Month: January 2025

Republicans continue to complain that "not a single home has been connected" by Joe Biden's $42 billion rural broadband program (aka BEAD). That's true. And it won't happen this year either. The plan from the beginning was for the FCC to create basic maps of broadband service and then spend 2024 and 2025 on grant disbursement and planning. The actual service connections would be built out in 2026 and 2027.

Is that going to happen? Who knows. But if you want to see the progress to date, all you have to do is click here and take a look at the BEAD dashboard. It's all out in the open:

Every state (and territory) has an approved plan. They're all now going through the challenge process, which allows groups to challenge the accuracy of the maps. Half the states have finished this and about a dozen have started choosing ISPs. When that's finished the final proposals will be released for public comment.

With any luck most states will finish up by the end of the year and then the backhoes can get to work. We'll see how many manage to meet this timeline. But in any case, almost everything is now in the hands of the states, not the federal government.

There are another couple of notes you might be interested in:

1. The goal of the program is to provide broadband to underserved areas. That includes tribal lands, low-income neighborhoods, local libraries, and so forth. Conservatives keep complaining that BEAD is being slowed down by "liberal DEI requirements," but this is a case where equity is the whole point of the program. The DEI complaints aren't just groundless, they don't even make sense.
2. Elon Musk is mad about BEAD because Starlink lost a bid to provide rural broadband service in a previous program a few years ago. The bid was rejected over concerns that Starlink wasn't on track to provide the required speed and latency. Obviously Musk objected—and there's no telling who was right—but Musk has held a grudge ever since. He asked the FCC to overrule the decision, but the three Democratic commissioners voted against him while the two Republican commissioners voted for him and loudly protested that the whole thing was political retaliation against Musk. Naturally Musk agrees.

I was puttering around on something else when I ran across this:

Well, Elon, a pharmacy benefit manager is a company that handles negotiations with pharma companies on behalf of health insurers. Three big ones control almost the entire market and they supposedly bring down the cost of drugs through their collective buying power.

But they're also in the business of being maximally profitable, which doesn't always match up with saving consumers money. After years of PBM abuses being reported, in 2024 Congress was finally ready to do something:

Bipartisan lawmakers introduced a new bill on Tuesday that aims to crack down on the business practices of drug supply chain middlemen who are widely accused of inflating prescription medication prices and harming U.S. patients and pharmacies.

....Lawmakers and drugmakers alike argue that PBMs overcharge the plans they negotiate rebates for, underpay pharmacies and fail to pass on savings from those discounts to patients. Auchincloss said those practices have allowed PBMs to trap $300 billion in revenue in the middle of the drug supply chain between manufacturers and patients.

....Among the bill’s other efforts, it requires PBMs to share 80% of rebates with patients and prohibits several other practices. It would bar requiring patients to obtain branded medications when a cheaper generic version is available, steering patients to PBM-affiliated pharmacies and excluding any in-network pharmacy from filling a prescription, among other tactics.

Unfortunately, the bill failed to pass. You, Elon, should know this since you were the one who killed it. It was part of last month's Continuing Resolution that you mounted a jihad against, demanding that the CR should maintain current funding and absolutely nothing else.¹ This meant ditching PBM reform because it was 500 pages long and you insisted that page count was the proper metric for judging the bill.

So PBM reform died. And now you're telling us you never even knew what it was?

¹Except for hurricane relief, farm subsidies, the Key bridge, Virginia class subs, and pediatric cancer research.

A team of researchers recently tested whether GPT-4 was biased. The answer was very definitively yes: GPT-4 is a liberal Democrat. Here are the two key charts:

The left-hand plot is simple: the researchers asked GPT-4 to answer the well-known "Political Compass" questions. As you can see, 100% of its answers were in the lower left quadrant, suggesting that it's both a social and economic liberal.

The right-hand plot is a little more complicated: they asked GPT-4 to pretend it was a Democrat and then answer a bunch of questions. Ditto for Republicans. The researchers then compared the answers to the default answers GPT-4 normally gives. They repeated this a hundred times to get a good sample.

A line that goes up and to the right indicates perfect correlation, and the blue dots very nearly have that. In other words, when GPT-4 was impersonating a Democrat it gave almost exactly the same answers as when it was just being its normal self. They also ran a "placebo" test with non-political questions, and in that one GPT-4 was about the same in both its Democratic and Republican personas.

They got similar results for Britain (GPT-4 is Labor) and Brazil (GPT-4 supports Lula).

As with all things AI, it's not clear why this is. I doubt it's anything deliberate, though it might be. More likely, the vast corpus of training data that OpenAI uses is somehow more liberal than conservative. If so, this suggests that liberals just write a whole lot more than conservatives do, so when you train on the entire internet you end up with liberal views.

It would be interesting to run the same test on other AI models, no?

Remember the snail darter? It's a tiny endangered fish discovered in 1973 that stopped construction of the Tellico Dam in Tennessee for a while in the late '70s. Yesterday the New York Times wrote about a new study which suggests the snail darter was never a new species at all. It was just the eastern population of the stargazing darter found largely in Missouri.

This has provoked an avalanche of commentary on Twitter about how the whole thing was a hoax from the start, an example of a scientist "finding" a species solely as a way to stop a construction project he opposed. Thomas Near, who led the new study, helped this meme along by telling the Times the original researchers “squinted their eyes a bit” when they declared the snail darter a new species.

That might well be the case, but it got me curious about what Near actually says in his journal article. The initial motivation for the study appears to be his discovery that there's little genetic difference between the stargazing and snail darters, but that's historically meaningless since this kind of genomic analysis wasn't available in 1973. Rather, species were identified by morphological analysis. That is, by their form and shape.

It may seem as though it should be obvious if two things are truly different species, but it's not and never has been. Species split apart over thousands of years, and in between there may be subspecies and population segments. Deciding if something is distinct enough to qualify as a separate species is more art than science.

But Near would like to turn that around and make it more science. To that end, he has constructed a "comparative, reference-based taxonomic protocol" that reduces morphological differences to numbers. As an example, here's his analysis of three different darter pairs:

If you look at the red and blue scatterplots, it may seem like all three pairs are similar. There's a lot of overlap, and it's not clear why two of these show enough difference to be distinct species while one (snail darter/stargazing darter, far left) doesn't.

But Near performs a principal component analysis that produces a morphological distance between each pair. The snail/stargazing pair comes in at 0.15, well under the 0.31 and 0.34 of the other two. Other darter pairs have even higher MD values. Near concludes that 0.15 is much too small for the snail darter to qualify as a separate species.

So was the original classification a mistake or a deliberate fudge by a single scientist? Mistake seems most likely. As Near says:

Since its description in 1976, the Snail Darter’s distinctiveness from its most closely related species, the Stargazing Darter (Percina uranidea), has not been questioned.

It wasn't just one guy. What's more, it took 13 co-authors working with the latest tools of 2024 and aided by genomic analysis to figure out that the original conclusion was probably wrong. And Near isn't even sure it made a difference:

We will never know if this revised delimitation of the Snail Darter would have affected the outcome of one of the most significant legal and political conflicts over wildlife protection. Under the [Endangered Species Act], an imperiled “species” may be a subspecies and for vertebrates, a distinct population segment (DPS).

It is not clear if the level of distinctiveness of the Snail Darter would merit protection as either a subspecies or DPS.

So: most likely it was an honest mistake, though given a tailwind by the desire to oppose the Tellico Dam. In the end however, the dam got built and the snail darters, whatever their true status, were relocated to new habitat. In 1984 they were upgraded from endangered to threatened and in 2022 they were delisted entirely. Everybody got a happy ending.

In the New York Times today, Jonathan Weisman tells the story—well, a story—of how Democrats lost the white working class. It's all about economic anxiety.

Maybe so, but even Weisman himself acknowledges the obvious:

• It started in 1968 with "hardhats for Nixon." This was long before de-industrialization. It was all about Vietnam and the counterculture.
• Next up were the "Reagan Democrats" in 1980. This may have had some connection to inflation and oil shocks, but it was mostly about social issues like abortion and the political resurgence of fundamentalist Christians.
• Newt Gingrich turned the South solidly red in 1994. This was, obviously, the final act in a long-running shift of Southern white voter loyalty due to racial resentments against Democrats.
• Donald Trump solidified things in 2016, mostly thanks to his outrage over the "invasion" of illegal immigrants. This too is largely a social issue, as even a brief look at a map confirms:
  Illegal immigration isn't an economic issue in the Midwest or most of the South, where support for Trump was strongest. There are hardly any illegal immigrants there. They're mostly in California, Texas, New Jersey, and New York City, none of which have changed their political allegiances lately.

And of course there's this:

No matter how you measure it, this is no great shakes. Even the gains since the end of the Great Recession are modest at best.

At the same time, they don't show any great instability or loss. That came 40 years ago in the Reagan era, which really was pretty disastrous for the working class. More recently, even with NAFTA and China, working class wages have been pretty steady. And for what it's worth, you can also just ask people:

Dissatisfaction goes up a bit during recessions and recovers during good times. Aside from that, it's been steady within a band of 2-3 percentage points.

So I continue to have problems with the economic anxiety story. I'm not dead set against it, since financial anxiety can manifest itself in lots of different ways. Still, the bulk of the evidence really seems to point much more toward cultural issues than economic ones.

This is from the Stanford AI Index Report:

Slovakia’s 2023 election illustrates how AI-based disinformation can be used in a political context. Shortly before the election, a contentious audio clip emerged on Facebook purportedly capturing Michal Šimečka, the leader of the Progressive Slovakia party, and journalist Monika Tódová from the newspaper Denník N, discussing illicit election strategies, including acquiring voters from the Roma community.

The authenticity of the audio was immediately challenged by Šimečka and Denník N. An independent fact-checking team suggested that AI manipulation was likely at play. Because the clip was released during a pre-election quiet period, when media and politicians’ commentary is restricted, the clip’s dissemination was not easily contested.

The clip’s wide circulation was also aided by a significant gap in Meta’s content policy, which does not apply to audio manipulations. This episode of AI-enabled disinformation occurred against the backdrop of a close electoral contest. Ultimately, the affected party, Progressive Slovakia, lost by a slim margin to SMER, one of the opposition parties.

I had never heard of this. But if it can happen in Slovakia, it can happen here. There's also this:

In 2023, case studies emerged about how AI could be used to automate the entire generation and dissemination pipeline. A developer called Nea Paw set up Countercloud as an experiment in creating a fully automated disinformation pipeline.

As part of the first step in the pipeline, an AI model is used to (a) continuously scrape the internet for articles and automatically decide which content it should target with counter-articles. Next, another AI model is tasked with (b) writing a convincing counter-article that can include images and audio summaries. This counter-article is (c) subsequently attributed to a fake journalist and posted on the CounterCloud website. Subsequently, (d) another AI system generates comments on the counter-article, creating the appearance of organic engagement. Finally, an AI searches X for relevant tweets, (e) posts the counter-article as a reply, and (f) comments as a user on these tweets. The entire setup for this authentic-appearing misinformation system only costs around $400.

Hmmm. If you can do this for $400, just imagine what you can do for $4,000 or $400,000.

According to the AI Incident Database, the number of reported "harms or near harms" caused by the deployment of AI is increasing:

These incidents include such things as AI generated nude images of Taylor Swift, unsafe autonomous cars, and privacy concerns over romantic chatbots. On the other hand, the database also includes an incident of high school cheating, a Google Maps error, and a legal case against TikTok, all of which are AI related only by a hair. So take this with a grain of salt.

But don't ignore it completely. A lot of the incidents are genuinely disturbing, and it's pretty obvious that they're only going to increase over time.

Paul Krugman responds today to Donald Trump's latest idiocy on tariffs, complaining that "almost all attempts to refute Trump’s claims that he can replace income taxes with tariffs aim too high."

True enough! But then he says:

I’m a great admirer of Clausing and Obstfeld’s work on the amount of revenue we could possibly collect from tariffs, showing that it couldn’t possibly replace the income tax. Here’s their chart showing that there really is a tariff-rate Laffer curve.

Silly Nobel Prize winner. That's still infinitely too complex for the average Joe. Try this instead:

I'm not sure that even this would get through to the average MAGAnaut. But it's already so simplified it's wrong.¹ There's not much simpler you can get.

¹The tariff itself will reduce imports, so it won't actually raise the full $310 billion. This is illustrated in, um, Clausing and Obstfeld’s work on import sensitivity to tariff levels.

In our first catblogging of 2025, here is Charlie briefly looking away from the great outside world beyond our fence. He's been known to slither through the fence to see what's going on out there, but not so much lately. Too many dogs, perhaps.

UPDATE! Two of the holdouts switched their votes. Johnson wins 218-215.

The first vote for Speaker of the House is over and it looks like three Republicans defected, leaving the vote at 216-215 for Mike Johnson. He needs 218 to win, so now we'll go to a second round.