Paul Rietschka
prietschka.bsky.social
Paul Rietschka
@prietschka.bsky.social
Data science/machine learning in the Pacific Northwest. In Minneapolis through early 2025.
What are the odds of President Business boring with his Venezuela invasion and just abandoning the whole thing?

His brain is oatmeal and he never had much of an attention span before dementia set in, after all....
January 4, 2026 at 2:47 AM
I cannot think of a better illustration of libertarianism in practice than fascist regime-affiliated individuals using inside knowledge to trade on electronic betting platforms.

I guess crypto is up there, but this is a great illustration of what libertarianism looks like in reality.
A side story is there’s a ton of insider trading on these prediction markets. New accounts making hundreds of thousands on their first and only trade, that maduro would be ousted. Seems like insider trading is legal if not encouraged on these platforms
January 3, 2026 at 8:43 PM
We've been pretending LLMs are a functional technology for 3+ years now, and I'm done.

A technology that doesn't work is not a technology, it's make believe.

LLMs are make believe.
I’ve been testing out LLMs again as part of my unfortunately necessary research as part of my job and I am amazed at how fucking awful it still is at nearly everything. A ton of superficial improvements to cover that it’s still a foundationally broken technology for anything outside of NLP.
January 3, 2026 at 8:14 PM
Saw a post about LLM hallucination rates and wanted to note:

1.) Hallucinations will not be "solved."
2.) Assessing hallucination rates is hard, so estimates are varied and unreliable.
3.) Even the low-ish results shown here by Vectara indicate the tech doesn't really work.
GitHub - vectara/hallucination-leaderboard: Leaderboard Comparing LLM Performance at Producing Hallucinations when Summarizing Short Documents
Leaderboard Comparing LLM Performance at Producing Hallucinations when Summarizing Short Documents - vectara/hallucination-leaderboard
github.com
January 3, 2026 at 6:53 PM
It takes only a cursory look at the finances of this it's-not-a-company-it's-a-cash-incinerator to see it has no future.

But they refuse to admit it. They refuse to see it. They refuse to even look.
January 3, 2026 at 4:37 PM
Wired is spending Saturday morning beclowning itself.
January 3, 2026 at 4:13 PM
The rise of crypto, i.e., electronic counterfeiting designed to create a shadow banking system to enable fraud/criminality, plus the growth of "betting markets" like Kalshi, et al, are both the apex of libertarianism and the most caustic developments of the last decade.

Libertarianism is a disease.
Insider trading on a crypto backed probability market on whether we bomb a foreign country with no justification is like the ur-spirit of the USA right now
This person went on a buying spree over the past 24 hours. Fresh wallet. Only existed since Dec 27th and has only bet on Venezuela-related markets.

polymarket.com/@0x31a56e9E6...
January 3, 2026 at 2:13 PM
The neighbors at the condo in St Paul invited me and the niece over to meet their new (to them) cat: they did what I do, which is go to the shelter and say "who has been here the longest and is likely missing some body parts?" and then you walk out with most of a cat.
January 3, 2026 at 1:45 PM
This is all true, but, my god, it's like people have caught some sort of prion disease.

And Fulk is, of course, right. An algorithm that probabilistically generates an output sequence is not "search," such an algorithm has as much a place in search as strychnine does in a cookie recipe.
January 3, 2026 at 1:31 PM
I’ve never seen a discourse this demented shared by so many.

Mass psychosis. Quite literally.
The way the press is treating chatbots, like they have agency and they “know” things all of a sudden, crossed the line recently from dumb and obsequious to something else entirely. Treating them as alive and making conscious decisions is an almost religious belief. It won’t age well.
January 3, 2026 at 6:46 AM
That quote is demented.
Microsoft CEO Satya Nadella says people should move beyond calling AI 'slop'

"We need to get beyond the arguments of slop vs sophistication and develop a new equilibrium ... that accounts for humans being equipped with these new cognitive amplifier tools as we relate to each other"
January 3, 2026 at 4:49 AM
"OpenAI’s app strategy, launched this fall, is reminiscent of the early days of ChatGPT: An occasional wow moment surrounded by dysfunction, according to Wall Street Journal tests."

Agents are a nonexistent fantasy technology. I'm not being hyperbolic.
OpenAI Is Taking On Apple’s App Store. It’s Got a Long Way to Go.
ChatGPT apps are a key piece of OpenAI’s long-shot bid to replace Apple. Many aren’t yet useful.
www.wsj.com
January 2, 2026 at 10:53 PM
The nephew is into Nigerian hip hop, and recently his friends introduced him to Shabazz Palaces, which, I must say, are the only hip hop-adjacent group I've liked in the past decade.

American hip hop, exemplified by the terrible Kendrick Lamar, is in this awful, stagnant place.
Ode to Octavia part 12: (Sparkles)... Recollections of the Wraith
A short film presented by The Black Constellation Part twelve of thirteen of the Ode to Octavia Series Presented initially during Moment Magnitude at the Frye…
vimeo.com
January 2, 2026 at 10:23 PM
This is what scares me most about the next three years: the ruling class bought themselves, in Nov. 2024, a massive tax cut + four years of regulatory paralysis.

That was what Trump represented to them.

There will be a crash, and soon. But we don’t have a federal government capable of response.
and not, say, on the companies with untold billions of dollars who are uniformly and non-transparently dictating all the contours and applications of AI in a country with no working regulators and a congress too corrupt to function
January 2, 2026 at 6:53 PM
Remember the reporting from six or so months ago that Nadella uses something like 70-plus “agents” to manage his day?

Now we all can recognize this reporting to be a tissue of lies, but it does say something about the deranged, mentally ill discourse around AI.
January 2, 2026 at 6:49 PM
The current ML research landscape is a kaleidoscopic field of perversions the likes of which I’ve never seen.

It’s a field of opium poppies populated by “researchers” alternating between hitting that opium pipe and sniffing their own farts, all posted for posterity on the 8chan of academia, Arxiv.
ML is completely corrupted by LLM money now, you get LLMs through lots of new papers (it's where the $ come from), NeurIPS accepts papers with "we spun the LLM roulette wheel" as part of the methodology

even the mathematics itself won't survive the bubble popping
January 2, 2026 at 4:29 PM
Totally normal response from a totally normal, undeniably respectable company.

A company that, I’ll note, loses $1.2B per month.

Less a company, more a cash incinerator.
January 2, 2026 at 3:05 PM
The target audience of the NYT is a managerial class that doesn’t, in fact, exist. It’s a sort of imaginary managerial class, a sort of theoretically-perfect c-suite that never, at any point, makes contact with reality. www.nytimes.com/2026/01/02/b...
Why Do Americans Hate A.I.?
www.nytimes.com
January 2, 2026 at 1:31 PM
It’s the little things that tell you we are living in a cursed timeline.

Gehry, a postmodern boil on the world, is one of those little things.

That this monstrosity is his last major project, and that it lives in the UAE, home to so much garbage architecture, is very fitting.
January 2, 2026 at 1:22 PM
Audio is an awful user interface. Generally, people find in clunky and dislike it immensely.

Truly hope Altman and Ive go all in on it for their upcoming, already-delayed device.
OpenAI bets big on audio as Silicon Valley declares war on screens | TechCrunch
The form factors may differ, but the thesis is the same: audio is the interface of the future. Every space -- your home, your car, even your face -- is becoming an interface.
techcrunch.com
January 2, 2026 at 1:07 PM
Erik the Rude is going off and I’m here for it.

I’ll only note a couple things:
1.) We’ll likely see this issue evaporate if/when data centers move to closed loop cooling systems.
2.) That data centers haven’t already switched to closed loop cooling is travesty and profligacy. Sheer laziness.
"DATA CENTER WATER USE DOESN'T MATTER" is one of the silliest arguments I see floating around. unless you think re-writing 200 yr old water law (OG political third rail in the US) is a simple matter, the marginal consumption of new water is literally make or break for much of life in the western US
January 2, 2026 at 6:28 AM
And 2026 is the year we find that:

1.) AI hasn't progressed.
2.) Costs have been subsidized and no one is willing to pay true cost.
3.) Metrics are gamed and useless.
4.) "Agents" are make believe.
5.) True demand is only a small fraction of what has been generally assumed.

Welcome to Hell.
In 2025, too many congratulated themselves on AI progress using the wrong measures. Lines of code added. Pull requests opened. Teams saying they “feel faster.” Those are activity metrics that we long considered junk. They are easy to inflate, easy to celebrate, and largely meaningless.
January 2, 2026 at 3:38 AM
A little late, but, please, if you need a new year song please thing of Yeasayer in lieu of that terrible Death Cab For Cutie song I wish to never hear again.
Yeasayer- 2080
YouTube video by YeasayerTV
youtu.be
January 2, 2026 at 12:04 AM
The year in which the lack of actual demand becomes clear.
As the world gets used to the technological feats of AI, the focus is shifting. In 2026 expect its economic, financial and social consequences to grab attention
AI’s true impact will become apparent in the coming year
Will it be boom, bust or backlash? In 2026 expect the economic, financial and social consequences of AI to grab attention
econ.st
January 1, 2026 at 10:02 PM
Please, please, please.

I beg of you, people: do not play that horrible "this is the new year" song by Death Cab For Cutie today.

Resist the urge. Please.
January 1, 2026 at 3:55 PM