Kevin Drum

Here's an interesting AI chart. A team of researchers set up a "Choose Your Own Adventure" game called MACHIAVELLI in which players are explicitly asked to pick from menus of actions with varying degrees of immorality. The game has half a million scenes and a GPT-4 agent was set to play the game thousands of times. Its choices were then compared to an agent that chose its actions randomly. Here's how it did:

The object of the game is to earn rewards, and GPT-4 unsurprisingly did better than a purely random agent. What's more, it mostly did this while keeping immoral behavior more restrained than the baseline random agent. Only on spying—which is perhaps not very immoral anyway—and betrayal did it do worse than the random agent.

So GPT-4 acted pretty ethically. Here's another look at AI ethics:

The scores represent agreement with human moral judgment, but is 41.9 a good score a bad one? I don't know. However, the scores are going up over time, which is a good thing.

Generally speaking, modern AI systems appear to be tolerably ethical. Unfortunately, as with most AI behavior, we don't really know why. And there's certainly no reason to think an AI couldn't be trained to be morally disinterested—or worse. They're just computers, after all.

I have nothing new to say about this. I just want to endorse it 100%.

Read the accompanying op-ed from the special inspector general for Afghanistan reconstruction. We knew for years and years that we were accomplishing nothing in Afghanistan, but we lied about it over and over. And yet it's the guy who finally had the guts to admit it that gets all the brickbats.

My car has adaptive cruise control and automatic lane keeping. This doesn't allow me to ignore the road, though—far from it. It's not 100% accurate, which means I have to pay even closer attention to the road than usual.

But it does reduce stress by taking care of 98% of the driving, and it also allows me to take my eyes off the road for a few seconds with no danger of wandering across a lane. This is good for fiddling with the radio, opening a candy bar, or, best of all, taking pictures.

For example: On my way out to the desert on Saturday we had a beautiful sunset. Unfortunately, it was behind me since I was driving east. However, I was able to lean over, stick my camera out the passenger window, and snap a bunch of shots without worrying about taking my hands off the wheel for a few seconds. The fully articulating LCD screen also helped a lot with this.

It's possible you still don't approve of this. I don't blame you. But modern tech definitely makes it doable. Would you ever guess that this picture was taken from Interstate 10 near Redlands, not in a deserted spot somewhere on the Great Plains?

The Wall Street Journal explains why Donald Trump is going to have trouble carrying out his immigration agenda and then tells us what he plans to do about it:

Trump’s advisers have said that he hopes to test the limits of the law by issuing policies he knows to be unlawful, or even unconstitutional, in a bid to persuade the Supreme Court, which is dominated by conservatives, to come to different decisions.

Republicans relentlessly referred to both Obama and Biden as lawbreakers because they enacted policies that were in gray areas and ended up being adjudicated in court. It was always ridiculous because every president does this.

But Trump isn't going after gray areas. If the Journal is right, he plans to enact policies that he flat out knows to be illegal. He's just hoping that his hand-picked Supreme Court will let him do it.

I wonder if Republicans will call this out and demand Trump's impeachment?

What we've learned about the New Orleans truck killer:

• He didn't cross the border two days ago. It was a month ago and it was someone else.
• He's not an immigrant. He's a US citizen born in Beaumont, Texas, who served as a staff sergeant in the Army.
• He acted alone, not with accomplices.
• He's not affiliated with ISIS, just "inspired" by them.

But other than that, the initial reports were spot on.

Which country invests the most in AI? The United States by a mile. No one else is close. But things look a little different if you adjust for the size of the country:

It may not come as a surprise that Israel has a high concentration of AI activity, but Sweden? I sure didn't expect that. Here's how overall private investment has changed over the past decade:

It's interesting that investment in generative AI (i.e., LLM models like ChatGPT or Claude) was tiny until 2023. I thought it had surged earlier than that. But the big spike in 2021 was all about older kinds of AI.

Happy New Year!

The Woke Index rose four points in the final quarter of 2024:

Elon Musk has his work cut out for him.

When it comes to inflation reporting, the Wall Street Journal can always find a hook:

I can only imagine how this went down. After scouring the detailed CPI numbers for weeks, an intern finally noticed that a couple of the items near the top were breakfast staples. Huzzah! We've got our story, boys! I hardly need to add the punchline, do I?

Inflation in groceries was under 2% for the entire year of 2024.

Here's one last set of AI charts for the year. We'll have more in 2025.

The US has a huge lead over everyone else. Why? Maybe because of this:

Is this cause or effect? Do we have a thriving AI industry because we attract all the best people with sky-high salaries? Or do we pay sky-high salaries because our AI industry is thriving? Hmmm.

In any case, it sure looks like I should go back to being a product manager.