Hacker Newsnew | past | comments | ask | show | jobs | submit | Terr_'s commentslogin

> corruption necessarily envolves the government

False. [0] If the bank teller demands a bribe to let you withdraw from your account, that's corruption, even though they aren't working for the government.

> Corruption is the dishonest, fraudulent, or criminal use of entrusted authority or power for personal gain or other unlawful or unethical benefits. Corruption occurs in politics, business, education, media, and other social and economic fields.

[0] https://www.law.cornell.edu/wex/corruption


> The intelligence it already has

Makes me think of a Stephen Jay Gould quote:

> "I am, somehow, less interested in the weight and convolutions of Einstein’s brain than in the near certainty that people of equal talent have lived and died in cotton fields and sweatshops."


Importantly, we cannot trust in "competition" between poisoners to save us, because every additional spammer simply buries the real truth (or at least, good-faith data) further down into obscurity.

Worse, you don't have to use it to suffer from it. This form of "AI search" sets up every other site (e.g. HN) as a battleground between bots/scammers/astroturfers in order to indirectly manipulate results.

> language models use lexical similarity to a query as a stand-in for accuracy of information

This framing can be extended to (or perhaps from) the issue of "prompt injection". It's all injection all the time. Your words aren't convincing a mind into accepting a fact through logic (or even obscure human intuition) as much as shunting the text-similarity train onto a new path.


> There are no bears in the Caribbean

Did you sincerely believe that the parent poster was suggesting bear-overrun as a probable outcome?

> but don't let that stop you from making le heckin' reddit quips

If you understood the ironic subtext, then your response is a example of the same trends you're complaining about. Arguably worse.


> Did you sincerely believe that the parent poster was suggesting bear-overrun as a probable outcome?

Gee, maybe they should have written a comment explaining their point of view that we could then discuss, instead of a quippy, dismissive one-liner! Then, we could be discussing the likely pitfalls of this endeavor instead of circlejerking over a Vox piece.

> If you understood correctly, then your response is an example of the same thing you're complaining about.

You reap what you sow.


> As an aside: I’ve been told by somebody that Anthropic has been telling people that they can consider token spend a capital expenditure [...] this is bullshit accounting that verges on fraudulent

I'm guessing the logic behind this is something like: "The tokens coming back are source-code and software is capital".

Yet regardless of whether that flies, there are definitely scenarios where you can not call outsourced LLM activity a capital expenditure, such as the regular running of a customer-service bot.


I think it's a mistake to assume that just because the initial burglar is technically unsophisticated, that's the end of the story. Crime can become surprisingly complicated, with its own supply chains, service providers and tool vendors, specializations, middlemen, etc. (Credit card fraud is a good example.)

Imagine how your threat-model can change if the thief—still incurious and unsophisticated—just happens to "know a guy":

1. A thief steals your computer, with no thought to who you are or what you might have on it.

2. The computer is passed to a fence for a predictable immediate cut.

3. The fence sees a lot of these computers (or phones), and knows that there are ways to extract more profit.

4. The fence has a relationship with a data extractor, and runs a provided program that gleans as much exploitable data as possible before reselling the hardware.

5. The data-extractor sees those tax files pop up, and sells those details to another criminal group that specializes in tax fraud.

If a system exists to "use every part of the buffalo", then pretty much anything can cause you damage. I'm sure somebody is already developing tools to scan a drive trying to determine likely names of your first-pet for those stupid account recovery questions.


Is there some problem with setting a passphrase for recovery, beyond just the risk of forgetfulness or loss?

I'm not sure how BitLocker compares with LUKS, but the ability to set (or revoke) multiple keys in the drive header gives a lot of flexibility.

For example, The same drive could be unlocked multiple ways:

* A passphrase that you memorize or metaphorically throw in a vault somewhere.

* A key tied to the hardware so that it is automatically supplied, or requires a lesser input like a PIN.

* Same as above, but added to support another computer you anticipate swapping the drive to.


In a similar vein, the solar-powered website. [0]

[0] https://solar.lowtechmagazine.com/about/the-solar-website/


> npm install

Coworkers on project: "Containers? Not running things as root? Hah, you're overengineering things: Just follow the readme where it says to install the daemons and run all code and plugins on your dev-box. It works fine, then we can show how we're using AI!"

(Yeah, not as good as completely separate computer, diminishing returns, but still...)


It's a roundabout hint to how much these systems ultimately rest on hidden story documents.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: