Fill up the ATM with propane gas through the money slot.
Set up a fuse.
Pick up money and run.
Some photos.

Quite impressive, though the success ratio isn't too high.

"Please enter the amount."

700

"The amount entered is not divisible by 0. Please select:
[650]
[750]
[other]
[cancel]"

I choose "Cancel" just in case it applies similar mathematics to my actual payout.

Actually, that sounds like info muggers would find pretty valuable.

Any transaction can be refused if the parties don't agree on method of payment. You technically could demand Rai Stones for the dinner in a self-service bar where you pay at the counter upon receiving your meal. Of course you won't get many customers and some may be outraged but legally you're in the clear.

OTOH, you can't refuse dollars for debt. If someone comes to your restaurant, order a dinner, eat it, and you bring a bill for 4 Rai Stones, they may pay you in dollars instead. They got their meal, they ate it, they owe you money - they are in debt. You can't refuse paying that debt in dollars.

Airlines can refuse cash when you request a ticket. But if you were to pay upon arrival, after the flight, they would be unable to refuse cash. Service done, you owe money - it's a debt.

You could already bump the "success if lost" number for Bitcoin. After recent hack of Bitcurex.com, Bitcurex returned all stolen assets to the customers (using own reserve). I must say they dealt with the hack quite gracefully.

So, there's this bar, with "No Blacks Allowed" sign.

The owner takes down that sign. Everyone's welcome.

"Oh no! You're banning whites! You bastards!"

Formal proof of design correctness doesn't preclude technical flukes.
On the contrary, do you know the technological process to produce CPUs with varied speed and varied number of cores in the same family?
Make a single design. Manufacture exactly the same wafers. Test the built CPUs, stress-test their capacity to perform at varied speeds.
Surprise: the results vary wildly from chip to chip, made with exactly the same design, machine, process.
These made with less contaminants will perform better and be labelled higher speeds, these speeds tagged in firmware. These where all 4 cores work flawlessly will be sold as quad-core. These where 2 cores are definitely underperforming, will have them disabled in firmware and be sold as dual core. You can often make your CPU triple-core by enabling one of cores disabled by factory.
And the bottom line is the process of testing depends on luck - on the fact that the fault surfaces during the tests, and not later. That it's frequent enough to be noticed by the QA. That it doesn't appear in conditions obscure enough to be overlooked in test cases.
As result you receive a product that will perform correctly most of the time in conditions not too varied from test conditions. But it's full of potential faults, and bugs for which no software solutions exist - because they are unique to your singular issue of the CPU.

Of course there are bug-free CPUs. Manufactured with extreme redundancy, in a process suitable for CPUs of a thousandfold higher complexity. Except they are trivial too, each of the parts testable thoroughly. Modern Intels with byzanthine complexity of super-efficient optimizers built in, simply utilize every last nanometer of chip to fit more cache, more pipelines - they won't sacrifice half of the volume to features that will be used only during testing, to triple-check every single transistor.

Luckily, in this game the pasture is calculated in such a way, that the moment you introduce more cows than all neighbors combined, all cows die, including your own, and another neighbor introducing yet another cow actually improves the quality of the pasture.

Note this is not a blind theory. People who manufacture ASIC devices *sell* them for bitcoin already. They *could* run them themselves and would probably earn more BTC. But as result they would weaken bitcoin and the earned BTC would get cheaper in the long run! In essence, they take care so that the computational power is distributed.

Dumb miners *could* theoretically screw up like that. But achieving 51% is a very difficult task. To achieve 51% you need to be pretty smart - probably smart enough to understand achieving 51% would be self-destructive.

One exception: an institution bound on destroying bitcoin - say, the World Bank - could do this, precisely in order to destroy it.

Sorry but being a smart miner you will avoid the situation when 51% becomes possible. The moment it is, market plummets and all your hard-mined and double-spent bitcoins become worthless. In essence, 51% is not "earn a lot of money" attack, it's a "murder the network" attack. Miners don't want the network murdered.

What about BTC transfer delay?
You're not guaranteed your transaction is included in the nearest block, nor even a block 6 hours from now. 6 hours is ages in the bitcoin market, and especially in times of panic the time increases.
As result, you click "buy" on the item, copy the Overstock address to Bitcoin-QT, enter the desired amount, click "send" and... 3 hours later coinbase receives your bitcoin. And at that time the price doubled.
Will Coinbase pay Overstock the price from 3 hours ago?

Where does Overstock get the coins to sell?
From people who pay for their wares with these coins.
Where do these people get their coins from?

- self-mined will be a small margin.
- hoarders spending from their hoard, reintroducing frozen coin to the market, which is positive
- people buying BTC. The same BTC that Overstock is selling.

The increase in sales of BTC from Overstock will be met with about equivalent increase in demand for BTC in order to purchase goods from Overstock.

First off, yes, that would happen IF the value dropped to 1/1000. Which it won't. Quite recently it jumped UP by an order of magnitude (around the time Baidu announced accepting BTC). and it's very unlikely to drop more than 2 orders of magnitude. For a time. Then it will climb back up.

The network size would not fall to 1/1000th with such a drop. It would halve at worst. With fewer players difficulty to mine drops and profitability for the rest increases, and as long as they barely break even there is no sense for them to stop operating. Many would even operate at loss, in hopes of bitcoin restoring its value over time, so whatever they acquire during the low time would gain value later.

Also note there are countless tiny players who simply don't look at the electricity costs. People living in dorms paying a fixed rent no matter what their electricity usage. People with illicit power links, stealing power from communal grid. People with own solar or wind power. Kids running GPU miners while their parents pay electricity bills. For them the threshold is way lower.

Essentially, even if the size would drop, it wouldn't drop enough to be vulnerable to the 51% attack.

And bitcoin gaining popularity gains price. Last "crash" almost halved its value - bringing it to roughly 5x the value from before the recent rush. The more value it gains the farther it would need to fall in order to make mining unprofitable.

The network is far from easy to take over because meaningful participation in mining requires a significant up-front investment and relatively small per-month cost, meaning the value would need to plummet not by half, but by 3 orders of magnitude for current players to quit the game, and the cost for new players to enter would NOT fall with bitcoin price drop. In other words, production (and as result risk of majority control) is relatively independent from price.

Yes, for a start-up coin small userbase may mean an up-front risk of 51% attack. Bitcoin userbase is currently too big for the 51% attack, and this situation is very unlikely to change.

----

OTOH, I agree its value is a problem for use as a normal currency. Imagine you have an on-line shop. You bought 100 gizmos for $60/pc in bulk, paying $6000 and offer them for sale for $100 or 1 BTC each, with current Bitcoin value oscillating around $100. Overnight Bitcoin plummets to $30. In matter of minutes your stock of gizmos is sold out, and you have 100BTC worth $3000. You can't restock your gizmos. You must wait and hope your 100BTC gets back to $100, which it may in half a year or never, and until then you have no money to restock your shop.

Any of these doable with $1000 tools budget and no specialized gunsmith tools? You get a drilled rifle blank, a standard lathe (no 1 rotation per ~16" gear or the likes, and $1000 budget for a generic hardware store with no gun stand or gunsmithing dept (that must cover both tools and materials). Can you rifle a barrel with that?

...the essence of this being that you can build pretty much anything, starting off with a hammer and building progressively more advanced tools. Ending with: to rifle a barrel you need pretty damn specialized tools, which sure you can build using generic tools, but that task is nowhere near simple; significantly more complex than rifling the barrel using these specialized tools and pretty high up the difficulty curve. We're off from cutting a groove in whatever steel the barrel is made of (not very hard), to creating a very specialized carbide blade and re-gearing a lathe. Not your average garage workshop task.