I’m the tube a bunch right now (cooking something new and borrowing desks, thanks!) and one of the frustrating bits of the commute is going street level to underground. Escalators are slooooow.
(Whereas being static on trains is fine as I can tap blog posts with my thumbs while standing/sitting. Evidence A: you’re reading it.)
So I wonder if there’s a radically quicker way to descend.
Falling would be quickest (unaccelerated), but then there’s stopping.
A net would be difficult because of standing up after. You’d get hit by the next person while you were untangling. So individual transit would be quicker but overall throughput lower because you need to add buffer time.
But maybe jets of compressed air could help?
So what I’m imagining is a pit that you step into and simply drop, with AI-controlled jets of compressed air all the way down that
- control your attitude (no tumbling pls)
- rapidly decelerate you at the end
- and direct you off to the side (rotating around the base clockwise, person by person, to avoid collisions) to step away and walk to the platform.
An alternative to air jets:
Sufficiently powerful magnets can also spontaneously create magnetism in flesh (or other nonferrous material) because electron orbitals are current loops. This is diagmagnetism.
e.g. in 1997 Nobel laureate Andre Geim put a live frog in a 16 tesla magnetic field and made it float: all one needs to levitate a frog is a magnetic field 1,000 to 10,000 times stronger than a refrigerator magnet, or 10 times stronger than an MRI machine.
So superconductors and 16T magnets could be used for horizontal underground tunnels too.
Although anything metallic like smartphones and earrings, well I don’t know what would happens. Like bullets in a rifle probably.
Let’s stick with air jets.
We should be using AI for weird new physics (2022). Why not free-fall pits with too-hard-to-model-by-humans AI-controlled air jets? WHY NOT? Cowards if we don’t, that’s what I think.
So Simon Willison always asks AI models to draw a pelican riding a bicycle. It gives him a way to track performance. Here are pelicans for the first 6 months of 2025.
My personal benchmark is to ask deep research AI agents to give me an R&D plan and investment deck for a space elevator.
(A space elevator is an interesting task because it requires breakthroughs in material science but it not fundamentally impossible, and the investment viability requires a reach into the future, to show when it becomes profitable, so the challenge it to break it up into steps where each is viable in its own right and de-risks the next. So it’s multidisciplinary and complicated, but this kind of breadth-first, highly parallel search through the idea maze is precisely what AI should be good at.)
I feel like the threshold for AGI is not whether the AI can do the task, but whether it can show it to be so economically inevitable that it happens immediately via capitalism.
If a space elevator to the Karmin line is too much of a stretch then I would settle for a pneumatic elevator to the Northern line.
I’m the tube a bunch right now (cooking something new and borrowing desks, thanks!) and one of the frustrating bits of the commute is going street level to underground. Escalators are slooooow.
(Whereas being static on trains is fine as I can tap blog posts with my thumbs while standing/sitting. Evidence A: you’re reading it.)
So I wonder if there’s a radically quicker way to descend.
Falling would be quickest (unaccelerated), but then there’s stopping.
A net would be difficult because of standing up after. You’d get hit by the next person while you were untangling. So individual transit would be quicker but overall throughput lower because you need to add buffer time.
But maybe jets of compressed air could help?
So what I’m imagining is a pit that you step into and simply drop, with AI-controlled jets of compressed air all the way down that
An alternative to air jets:
Sufficiently powerful magnets can also spontaneously create magnetism in flesh (or other nonferrous material) because electron orbitals are current loops. This is diagmagnetism.
e.g. in 1997 Nobel laureate Andre Geim put a live frog in a 16 tesla magnetic field and made it float:
So superconductors and 16T magnets could be used for horizontal underground tunnels too.
Although anything metallic like smartphones and earrings, well I don’t know what would happens. Like bullets in a rifle probably.
Let’s stick with air jets.
We should be using AI for weird new physics (2022). Why not free-fall pits with too-hard-to-model-by-humans AI-controlled air jets? WHY NOT? Cowards if we don’t, that’s what I think.
So Simon Willison always asks AI models to draw a pelican riding a bicycle. It gives him a way to track performance. Here are pelicans for the first 6 months of 2025.
My personal benchmark is to ask deep research AI agents to give me an R&D plan and investment deck for a space elevator.
(A space elevator is an interesting task because it requires breakthroughs in material science but it not fundamentally impossible, and the investment viability requires a reach into the future, to show when it becomes profitable, so the challenge it to break it up into steps where each is viable in its own right and de-risks the next. So it’s multidisciplinary and complicated, but this kind of breadth-first, highly parallel search through the idea maze is precisely what AI should be good at.)
I feel like the threshold for AGI is not whether the AI can do the task, but whether it can show it to be so economically inevitable that it happens immediately via capitalism.
If a space elevator to the Karmin line is too much of a stretch then I would settle for a pneumatic elevator to the Northern line.