and you retract the paper? Maybe resubmit once they come back from Belgium....

2204

Matt Krause @prokraustinator.bsky.social · 20/05/2026

Well... You're just describing a stomach!

190

Matt Krause @prokraustinator.bsky.social · 07/05/2026

Which one?! (I have a few candidates!)

000

Matt Krause @prokraustinator.bsky.social · 22/04/2026

This sounds amazing....and I have all those things already!

100

Matt Krause @prokraustinator.bsky.social · 16/04/2026

May I ask where you got it? This would be super-useful for monitoring stuff in our lab!

110

Matt Krause @prokraustinator.bsky.social · 16/04/2026

That’s a neat way to make a virtue of necessity! A lot of (normal) webcams would be IR sensitive but have a cut filter glued in front of the sensor. I think a lot of the « night vision » cameras are the same but with a low-pass instead.

110

Matt Krause @prokraustinator.bsky.social · 15/04/2026

Wikipedia has a plot of the data and it doesn't look like a gradual converge onto true value at all.

A plot of estimated electron charge vs. time. I see a step change c. 1930, not the pattern Feynman describes.

By Christian Hill - https://scipython.com/blog/measurements-of-the-electron-charge-over-time/, CC BY-SA 4.0, https://commons.wikimedia.org/w/index.php?curid=114882411

030

Matt Krause @prokraustinator.bsky.social · 01/04/2026

Do you....not write cover letters?

100

Matt Krause @prokraustinator.bsky.social · 27/03/2026

What do you mean by "neuromodulation"?

100

Matt Krause @prokraustinator.bsky.social · 26/03/2026

Good thing you don't work at a university. Oh....

100

Matt Krause @prokraustinator.bsky.social · 18/03/2026

I'm afraid I don't... My point was just that disinterested people *have* found some of the claimed effects with the same method (but a different implementation), so this isn't obviously hogwash.

110

Matt Krause @prokraustinator.bsky.social · 10/03/2026

We're having a run of fun talks this week at McGill. Rich Krausliz today on higher visual function in the colliculus (4pm at the Neuro). Then, on Friday, @crisniell.bsky.social will talk about visual coding in mouse and OCTOPUS(!!!).

010