Jake Wildstrom

Jake Wildstrom

@dwildstr.bsky.social

Mathematician, tinkerer, crocheter, freelance geek.

100 followers72 following540 posts

Jake Wildstrom reposted

Replying to @asheeshks.org

AI has exposed the austerity con game in higher ed. For years, we have been told there is no money for students, faculty, staff, capital improvements, libraries. But somehow universities have suddenly come up with the money to pay for AI services??

Hey, Democrats, I know you have people to write your talking points for you, but they suck. You're welcome to use some of mine. I'm not any sort of fancy politics-talking person, but I think I can do better than the fancy people you have.

View full thread

When you're asked about surgical transition in youth: "There's well-established medical guidance for when social, hormonal, and surgical transitions are appropriate, based on sound research. Those guidelines are effective in making sure everyone receives age-appropriate care."

When your PR people try to make you triangulate: "The liars of the right will paint us as genital-mutilating perverts no matter what we say or do, and the sort of people inclined to believe them will do so. For everybody else, being kind and thoughtful looks better than capitulation."

Hey, Democrats, I know you have people to write your talking points for you, but they suck. You're welcome to use some of mine. I'm not any sort of fancy politics-talking person, but I think I can do better than the fancy people you have.

When you're asked about trans people in sports: "The government doesn't need to mandate anything here. Professional leagues can make their own rules based on the aspects of human performance they mean to showcase. Amateur sports build community and character and should generally be open to all."

Does any app have a hamburger button that looks like a literal hamburger, i.e 🍔 instead of ≡? Seems like a great opportunity to invent a sort of reverse skeumorphism.

Stick a fork in this trend; it's done. I predict that in about three years legacy newspaper cartoons will discover the "6-7" trend and incorporate it into strips to the befuddlement of their three remaining readers (@jfruh.bsky.social and two other poor souls, presumably).

(Reposting because image upload failed the first time for some reason) Today I was nearly brained by falling hardware while teaching! OK, I exaggerate. I was a good 10 feet away, but on any given day, I quite reasonably could have been near or under the projector screen as it fell.

Over four months later, the university having scheduled classes into NS212C, facilities finally looked into the fallen projector and figured, "huh, maybe we should do something about this before the class scheduled next Monday in this room."

I'm not Hasan Piker's biggest fan but if he can exorcise the malignant spirit of James Carville from the Democratic Party then he can't be _all_ bad.

Donie O’Sullivan@donie.bsky.social

"I am not going to be in the same party with Hasan Piker ... if he becomes a force in the Democratic Party, I am out of here," James Carville says on Fox this morning. Reminder: The Left Insurgency airs tonight on CNN at 8p ET.

Apropos of nothing, looking at trade dress for various editions of the Lord Peter Wimsy stories. It's interesting because they're both murder mysteries — and thus gritty — but also moderately comic, with the central character being a faintly Wodehousian upper-class twit, and art can go either way.

View full thread

Harper's design seems very American, and very mid-90s. I feel like if I had a brain injury where I couldn't decipher text and I saw these, I'd reckon the author was Carl Hiaasen or Elmore Leonard or something: still crime stories and funny, but a bit more gonzo and a bit less British.

But there's one I keep stumbling upon which I _think_ is New English Library, but it's definitely trying very hard not to be fun. Murder is a big deal, y'all, and the NEL wants you to know these are Serious Books for Serious People.

Apropos of nothing, looking at trade dress for various editions of the Lord Peter Wimsy stories. It's interesting because they're both murder mysteries — and thus gritty — but also moderately comic, with the central character being a faintly Wodehousian upper-class twit, and art can go either way.

Open Road Media definitely leans towards "silly and fun", but their trade dress is weirdly inconsistent, with a complete changeover in style from _Have His Carcase_ to _Murder Must Advertise. Both are basically lighthearted though; the "empty suit" design particularly seems to fit the style.

Hachette also keeps things lighthearted, except that they include a little spray of blood to remind you these are murder stories. They have a very consistent style. Notice how they _also_ have a monocle, albeit a stylized one.

Between this and the discovery that solar panel shade can actually _increase_ certain crop yields, people are discovering that there are truly extraordinary opportunities to make better use of the earth's surface area with solar panels. There's so much power waiting to be harnessed!

Ada Palmer@adapalmer.bsky.social

A Chilean salmon farm now draws about 57% of its power from panels laid over the water, cutting roughly 136,000 litres of diesel a year and silencing the generators. With the noise gone, thousands of salmon have gathered in the shade beneath the arrays. Ecoportal buff.ly/CwczkbP #ShareGoodNewsToo

Passing thought for the morning: slow-blinking at a cat and playing peekaboo with a baby are essentially the same thing. "My eyes were gone, and now they're back! This demonstrates my trustworthiness and/or entertainment value!"

Huh, I had somehow missed the news that, since 2021, Maryland no longer had a state song, which is actually good news, since "Maryland, My Maryland" was the worst set of lyrics ever to be graced with the name. But it ought to have a new one. One that doesn't suck.

A hot take thread on LLMs, undercutting the market, and how their strategies are (and are not) like the Uber/Lyft market-invasion plan. The tl;dr: AI is trying to do for creative labor what Uber did for hired drivers, and it's not going to work. Disclaimer: I am not any sort of relevant expert.

View full thread

And with LLMs repriced to a point where ordinary human meatsack work is cheaper, they're going to do just that. Yes, businesses hate having to pay querulous humans and meet their dumb biological demands, but I promise you they like money more than they hate people, and people are cheap.

In conclusion, I figure we are either in for (a) commercial LLMs pricing themselves out of the market and imploding once demand disappears and revenue nosedives, or (b) LLM providers frantically going back to subsidizing models and retaining revenue but spooking investors.

A hot take thread on LLMs, undercutting the market, and how their strategies are (and are not) like the Uber/Lyft market-invasion plan. The tl;dr: AI is trying to do for creative labor what Uber did for hired drivers, and it's not going to work. Disclaimer: I am not any sort of relevant expert.

View full thread

Well, we're seeing it in real time now. Token-based pricing makes everyone who is using LLMs reconsider if it's actually a good deal. With rideshare this didn't happen. After killing taxis, Uber could reprice higher than taxis and it was still a better value proposition than the extant alternatives.

But those creatives that OpenAI and Anthropic promised to make obsolete? They're still there. Most of them haven't even been actually replaced yet by LLMs. Big business is only very tentatively integrating LLM products into their workflow. They can still very easily roll back to their old plan.

A hot take thread on LLMs, undercutting the market, and how their strategies are (and are not) like the Uber/Lyft market-invasion plan. The tl;dr: AI is trying to do for creative labor what Uber did for hired drivers, and it's not going to work. Disclaimer: I am not any sort of relevant expert.

View full thread

Sooner or later, even the dumbest investor is going to realize that you have to raise prices to the point where you sell the product at a profit, and venture capitalists are surely putting pressure on their debtors to actually, y'know, try to actually make money instead of losing it.

The actual details on (3) are pretty extensively covered by other people who dug deeper into the nubmers than me (Ed Zitron most notably), so I'm not going to go into that. But having established that that leads inevitably to jacking prices too early, why's this kill commercial LLMs?

A hot take thread on LLMs, undercutting the market, and how their strategies are (and are not) like the Uber/Lyft market-invasion plan. The tl;dr: AI is trying to do for creative labor what Uber did for hired drivers, and it's not going to work. Disclaimer: I am not any sort of relevant expert.

View full thread

The LLM industry really can't, not without making their product an attractive value proposition for long enough that skilled human labor in their field totally dies out. And that leads into related problems (3) LLMs massively exceed human labor in cost, and (4) they jacked prices too early.

First off, how (3) leads into (4). Savvy investors have finally noticed that pretty much every operator in the LLM space except for NVIDIA is hemorrhaging money horrifically not only on capex but also operations. Again, Uber did that too for a good long time, but never to the same degree.

A hot take thread on LLMs, undercutting the market, and how their strategies are (and are not) like the Uber/Lyft market-invasion plan. The tl;dr: AI is trying to do for creative labor what Uber did for hired drivers, and it's not going to work. Disclaimer: I am not any sort of relevant expert.

View full thread

"Creatives" (or, if we were to drill down to a specific segment of creatives, like "software developers") are a bunch of individual humans who will retain those skills, and likely an enthusiasm to use them, even after you somehow convince their employers to replace them with robots.

They aren't going to come work for the AI companies, because the AI companies's whole schtick is not employing humans. They're going to do something else with their lives and come rushing back into the marketplace the moment companies start wanting human labor again. Uber could kill alternatives.

A hot take thread on LLMs, undercutting the market, and how their strategies are (and are not) like the Uber/Lyft market-invasion plan. The tl;dr: AI is trying to do for creative labor what Uber did for hired drivers, and it's not going to work. Disclaimer: I am not any sort of relevant expert.

View full thread

This leads us to (2) "creative labor" is not already monopolized. Uber succeeded because they were targeting a business, and a highly monopolized business at that. Taxi sevices varied locally, but everywhere they were incorporated, and many places they had weird, exclusive "medallion" systems.

This meant: kill the taxi businesses, and you get to be the only game in town. The individual taxi drivers aren't your problem; hell, a lot of them will come and work for you after you eliminate their current employer. Let's contrast with the market LLMs seek to replace.

A hot take thread on LLMs, undercutting the market, and how their strategies are (and are not) like the Uber/Lyft market-invasion plan. The tl;dr: AI is trying to do for creative labor what Uber did for hired drivers, and it's not going to work. Disclaimer: I am not any sort of relevant expert.

View full thread

This is the least convincing point because, on pure creation (i.e. fiction in any media), LLMs make superficially plausible products whose inferiority to human work is arguably a matter of taste. But when asked to make functional work (e.g. law briefs, research articles, and the all-important CODE)

they fuck up in ways which aren't purely subjective (nonexistent citations, bad statistics, bugs). LLM-adoption while tokens are cheap assumes this inferior product is worth the savings. But even if you don't think LLM product is much worse than human labor, there are big, economic problems.

A hot take thread on LLMs, undercutting the market, and how their strategies are (and are not) like the Uber/Lyft market-invasion plan. The tl;dr: AI is trying to do for creative labor what Uber did for hired drivers, and it's not going to work. Disclaimer: I am not any sort of relevant expert.

View full thread

Looking at the current inflection point in client-side LLM pricing, it's not too hard to draw a parallel between Uber's cheap rideshares to undercut established taxi services, and the service providers' subsidized tokens to undercut a huge number of creative fields (most overtly coding).

But the LLM adoption of this strategy has big, business-killing problems that didn't plague the rideshare biz. To start with the least convincing: (1) LLMs aren't actually capable of matching human labor. In contrast, Uber's competent but unspecialized humans were decent competition for taxidrivers.

A hot take thread on LLMs, undercutting the market, and how their strategies are (and are not) like the Uber/Lyft market-invasion plan. The tl;dr: AI is trying to do for creative labor what Uber did for hired drivers, and it's not going to work. Disclaimer: I am not any sort of relevant expert.

First, a quick primer on what Uber did: they set up a platform better than, and rates cheaper than, local medallion taxi services. Once local taxi services lost market share to the point they couldn't compete, they raised prices. They lost money hand-over-fist on the first half of this deal.

This is not a particularly innovative strategy, and it's a standard part of the monopolist playbook. If you're big with capital reserves, you can squash competition by undercutting them longer than they can survive. Uber did it as a scrappy upstart with VC, but otherwise played the same game.

Jake Wildstrom reposted

ICE is not about immigration. ICE is about normalizing a paramilitary force on peaceful streets. Abolish ICE.

Abdul El-Sayed@abdulelsayed.bsky.social

ICE is not about immigration. ICE is about normalizing a paramilitary force on peaceful streets. Abolish ICE.

I'm using OpenAI Whisper to subtitle a video and am pleasantly surprised by the accuracy. I probably shouldn't be, because AFAICT Whisper uses a trained neural-net model like most cool "AI" did from, oh, 2002 through 2019. It's only the injection of LLM that makes modern "AI" horseshit.

Listening to the America cover of "I Don't Believe in Miracles", thinking, "Colin Blunstone went harder." It's hard to imagine going less hard than Blunstone, but America pulled it off.