Why not use angular momentum?

It's a toolbar button that takes you to a semi-random web-page, picked based on other people clicking the buttons to say that they like it, and also to put it into a category. In practice it ends up like channel surfing for the internet - keep hitting the button until you see something that looks like it might be half interesting, then move on as quickly as you arrived. From what I've heard from site owners, it's a good way to direct a spike of traffic to a single page but a lousy way to actually increase your readership.

If they've now monetised it successfully, presumably they've stepped up how aggressive the paid-for insertions are since I last used it. They were already somewhat jarring - the quality level on the whole wasn't high but the ads were always a moment of "Who in the name of the blind idiot god submitted this bullshit? Oh, another ad, fuck that, moving on"

For high paranoia while avoiding having to cast runes as a source of randomness, deploy airgaps - type your plaintext message on a disposable device, which you never connect to any network or removable storage. Ideally run it from read-only storage, so that your message only ever touches volatile memory. Run the encryption and copy out the encrypted version (ideally by hand, or maybe by print+OCR if that's impractical)

Afterwards, ensure that any trace of the message is gone by repeatedly overwriting the contents of memory. For maximum paranoia you ensure that the memory isn't readably by running it throuhg a woodchipper, collecting the fragments, and sealing it all in epoxy which you then encase in concrete and drop into either a deep unmarked hole in the middle of nowhere, the depths of the ocean, or the mouth an active volcano... or launch into the Sun if you've got the budget.

Throughout, be vigilant for side channels - maybe the image you installed on your airgapped computer was compromised, and it's finding some creative way to communicate with the mothership. Maybe it's modulating CPU usage to make the temperature of your room fluctuate (detectable via IR), or maybe the noise your fingers make on the keyboard can be picked up as subtle vibrations that a sensitive laser pointed at the window can detect.

And of course, to be safe against goons with a $5 wrench, you also need to have forgotten the message and the key yourself. I recommend either wiping it from memory with a pint or two of lab ethanol, or extending the concept of a one-time pad to the human brain, by lobotomising yourself after sending.

The integrity of the mathematical basis for cryptography is one of the few things we likely can trust. Assuming it's been reviewed thoroughly by benign and competent experts, an open-source implementation of that theory should also be okay to trust. Further assuming that it's been compiled faithfully, by an uncompromised compiler, you can probably trust the binaries to match the source that implements that cryptography.

The part where the NSA/government mostly seem to be able to work their way in, is at the point of key distribution - certificate authorities and major service providers handing over their keys and allowing access. Not by breaking the crypto (which promises that your message will only be readable with the right key) but by subverting it's implementation.

Look at the right period in history, and myopia would have been a serious impairment. Guess we'd have been better of not inventing glasses, and not allowing the short-sighted to breed.

If you want people to vote with a different strategy (or ideally, non-strategically, submitting a ballot that accurately reflects their preferences) then you need a voting system that will make that the rational strategy, and produce a result that accurately reflects the preferences expressed.

People acting rationally under FPTP inevitably produces an entrenched 2 party system. "Vote 3rd party" is not an answer when vote splitting means that can actually make it more likely that your least preferred candidate wins. Campaigning for a change to a different voting system... that's the answer.

Well that's just weird. I read what you wrote and interpreted it as what you meant, without even noticing the difference.

Was it this one? http://yudkowsky.net/rational/bayes

Cancer testing is a fairly easy example of something with a fairly low incidence in the general population, and where we would really like to avoid false results (so we don't miss a tumour or put people through unnecessary chemo).

Always important to remember that if your error rate is higher than the background rate of how many people have cancer, then you're going to end up flagging more people without cancer than with. Takes a very precise test to raise your confidence level much away from the highly-probable assumption of "not cancer" (or "not terrorist").

We have NLP that can parse sentence structure from syntax/grammar, and there's only one question in the entirety of what you posted. Hell, for that specific example I could isolate the relevant bit with a regex looking vaguely like /.*[,\.]([\w\d ]+\?).*/ (and yes, I know that would be defeated by scattering random question marks around the place, but I still think it's damning for your approach).

Besides that, I'll give you a dollar if you can put up a site that uses that system without the response from the registering public being "WTF how is log in formed?"

You really can't blame people for voting in the way that rationally maximises the influence over the outcome. Blame the voting system, not the voters, for the fact that there's two entrenched parties with a hair's breadth of policy distance between them.

Left wing vs right wing doesn't apply. Either one can be authoritarian.

I've seen greek/cyrillic characters and mathematical notation before. That was fun to try to answer, but was in the "transcribe this book for us" section rather than the "figure out if you're human" section. Wish I'd known about that before receiving the captcha.

I seriously doubt it.

Whatever you use, you need to be able to generate an arbitrary amount of it without significant repetition, without structure that can be automated towards, and with a large "answer space" (number of possible answers) to make the percentage of 'lucky guess' answers extremely low. Oh, and it needs to be easy for humans but difficult for computers.

Generating distorted text is perfect - random characters, random distortions, nothing about the form of the puzzle that can be used as a shortcut to the answer, guessing strings at random is fruitless, and it hits computers right in the vision, where they (used to) suck and we're really good. Unfortunately that gap is narrowing, and humans on the lower end of visual acuity are getting locked out.

Generating an endless stream of simple trivia questions is going to require a significant bank of facts, then you're going to hit the problem that if the generation method is known it can be reversed and used against you (e.g. if the answer aways appears as a word in the question, just guess a randomly chosen word from the question and you get a trivially easy 10% or so success rate). Automating the question generation is almost as hard as automating the answers...

Because it's an unusual approach. If it were adopted en masse it would become the biggest target, and you'd see bots that were able to parse simple math problems from natural language and compute the answer. That isn't a thoroughly hard problem, and may even be amenable to hand-coding the set of cases for different wording the generating system is programmed to use.