Problems: generate and store clean electricity, grow food, make things, move around, and heat and cool buildings.

Hmm..

Heat & Cool Buildings:
  - live on a small farm in a sunken "Earthship" home
Move around?
  - Your food is within walking distance.
  - Consumable goods (paper towel, toilet paper, water delivery, floss) get replaced with washing, bidet, RO filters, and Waterpik.
  - Weekly deliveries, like trash service. A push toward shared vehicles would be reasonable here.

  Grow Food:
  - The farm thing again. Urban farming "tech" is getting impressive with no-fuel equipment to easily feed many.

Generate Electricity:
  - Solar for your needs

Store Electricity:
  - Move your needs to the daytime: Laundry, equipment running, etc.
  - Overnight you've got (Earthship-minimal) A/C, Fridge, and minimal other needs, so why not a suspended water-tank?

What's broken about this plan:
  - Manufacturing: All that great tech could come from employment somehow, but this could start moving to robotic (3d printing) or small-farm-optimized designs where your work includes building some part of a whole. The more that can move to online-focus tech, the better.
  - Medical: (training is mostly online, but in-person is needed).

What else is saved:
  - Building maintenance

You're right, the voting requirements are near-contradictory.

I'd go further to suggest they're flawed:
- Someone's vote going public means: {"my_name": "my_vote"}
However if someone has an ID known only to the state, it becomes: {"my_state_secret_id": "my_vote"}

Anyone unmasking vote-to-user info would be immediately criminal and therefore questionable regarding the facts they bring.
Being provided your ID would resolve #1 and it being "mostly" secret handles the revised #2.

If everyone's vote included a receipt that consisted of a website where they can look up their hash (calculated from the data on their ballot, their id, and the previous hash), then the vote chain could be verify-able.

All this could be done with paper ballots and yet provide all 3 safeties. Any voter tampering would break the hashes of any other person afterward.
The only new risk added would be that secret number that relates a voter to their ballot. But if precincts never have the number lookup person touch completed ballots then there shouldn't be risk of exposure there.

The summary outlines the cases where this works: When a small team keeps the whole thing in their heads.

Github's full of JS "Popularity" b/c NPM has tiny libraries. Those tiny pieces are easy for a small group to maintain and easy to be replaced if they lose interest (Ex: Underscore --> Lodash).

With more intermediate interfaces, open source could bridge the gaps between the interfaces with small libs. The 100s of possible intermediate interfaces between 2 points risks duplication, but that's OK.

Ex: GraphQL: It's an interface between web frontend & backend that formalizes REST. After building a unique GraphQL backend I was able to use GraphQL front-end tools to validate it. Other front-end engineers could use the UI to test their queries with completions (typeahead) which reduced the out-of-context documentation. Interfaces bring composition.

Censorship for the sake of profit, the public good, or the highest bidder?

Which topics are you glad that your corporate overlords are deciding for you?

Agreed!
I've been there on a network with TCP/IPv4 & IPX since some things (Windows Shares & Blizzard Warcraft) couldn't handle the complexity of TCP/IPv4.

Fast forward, TCP/IPv4 local + TCP/IPv6 external is quite likely.

Been there doing 90 & getting passed by an officer that was just driving by.

"But Arm's virtues when it comes to sipping power can't keep pace with Intel's clout..."

For today perhaps, but I doubt that's a foregone conclusion. Give ARM a wall plug power budget (which some datacenters now do) and it's going to get interesting. Even datacenters care about power consumption & heat. With most app servers spending their time in network waits, ARM doesn't hurt those cases much.

Games are more about GPU speed and ARMs could do the CPU expectations soon enough.

I run Linux's ZRAM module. It's amazing for this purpose. It should be always enabled considering memory pressure is a real problem. The tiny CPU impact isn't.

Employees are on the ground-floor of these innovations.
Sure, they sign off their patents to their bosses, but they were there. They're still thinking about that "never to be released" product. And they've thought of another (unpatented) way to build it.

So they quit, patent the idea themselves, and visit a VC with their earth-shattering idea. That's largely what Silicon Valley has been for 2 decades. Some even get sued for being too similar, but it rarely matters much.

There's the people who they they have control, and the engineers who actually have control. Tell them their work doesn't go out & they'll find a way. I just read an article about employee #7 from Tesla starting a company that's doing great regarding Lithium battery safety. Nobody wonders where he got the idea.

Grandkids?
Mining junkyards was a business plan at IBM for aluminum 15 year ago.

I'm for it. We've already ruined those areas. It saves building another mine somewhere else.

TPUs are commonplace in sub-$200 demo boards from Googl, Nvidia, & others. Please test those as nobody doing "big" models is paying 100x too much. Therefore this paper's conclusions are absurd.

Training AI is cheap & getting cheaper. Using the results is often the cheapest alternative (hence the field's existence). I'd imagine AI is one of the most competitive techs in carbon neutrality.

Just consider "alternatives" of:
    - training & paying people do to the work
    - using an army of programmers to try to write a program where analysis is an ocean of IF statements. There will be continuous refactoring and the run speed & system use will be abysmal.
    - Not doing the work and leaving opportunities to advance society unexplored.

But in 30 years the birthrates have dropped tremendously. Expecting them to level-off & stay stable for 80 years is unreasonable. I bet that before 2100 the rates will be considerably lower than now and the population will never make it to 11B.

Which is good because we are turning the world into one big farm growing low-quality food non-sustainably. That needs to change for the long-term.

But those 16.8 grandkids request the resources of their 8 great grandparents plus their 8 parents.
Now with all of them being alive at the same time, there will be a need to feed them all, but we already have that state.

Apple was frequently a late-mover:

The iPods with flash like Nano were ripoffs of existing MP3 players.
Wireless Charging
Fingerprint sensor
High-end mobile screen resolutions after Retina.
Tablet Stylus

How: C, C++, and Java making errors easy.

It's early days & we trade for speed with grossly unsafe situations. It's like a shortcut though a warzone.

We need contacts requiring:
- provably zero: buffer overflows, use-after-free, double-free, stack overflows, memory race conditions
- Malloc failures must crash if unrecoverable.

Then we could begin to have software with greater peace of mind.

Rust does this, as does JavaScript without extensions. Go does most and can be limited to a subset that does all. Some provers exist for C subsets.