@moosewin
DOGE-displaced former civil servant technologist, current developer on analytics/stats for private and nonprofit sector. Amateur poet, recreational philosopher of math, and former photographer. they/she. 🚄🏳️🌈 https://povertyofattention.com/
The underlying theory I have is the Microsoft/Adobe-ification of it all. Glossy, functional machines that churn out unremarkable but useful content for a job, built by companies paranoid about pushing past corporate platitudes, competed by enterprise providers.
bsky.app/profile/moos...
Well, to use the quip from aviation safety, standards are written in blood. Whether or not that's a good thing is left for brighter minds than I.
I'm no Uncle Bob Clean Code Evangelist, not an eXtreme programmer by the standards of Kent Beck, by any means. I, too, have rated against trello boards and story points and kanban like everyone else. But there was a there, there, that I think might get lost in the generation.
Capacity is always matched with constraint, power matched by fragility. The business analyst empowered by code generation, building insecure systems that are easy to attack because there's no least-access principles and solid data pipelines. The loss of DRY, Agile, and clean code principles overall.
My guess is that it becomes like "lean production", as Womack et al wrote, The System That Changed the World. A good many little tweaks here and there that streamline systems but also add fragilities as well, which empowers but also changes, and a trickle that changes the stream, ever so slightly.
I would like to note that I am not a defender of LLMs, or Microsoft and Adobe.
I'm a former government statistical researcher who did data work and now I've decided to spend my life taking vitals and drawing lab samples. Sun Microsystems is dead and so is tech, long live tech.
My more prognostication take is that a good many LLM providers focus products on serving enterprise customers and creating inoffensive, slick graphics and text that don't raise copyright and trademark issues.
Or, the inexorable arc of technological history bends towards Microsoft Office and Adobe.
We shall plumb the depths of cooperative game theory, from proximal policy optimization to the Shapley value, to train models that ensure that the Opus-Chat-Grok-Gemibi-Seek is 10% better at making a good excel formula.
I mean, yeah, my long term guess is that the median use for businesses becomes "excel formula language generator" and "make my email sound like I live and die to work at my 9-5" that occasionally helps people with PowerQuery and (rarely) SQL.
Mathematics as a science of formal systems, turned into a science of formal methods, with those methods as an object of study which includes systems of logic is an interesting argument. A proto-structuralism preceding Linnebo, as JP Seldin has argued.
link.springer.com/article/10.1...
I feel like some of the work Haskell Curry did here on formal systems and methods that are not systems of logic is relevant?
See the notes about the "System of Sams" in Section 5 of his IEP entry:
iep.utm.edu/haskell-broo...
Office automation all carcinizes to the great Excel and VBA machine. Run from it, hide from it, it comes all the same for small to medium size enterprises around the world.
My prediction is that the long term effect of LLMs for the average business is, on average, more custom VBA macros for Excel sheets. The ones have been maintained since the early 2000s as a database system for administration that really shouldn't be like that, but runs the global economy.
The 2028 Republican Primary Padishah Emperor Caudillo Libertadir Macus Rubius versus JD Vance
Relevant:
A search for factors for algorithm understanding results in multiple terms displayed as documents, including available, compact, and aligned. These are shown to be necessary and sufficient. Other, similar terms are shown in the background faded, like intuitive, rule-based, grounded, modular, linear, decomposable, accurate, symbolic, causal, and personalized.
Is the only way we can create algorithms that people understand to make them trivially simple? We argue, no.
People can predict the behavior of algorithms that are arbitrarily complex, if and only if they are available, compact and aligned.
arxiv.org/abs/2601.18966
I do oat milk and Special K instead of wheetabix with my estradiol, can say that my mental well-being is better off for doing so
Heartwarming: All The Worst People You Know Are All Fighting
I just created a series of seven deep-dive videos about AI, which I've posted to youtube and now here. 😊
Targeted to laypeople, they explore how LLMs work, what they can do, and what impacts they have on learning, well-being, disinformation, the workplace, the economy, and the environment.
Writing a GeoPackage(GPKG) library from scratch taught me a lot about some friction between the spec and the modern geospatial ecosystem🦀🌍
GPKG is SQLite. We take that for granted, but have you ever stopped to think about the advantages and disadvantages? Here’s my take.
dev.to/yutannihilat...
The Spring 2026 course website for my grad causal inference class is up and running. Was inspired by @andrew.heiss.phd and @mattblackwell.bsky.social to move my materials over to a standalone site and use as little of Canvas as possible.
www.antonstrezhnev.com/ps813/
I mean, the way I see it, half of it is strategizing? You push your preferred policy all the way because you know you have to get it through a legislature of varying interests.
Easy to start big and cut down to sensible, then start sensible and cut down to inadequate.
🔥I am super excited for the official release of an open-source library we've been working on for about a year!
🪄interpreto is an interpretability toolbox for HF language models🤗. In both generation and classification!
Why do you need it, and for what?
1/8 (links at the end)
> are they trying to dress like Trump too?
This is an insult to the Florida Panthers, they are more than capable of looking like idiots themselves
Not their fault that the President looks like a dipshit and ruins the good name of all earnest dipshits out there
> Sponsored by Jack Daniels
Unironically this is a dumb business decision for them.
As a Brown-Forman shareholder, the stock is down 23.6% YOY. Sponsoring CBS's old dude hour is not exactly the way to revive struggling Jack Daniels sales my dudes
1/ We found that deep sequence models memorize atomic facts "geometrically" -- not as an associative lookup table as often imagined.
This opens up practical questions on reasoning/memory/discovery, and also poses a theoretical "memorization puzzle."
You wanna read a new (hopefully improved, thanks to very helpful comments from a bunch of very smart colleagues) version of paper called "The Political Economy of Shitcoins"? Of course you do. @tompepinsky.com & I are at your service.
papers.ssrn.com/sol3/papers....
The TidyTuesday project has a whole bunch every Tuesday, going back to 2018!
github.com/rfordatascie...
Walder and Strijbis (2023) seems relevant here: the information environment matters.
www.sciencedirect.com/science/arti...
I mean, do voters know?
That's an honest question. We're talking about voters as if they're analyzers of policy, and not selectors of heuristics about partisanship, strength, and other factors including vibes.
We cannot build the future of science and medicine on quicksand.
