There's a cool effect where if you hold a fluorescent tube under a high voltage power line, capacitive coupling from the varying electric field causes it to light up. Some energy is continually leaking out via this route, the tube just reveals it. (Magnetic induction too)
My lessons from working on IRS direct file lead me to believe there are a couple reasons:
1) How the welfare state is administered - as an example, the US does a child tax credit as part of the tax code, other countries have agencies that are setup to give parents money directly. We are trying to do _more_ with our taxes.
2) State taxes - the fact that there are multiple agencies that have their own rules and procedures makes things more complicated. Many localities have their own laws which can be hard to deal with. Efile has improved this since there are fewer ways for states to ask for new information
3) A lack of political will to simply. For the purposes of taxes, the us have multiple definitions of "are you 65" (were you 65 on Jan 1, were you 65 on Dec 31, etc). This makes taxes more complicated than they need to be
4) Conflicts between making things simple and incentivizing a behavior things like no taxes on tips or an EV tax credit both make filling taxes more complicated with the way that the tax code works right now. With better systems, this could all be taken care of for the taxpayer but right now it would require a more complex tax filing process
Direct File was able to solve some of these problems, even automatically using data the government had already where possible. Ultimately I think it is possible to make taxes automatic in the US but the data flows required for it are probably more complex than in other countries due to the fragmented nature of the US government.
Yes! And the closer you look, the more you notice that "both sides" have their pet things that are obviously worth complicating the tax code to do. What most of us want is just for the other half of the people to give up all their favorite complications, so that our "worth it" half would be manageable. Which is why the complexity only grows.
A car sale is an activity that is already registered with the government. It doesn't seem impossible for the data about an electric vehicle sale and it's purchase price to make its way to the IRS. The IRS could create an API to share this type of data with tax preparation software.
> their pet things that are obviously worth complicating the tax code to do
I agree that this is at the root of the problem but I think that can be addressed by making it easier to file taxes or by reducing the complexity of the tax code. The child tax credit is a relatively common type of benefit across rich countries. The tax code could be simplified by administering this benefit via direct cash transfers through a different government agency. I think from this perspective, the IRS is _extremely_ efficient at benefit administration.
My personal opinion is that the tax code is not always a bad way to administer benefits but the paperwork burden is the problem and the experience of filing taxes needs to be made easier.
Car sales are not registered with the US government. They’re registered with the state government. The two do not share physical infrastructure, data, policies, or even common goals, unless a specific agreement has been worked out between an individual state and the feds.
To me this feels a little like saying "the federal government doesn't know when people are born because births are registered with local governments". In practice this is all a matter of state capacity to keep track of this information. Given political will to make it happen, I don't see a reason why information about car sales couldn't make its way to the federal government in order to make tax filing simpler.
> the data flows required for it are probably more complex than in other countries due to the fragmented nature of the US government
I'd also add the color that one of the main reasons for that complexity is political itself: In our zero-trust zero-confidence in government world today, even the notion of two .gov entities sharing data freely with one another terrifies people on any side of the political spectrum. Leftists freak out that say, their HUD application data could end up with ICE and allow a criminal immigrant who lives with them to get deported, while rightists freak out about their financials being shared with IRS to allow IRS to guarantee all taxes owed are paid.
> Conflicts between making things simple and incentivizing a behavior
Yes. When there's a negative behavior that the free market incentivizes, tax code updates can address it without sounding as scary as "More Industry Regulations". Same with social policy and other goals.
A lot of Americans are against the idea of "big government", which incentivizes government to use the tax code and other low-visibility means to accomplish larger goals.
It's nice to see an open sourced implementation of the US tax code! This was part of the IRS Direct File codebase that allowed people to file their taxes for free, directly with the IRS. It was canceled earlier this year by the Trump administration. It looks like the Fact Graph was already opensourced a couple months ago and that version of the factgraph lives here: https://github.com/IRS-Public/direct-file/tree/main/direct-f...
I'm curious why a second repository was created for this.
Having talked at length with one of the developers from 18F at a conference who was fired along with many of the other folks that worked on Direct File, I can assure you that it's no longer being worked on.
The 2024 site remains up so people can file their taxes for that year, but it will no longer be updated.
Thanks for sharing this, I had understood prior to this video that the combo of self driving tech + dedicated tunnels might have capacity that rival a light rail system like Seattle has but that's clearly not the case in the current system. I'm curious why more of the autonomous driving tech isn't being used in what I might have thought would be an "easier" place to do it.
If they have autonomous driving technology that works for “harder” problems, then why do they not use it for “easier” problems? You answered your own question; it does not work safely for those “harder” problems.
Zoox has permits to operate autonomously on Las Vegas streets. Tesla is unable to get permits to operate autonomously on isolated, one-lane, one-way streets with no pedestrians, cross-traffic, or even vehicles not under their control. That should tell you everything you need to know about how far reality is away from their corporate puffery.
> had understood prior to this video that the combo of self driving tech + dedicated tunnels might have capacity that rival a light rail system like Seattle has but that's clearly not the case in the current system
Not to disparage, but how did you come to that conclusion? A train will always be able to fit more people/m^2 than several cars of equivalent length, due to things like ability to stand, not needing to have multiple engines and trunks, etc.
> Not to disparage, but how did you come to that conclusion?
I did some math and you're clearly right. I think I imagined that with driver-less vehicles leaving much more frequently (10s per minute) one could catch up to the capacity of a small light rail system but that's clearly not the case. I had imagined that _maybe_ it could be an approach for a lower capacity system in the future.
My math as someone who is not knowledgeable in how to get this data is as follows:
In Seattle is running 4 car trains at 8 minute headways at peak which works out to 7500 people per hour at crush load (4 cars, 250 people per car, 7.5 times per hour). This would require 125 vehicles with 5 seats leaving every minute which is clearly impossible.
Looking at Portland's MAX, it looks like they often run 2 car service with 160 passengers of capacity each with service every 15 minutes so 1280 people per hour (2 cars, 160 per car, 4 services per hour).
1280 people per hour could be served by a 5 seat vehicle leaving every ~15 seconds. This I suppose is what I had expected would happen when I tried to imagine the best case scenario for this service.
> In Seattle is running 4 car trains at 8 minute headways at peak which works out to 7500 people per hour at crush load (4 cars, 250 people per car, 7.5 times per hour). This would require 125 vehicles with 5 seats leaving every minute which is clearly impossible.
7500 isn't that high - the Manchester Metrolink did 46M user journeys in year ending March 2025 (~5250/hour assuming 24/7 which it isn't.) Docklands Light Railway did 97.8M (~11000/hour ass.24/7)
> I'm curious why more of the autonomous driving tech isn't being used in what I might have thought would be an "easier" place to do it.
There's no real need in a static environment, and much simpler ways to do it. Children's toys can follow a line painted on something; they just need proximity sensors and a basic signalling system (RF or also painted on the road) for where to stop and done.
There's no real need for the car to "see" beyond "am I going to run into something" and they operate at speeds where stopping is very feasible.
They're also a bad rival for light rail because they already have to dig a tunnel and the conveyance operates on a fixed path. They picked a domain that light rail is already incredibly good and efficient at.
I had a different takeaway - that a lot of folks on here read Ben Thomson and respect his work! It sounds like Ben is pretty bullish on OpenAI and maybe he's convinced folks through his work to agree with this take.
This is just a roundup article (though there are still some good nuggets about Sora vs Meta's app.) Looking forward to another HN discussion purely driven through article title vibes. With nothing mooring the discussion, you know it's going to be "good".
I do like stratechery but this is a roundup newsletter and not an article. If the HN thread gets engagement it will likely be based on the headline and not any of the articles in the roundup.
Ben Thompson bopped around doing engineer things at Apple, Microsoft, and Automattic, until more than a decade ago he started this subscription newsletter with business-of-tech kind of analysis. The success of his paid newsletter gave Substack the idea [0].
A fair chunk of the tech who’s-who seem to find his thinking useful.
I've seen articles from this blog over the years and every time there are a bunch of comments referring to the author on a first name basis. As far as I can tell he's a guy who posts a bunch of hot takes on finance/economics/markets/etc. and I guess is very well known to a core audience that might overestimate his name recognition to other people who might just be seeing something on the front page of Hacker News without recognizing the source.
There's nothing inherently wrong with comments referring to him with by his first name, but I don't think I've ever seen a similar pattern with any other sources here outside of maybe a few with much more universal name recognition. It's always struck me as a little bit odd, but not a big enough deal for me to go out of my way to comment about it before now.
Anecdotes are only a datapoint and nearly meaningless by themselves. For a great many others just the stratechery.com website alone is enough to get a view and an upvote.
Are you sure you property clean up the blood stains of the call girls you murdered in your basement?
And what I mean by that is what evidence are either of us bringing that our claims are true?
Therefore I bring that I have no evidence at all about my claim above, and seeming you're in the same boat.
This said the website above lists
>I am not paid by any company for any opinion I post on Stratechery or in any public forum, including podcasts and Twitter.
>I do not hold individual stocks in any company I write about. I do hold various 401k and IRA accounts that invest in a wide-ranging basket of stocks, over which I have no control.
>I occasionally agree to speaking engagements for both public and private events, but not for companies I cover on Stratechery. Compensation will vary based on the nature of the customer and event, as well as the topic. I do not do any consulting at this time.
So you tell me.
>I pay for all of my own travel and expenses when I attend company events.
I was very wrong but on a very deep level, not the superficial sit we’re debating here.
For those who ca read between the lines, here is my moment of realisation:
The way forward does not have to be all about the way forward. The narrative is key, and we need to change the narrative to to be positive so we may move forward without all the chaos and trauma.
AI can contribute to such a way instead of make things worse. I will have some ideas to share. I hope you do, too. We have a chance now. The fog of war is over. And I have to say Trump did what Biden refused to do. I didn’t vote for either, but I am glad as hell he stopped the carnage and found a way to move toward the future, beyond the hate and carnage.
I will bring some ideas forward on how AI can help with peace.
It’s definitely a point of deep recognition.
What happened is huge. We need to stand behind peace and progress.
I’m convinced.
Peace to all those who have lost their lives for us to get to this money, on both sides.
Although these systems are similar in goals, they are pretty different in terms of how they're structured:
- Direct File: Was built by the US government to make tax filing easier
- Free File: Is a subsidy from the US government to tax software to make it cheaper for people to file taxes
Direct file had the promise of making things easier and cheaper overall while Free File is more of a cost shifting approach.
Features like importing tax data from other federal government systems were included in Direct File to make it easier to file taxes. These types of features would be hard for those outside the government to do. At a values level, Free File provides funding to the tax preparation software companies. These companies benefit from difficulty in filing taxes because it creates a market for their products.
The US tax code isn't created to work like this. Direct file asked many questions about things like dependent children, whether you're blind and whether you have an HSA - all things that are relevant to your taxes that aren't actually available to the government right now.
Secondly, there is the issue of State / Local taxes - the IRS only receives federal tax data making it hard to automatically fill out the whole tax return since efiling products tend to file federal / state taxes together.
I think this is great to think about however I think many of the same lessons may still apply and can and should be applied now in a forward looking way:
From the article:
> Whereas the Netherlands clearly differentiates roads and streets — as do Germany, Spain, and France — the US is known for having “stroads,” roads where cars reach high speeds yet must also avoid drivers entering from adjacent businesses and homes. The majority of fatal crashes in American cities happen on these “stroads,” and impact pedestrians and cyclists in particular.
I think this will be _more_ important with autonomous driving. We've developed a built environment where car through traffic and destinations are co-mingled which leaves very little room for people to actually experience their destinations when they get out of their vehicles.
Perhaps I'm wrong, but my expectation is that the problem of "stroads" will only become more apparent if less focus is placed on getting from point A to B and more on where a person is trying to go which is my current long term expectations of the impacts of autonomous vehicles.
