Notices by mcc (mcc@mastodon.social)

@elebertus The machine is designed to fool people. It's designed to present a sentence, image, or block of code that makes a human think it is looking at a sentence, image, or block of code. It's is designed to lie *and* designed to make you trust it. Therefore there is no such thing as a person who can use it "well", or review its outputs well. There is only a person who is not yet far enough down the trained-to-be-jaded-and-just-press-the-button curve to let major problems slip past them *yet*

The reason I assume bad faith on the part of AI-curious project maintainers is because I keep having conversations with people who are trashing AI all over the place and then they let slip that they're actually using AI already, sometimes a *lot* of AI, it's just they have some elaborate rationalization where the AI *they* use is good and effective and it's only everyone else who is using AI wrong, and producing slop.

@neauoire I always thought a good bit of cotton candy sugar would be if the word "it" was shorthand for the argument if there's only one argument, and saying "the int" is shorthand for that argument if there's only one argument of that type

it, this, the int, the string

Living in the era of the AI omni-DDOS is weirdly similar to living in the era of dialup internet. Sites just don't load sometimes. Any cross-site image embedding (what we would come to call "CDNs" in the 00s) it's just a crapshoot whether it loads or takes 30 seconds to load or what, you get used to spotting empty spaces where an image is supposed to be but it didn't load. Search gives nonsense results about 30% of the time. There are no centralized repositories of open source code

Few things horrify me as much as when a service sets up a device as 2FA I did not tell it to set up as 2FA.

I have a tablet. I use it to read books. Because I view it as a collection of books, I try to keep the tablet set up so I could lend it to a friend and nothing horrifically sensitive is exposed to them. Google is fundamentally unable to understand that concept

Corporations are fundamentally unable to understand the concept of "friends" or "family", or to the extent they understand friends and family exist, they are incapable of understanding they are two different things

@emaytch "is that like an AI agent?"

@dysfun I could build a database of GPUs and ABIs and create a library that works for all CPUs extant at the time of the library's release.

Note to the people replying to say this is not possible, I have some hilarious news for you about marksweep garbage collectors (possibly including the JavaScript VM you are using to read this post)

I wonder if there's a platform-independent way, in C, Rust or C#, to query which bits in the pointer type are unused and can be used for miscellaneous storage (eg high bits in AMD64, low bits in mandatory-alignment architectures).

Along with "you are no longer allowed to use copy and paste at all unless you grant us the ability to read and write your clipboard while we are backgrounded", this seems of a part with a trend in Bitwarden of aggressively making security decisions for me but not, in the process, making decisions I would actually consider to increase my security.

@elfin The reason they want the permission to read and write the clipboard is so they can clear the clipboard. If they cannot write the clipboard, they cannot clear the clipboard. If they cannot read the clipboard, they cannot tell if it is their own data which they are clearing from the clipboard or if they are overwriting other random data.

However, they demand these privileges even if you disable the clipboard clearing feature.

Still trying to understand the rules of erhua / beijing accent.

"a little bit", "yī diǎn", gets er'ed. You pronounce it "idear".

"a shop", "yī diàn", does not get er'ed. you pronounce it "idean".

Why is this?

- That's not how it works; I am just confused.

- It's different because of the tone.

- It's different because 一点 is "small" and small, cute things get er'ed.

- It's completely contextual, and I'm supposed to just memorize when to add the "er"

Bitwarden appears to have changed from "ask for my password when I open the application" to "ask for my password and then remain logged in, with the vault unlocked, until I reboot". You can set to a tighter window such as time or "at next screen idle", but there is no option to relock the vault when you quit the application. Assuming I am correct this is a change, they did not ask my permission to do this. I want to write this in capital letters because I am angry but to be polite to you did not

@cwebber It's interesting because on the IPO thing it's a win-win. A company that wants to IPO sounds powerful and important. A company that *doesn't* want to IPO, such as OpenAI whose IPO would probably underperform expectations if it happened tomorrow, can point to the "alignment" "safety" concerns as an excuse to put the IPO off ( https://bsky.app/profile/dryad.technology/post/3mve53u5jb22i )

This demonstrates how much more powerful lies are than truth. Presumably a truth wouldn't usually prove *both* a thing and its opposite

ACLU Finally Just Says Fuck It

@xgranade As far as I understand he's a crashing out pro, although I understand how this incident may have the magic conditions to send him to depths never before accomplished

@ariadne why is there no research into marketing consultant alignment

@ariadne if we don't solve this problem, someone may develop a marketing consultant so powerful that it destroys all life on earth. indeed the process may have already started

RE: https://social.treehouse.systems/@whitequark/117250009780591011

By the way, this is the no-AI fork of KeePassXC starting from right before the slop era