Skip to content

Saturday was astronomy night, so I headed out to the desert to take a picture of the Heart Nebula. I've done this already, but my telescope has too long a focal length (900 mm) to capture the whole thing. All I got was a part of the nebula and you couldn't really tell that it was heart-shaped.

So this time I did a mosaic: four separate images that get stitched together in post-processing. It was a lovely night, but unfortunately not everything went well: it seemed as if the telescope wasn't moving in order to capture the four images. It took me a while to confirm this, and eventually I discovered there was a hidden setting called "Slew to target." But why is this a setting? If you've specifically set things up to take four separate frames, why would you not want to slew to each of them? I still have no idea.

In any case, I fixed that but by then I only had a few hours of darkness left. That meant I got only about 40 minutes per frame, and even on a good night that's not enough to get a really detailed image.

Still, as you can see, it worked. The image is decent, the slewing and stitching worked great, and it's a clear proof of concept that I can do this. I even adjusted the rotation of the camera so that the heart was right side up. At least with this one you can tell where it got its name.

December 28, 2024 — Desert Center, California

If Jeff Asher's Real-Time Crime Index is to be believed—and it probably is—our 2020 surge in homicides is now all but over:

On an international basis, our murder rate is very high—thanks mostly to the wide availability of guns—but it's now barely higher than it was before the pandemic. Overall violent crime and overall property crime never surged in the first place and are also at their old levels.

Another 10% decline and the pandemic murder surge will be completely over. At the current rate that will happen by this time next year.

Here are a pair of charts that show the skyrocketing cost of training new AI models. The top chart shows compute cost, with Google's Gemini the current record holder at about 100 billion petaflops. That's 100 yottaflops.

The bottom chart shows training costs in dollars. Gemini clocks in at nearly $200 million. GPT-4 cost nearly $100 million.

Jimmy Carter left office under a hail of scorn, but his reputation has improved over the years. I personally would take this only so far. Carter wasn't a great leader. His malaise speech was a bad idea even if he never used the word malaise. Purging his cabinet in 1979 was an act of public desperation. And while he was unlucky in the timing of the Iranian revolution, he probably also mishandled the situation with the shah. He was also politically tone deaf and straitlaced, as he showed right out of the gate by primly opposing politically untouchable water projects in the West.

At the same time, his substantive legacy was really quite respectable. He was the first president to make human rights a centerpiece of our foreign policy, a stance that Ronald Reagan adopted to great effect and something that’s been a part of American diplomatic relations ever since. He managed to return the Panama Canal to Panama, a courageous but politically costly fight that could have been disastrous if he’d lost it. He helped make peace between Israel and Egypt at Camp David. He appointed Paul Volcker as Fed chairman and allowed him to begin squeezing inflation out of the economy — something that very possibly cost him the 1980 election. He began the wave of deregulation that Reagan and others extended for the next 30 years. He started the war against the Soviets in Afghanistan. And although his efforts to push energy conservation died after he left office, they look prescient now.

It's also worth taking a second look at Carter's economic policy. "Stagflation" was the word that haunted him, and the economy was hardly stellar during his presidency. Still, economic growth was surprisingly strong:

Wages declined under Carter, but starting in 1973 they declined under every president until Bill Clinton. Whatever happened, it had nothing to do with Carter:

Finally, there's inflation, the great bugaboo of the Carter years. But memories aren't always reliable. Take a look at the record:

The great inflation of the '70s started under Nixon and Ford, but Carter was the first to do anything about it. Both the much derided G. William Miller, followed by Paul Volcker, raised interest rates substantially to fight inflation—with Carter's full approval even though it promised to be politically damaging. But it worked. Inflation peaked in 1980 and had already fallen nearly three points by the time Reagan took office. Rates kept coming down after that, but it was the doing of Volcker and the man who appointed him. Reagan played no role at all.

Unemployment was probably Carter's biggest real failure. It went down and then up and he left it no better than he found it. It took the end of the Volcker recession to bring it down for good.

So that's the story. Carter presided over strong growth; got caught in the middle of a two-decade wage decline; and showed more courage in fighting inflation than any president before or after him. It all came undone at the end, but that was due to the Iranian revolution and the oil embargo, events that were outside of anybody's control.

Carter shouldn't be remembered as an economic failure. He should be remembered as a president who was dealt a bad hand but generally played it honestly even at the risk of political blowback. That's better than most.

Not everyone is happy about the money the IRS got a couple of years ago to beef up its tax enforcement of rich people:

This is a common conservative trope: You've spent all this money and gotten nothing! Its most famous use is that lots of money has been allocated for electric charging stations but only nine have been built. But that project was scheduled from the start to go through 2030. There's nothing wrong with the current construction rate (210 charging ports through Q3 of 2024).

The same kind of thing is true here. Let's go through it:

  1. The original funding level for everything related to the IRS was $80 billion.
  2. Of that, $45 billion was for enforcement.
  3. Republicans have either clawed back or frozen about half of that, so we're down to about $22 billion.
  4. That's for all enforcement. Probably around two thirds is for audits of the rich. That puts us in the neighborhood of $15 billion.
  5. As always, this is over ten years. It's probably somewhat front loaded, so it's most likely around $2 billion per year right now.
  6. So if Jayapal is right about the $4.7 billion number, that's $4.7 billion in return for about $4 billion in spending during the program's first two years.
  7. Needless to say, the intent is to ramp up enforcement over the next decade. These cases can take a lot of time, and the payoff will get bigger and bigger each year.

Griping about $4.7 billion compared to $80 billion is either ignorant or dishonest. Only time will tell how successful this enforcement program is over its lifetime, but right now it's perfectly reasonably on track.

A reader brought to my attention the 2024 Stanford AI index Report, and it's basically catnip for someone like me. I shall regale you with various charts from the report over the next few days, but let's start with the good and the ugly (we'll get to the bad later). Here's how basic AI performance has progressed over the past decade:

Impressive. Practically everything is now performing at human level except for high-level math, and it's been on a rocket recently, going from nothing to 90% in only two years. But here's the ugly:

Even the good AI models make up shit 20% of the time, and the bad ones might as well be Donald Trump. This is by far the biggest short-term Achilles heel of AI. Nobody can rely on it for anything serious until this gets cleaned up.

I've either discovered something interesting or else I've made a big mistake. I'm not sure which.

It started with a tweet from the St. Louis Fed that linked to a paper about wage growth. The authors compared wages (from the QCEW survey) to inflation in individual cities between 2019 and Q2 of 2024, but they displayed their results as a complicated scatterplot, which seemed kind of silly to me. Why not just calculate real wage growth in each city and display it?

So I extracted the same information they used for the 20 biggest metro areas and did just that. The only difference is that I started in 2020 because that's all the QCEW tool would give me:

This result is so startling I spent hours trying to figure out where I went wrong. But the QCEW wage data and the inflation data are both straightforward. What I discovered, if the QCEW numbers are reliable, is that since 2020 real wages are down in virtually every big city. I also did a spot check of smallish towns and found the opposite: real wages were consistently up.

Now, the QCEW wage data is more conservative than other sources. Here is nationwide nominal wage growth (i.e., without adjusting for inflation) from every source I could think of:

QCEW is a huge outlier. Partly these differences are because each estimate measures something a little different, but QCEW isn't measuring anything that different from, say, CES or CPS.

But even if you figure QCEW is off and just arbitrarily increase wage growth across the board to bring its national average up to the CPS national average, more than half the big cities in America still show real wage declines—some of them huge:

New York City wages are down 15% even after the arbitrary boost. That's a gigantic decline.

In other words, starting from just before the pandemic, big cities—which are strongly Democratic—mostly show real wage declines while small towns—which are more Republican—show significant real wage increases. This is true even in the best possible case. Does this explain anything about the election? Or is it just some weird statistical artifact? I lean toward this being real, but I'm still not 100% sure.

POSTSCRIPT FOR THE CURIOUS: It turns out that inflation varies widely in big cities across the country. Over the 2020-24 time period, it ranged from a high of 29% in Tampa to a low of 17% in San Francisco.

Which states have the highest violent crime rates? As always, we have two sources for this: the FBI, which compiles records from local police departments, and the NCVS, an annual survey of crime victims.

The FBI releases state totals routinely, so that part is easy. The Bureau of Justice Statistics, however, had to spend an entire decade to produce state-level breakdowns¹—and even then, due to concerns about "disclosing sensitive information," it only produced a single report for the 22 largest states.

And as usual, the data from the two sources had to be normalized. The NCVS calculates crime rates based on the population over 12, so the FBI data had to be adjusted to match. At the same time, the FBI violent crime index includes only aggravated assault, not simple assault, so the NCVS data had to be adjusted to match that. Finally, the FBI by definition includes only crime reported to the police, so NCVS has to be adjusted for that too.

Whew. After that's all done, here's what things look like:

There you have it. And here's one more thing just because. It shows victimization rates by household income:

The median household income is currently $80,000, so these income levels are all below average. Even so, you can see that violent crime is highest in the absolutely most poverty-stricken places—those with incomes under $25,000. Even an income of $30,000, about what a full-time burger flipper makes, is enough to get you into a neighborhood with an average-ish crime rate.

¹Seriously. Here's the timeline for this gargantuan effort:

The Supreme Court will soon decide if banning TikTok violates the First Amendment. Donald Trump has no opinion about that,¹ but he weighed in today with an amicus brief anyway. It's...... unusual:

President Trump is one of the most powerful, prolific, and influential users of social media in history.... Further, President Trump is the founder of another resoundingly successful social-media platform, Truth Social.... On September 4, 2024, President Trump posted on Truth Social, “FOR ALL THOSE THAT WANT TO SAVE TIK TOK IN AMERICA, VOTE TRUMP!”

Furthermore, President Trump alone possesses the consummate dealmaking expertise, the electoral mandate, and the political will to negotiate a resolution to save the platform.... Indeed, President Trump’s first Term was highlighted by a series of policy triumphs achieved through historic deals, and he has a great prospect of success in this latest national security and foreign policy endeavor.

I have no comment on this. I just wanted to share.

¹Page 4, Summary of Argument: "President Trump takes no position on the merits of the dispute."

I haven't checked up on EV sales recently, and I was reminded of it by a New York Times piece about Tesla's recent woes. So I decided to kill two birds with one stone and take a look at both:

EV sales are rising again since flattening for a while in 2023. But Tesla sales are in the tank. US sales have declined over the past couple of years and they've only barely made up for this with sales overseas. Despite this, Tesla stock has surged since the election:

Elon Musk may be the world's richest man, but only because Tesla stock is wildly overvalued. A company with stagnant sales and obvious problems in its biggest market (China) should be selling for a quarter of Tesla's current price—at most.

If that were the case Musk might no longer be the wealthiest person on the planet, but he'd still be worth north of $100 billion. That's not bad no matter how you slice it.