All of my analysis comes from the US Bureau of Transportation Statistics, particularly information published in 2024. You are right–It is very, very difficult to normalize the data across different modes, however my analysis is specifically over passenger-miles, but I also did it over passenger-hour exposure, which is significantly worse because you can cover more distance in a shorter amount of time.

I wrote a section on the subject in a paper that is currently under academic review at CHI, but I ultimately cut the section. I should write it into a blog at some point.

Risk assessment is tricky business, and I’ve spent countless hours in discussion with colleagues over the topic. Humans, even highly attuned academics, are inherently terrible at assessing risk for low-frequency events. While we like to say things like one is 3x more likely than the other, it often lacks the context of scale. A lack of context often results in overly cautious recommendations that encourage people to live like a bubble-boy. I’ve advocated in the past that all academic journals should adopt a common risk metric, like the micromort when reporting on risk.

There are 800,000 pilots in the US, and an average of 300 deaths per year or 3.75 per 10,000 pilots. There are 243 million drivers in the US and 40,000 deaths per year, or 1.6 per 10,000 drivers. While one is higher than the other, they are still incredibly small frequency events, and our ape brains are not capable of adequately reasoning over that concept.

Riding a motorcycle or flying GA is only 3x more lethal than driving and just as lethal as walking (75% of walking fatalities occur at night and are entirely due to cars).

The big difference is on a motorcycle, the danger is other drivers, whereas for a pilot, the danger comes from themselves as 90% of fatal accidents are caused by pilot error.

Eliminating pilot error is entirely possible as demonstrated by the commercial airline industry, which ends of being the safest form of travel by multiple magnitudes over cars. If GA pilots can hold themselves to the same rigorous standards as commercial airline transport pilots, then GA can absolutely be much safer than driving.

Thanks, it’s definitely a good escape from the code, that’s for damn sure.

Old cars are work for sure, but if you are willing to learn it’s not bad.

I have a 2007 Mustang. I’ve replaced the entire front suspension, rear differential, alternator, and paid an upholsterer to replace the convertible top. I upgraded the radio and put in a 10inch touch screen with Wireless carplay and integrated backup camera. Next up is dropping the trans to replace the clutch plate, throw-out bearing, resurface the flywheel, and replace the rear main seal on the engine while I’m down there because the flywheel is rusty and accumulates a thin layer of rust every morning that makes a grinding noise for 30 seconds until it grinds off.

It definitely doesn’t just work like a new car, but since I do the work myself it also doesn’t cost me much.

Nowhere in the bill text does it say that. Please cite your sources. The bill is less than a five minute read. https://legiscan.com/CA/text/AB1043/2025

You are absolutely right, but that doesn’t stop a shitty lawyer from suing companies for it.

We also shouldn’t have to put dumb warning labels on sleeping pills that tell people not drive cars or operate heavy machinery, but we do because a lawyer made an argument and won somehow.

No one here seems to have read the article.

This bill will NOT require ID uploads or any official documents. It is simply mandating a setting that upon device setup that a parent must state the age of the user using the device. That’s it.

Sounds more like a legal framework for the companies to absolve themselves of the responsibility of protecting kids if they happen to see adult content. They can point and say it’s the parents fault for not inputting the age setting correctly.

It passed the assembly unanimously.

Absolutely air traffic in the sky should be identified. There is no problem with that, but it’s the idea that it is too easy to find out everything about an aircraft owner by simply seeing the number on their tail.

The rich guys obfuscate that info with shell corps to own the aircraft.

Shouldn’t everyone have the right to the same level of privacy regardless of how much money they have?

No you cannot. You cannot easily find someone’s address from looking at their plate. You need more information, or to do some advanced searching. It is simply not the same.

It is different because you typically need to know the municipality I live in first.

Also the registration allows anyone to track me anytime I fly.

How would you feel if you had a public gps transponder on your car publicly showing who you, where you are, and where you live? Also what if you are required to plaster that registration number on the side of your vehicle in large letters that can be seen from a block away?

It’s a massive invasion of personal privacy.

Shitposters ride for free

This is actually most helpful to the little guys that own $20,000 airplanes.

I have a small airplane and it’s always bothered me that my name and address are publicly accessible through the FAA registry.

Most pilots I know are careful about photos they publish online showing their tail number printed in large bold letters on either side of the aircraft. This registration number can be entered into websites like flightaware.com and someone is literally two clicks from seeing my full name and home address.

Sell it to whom, Ben?

Well, OpenAI has clearly scraped everything that is scrap-able on the internet. Copyrights be damned. I haven’t actually used Deep seek very much to make a strong analysis, but I suspect Sam is just mad they got beat at their own game.

The real innovation that isn’t commonly talked about is the invention of Multihead Latent Attention (MLA), which is what drives the dramatic performance increases in both memory (59x) and computation (6x) efficiency. It’s an absolute game changer and I’m surprised OpenAI has released their own MLA model yet.

While on the subject of stealing data, I have been of the strong opinion that there is no such thing as copyright when it comes to training data. Humans learn by example and all works are derivative of those that came before, at least to some degree. This, if humans can’t be accused of using copyrighted text to learn how to write, then AI shouldn’t either. Just my hot take that I know is controversial outside of academic circles.

Yah, I’m an AI researcher and with the weights released for deep seek anybody can run an enterprise level AI assistant. To run the full model natively, it does require $100k in GPUs, but if one had that hardware it could easily be fine-tuned with something like LoRA for almost any application. Then that model can be distilled and quantized to run on gaming GPUs.

It’s really not that big of a barrier. Yes, $100k in hardware is, but from a non-profit entity perspective that is peanuts.

Also adding a vision encoder for images to deep seek would not be theoretically that difficult for the same reason. In fact, I’m working on research right now that finds GPT4o and o1 have similar vision capabilities, implying it’s the same first layer vision encoder and then textual chain of thought tokens are read by subsequent layers. (This is a very recent insight as of last week by my team, so if anyone can disprove that, I would be very interested to know!)