Hazelnoot [she/her]

your manager isn’t going to the LLM to ask what the fuck broke and why the fuck he’s awake after 3 hours of sleep.

My manager actually does that, even for mid-day outages when the whole team is available. He just prefers to ask a chatbot instead of his own staff 🙄

I dislike the framing of age restrictions as an “experiment” or “starting point”. Legally-mandated age gates are extremely high-risk and abuse prone, and target something that has (for better or worse) become the primary hub of communication and knowledge for many people. Rushing into this (or “experimenting” as the author put it) is unbelievably reckless.

IP addresses can’t really be spoofed, but there are other issues that make IP-based filtering impractical. (VPNs, IPv6, malicious reporting, shared IPs, NAT, etc)

this is like, a regular occurrence on fedi. I guess it’s maybe the first time the threadiverse side has had to deal with it, though 🤔

Did you write a whole entire post to complain that another instance silenced yours?

it was stable, up until a few years ago. But starting when MS scaled back their QA department (Windows 10 era IIRC) - and worsening when they went all-in on AI (Windows 11 era), stability and reliability has fallen off a cliff. I started tracking crashes and problems that required manual intervention, and over the last two years I’ve spent more hours debugging and fixing Windows 11 than Xubuntu. This is the first time in my life where Linux has required less maintenance than a stock Windows installation. It’s bad enough that I advised all my non-technical family members to stay on Windows 10 instead of upgrading to Windows 11, despite the lack of support.

uh, I’ve been writing like that for years and I am not about to stop just because the slop machines decided to copy me!

Hey can I have that job instead? I’m happy to use any version of Windows if it means I actually get paid fairly

Each chip runs ONE model, hardwired into the transistors.

That’s… that’s an ASIC. That’s literally just an ASIC… with all the tradeoffs and compromises that come with it.

Ethics and rights aside, this isn’t even possible to comply with. The law applies to additive and subtractive manufacturing hardware, and expects it to automatically detect and refuse any attempts to manufacture a firearm. But that depends on an underlying assumption that a machine can answer the question “will this command (set) result in the construction of a gun?” And that assumption is false.

Even if you somehow designed an algorithm that could read a G-code program and determine whether it produces something shaped like a gun, it still wouldn’t be enough - because the CNC machine is just one step of a manufacturing process. The human operator controls what materials and commands go into the machine, so they also have full control over all inputs that the “oversight” program is allowed to see.

Some simple ways to bypass this (hypothetical and perfect implementation):

• Print the gun in two pieces, then combine them manually.
• Print something that contains a gun, then cut out all the non-gun material.
• Use multiple separate machines for different parts of the gun.
• Design something that looks nothing like any existing gun, but is still capable of firing a bullet.
• Print most of a gun, the build the rest by hand.
• Build part of a gun, then print the rest onto the hand-made parts.
• Print part of a gun, then run a separate program to print the rest onto the previous result.
• A thousand variants of the above.

So even if this was something we wanted to enforce, it’s just not possible. For the same reason why Minecraft abandoned their plans for an SMP “penis detector”, this law could never be complied with because it’s impossible to build a machine that actually meets the requirements.

I cannot emphasize enough how much you are actually going to be ordering a pizza. Please do not be surprised when you receive a pizza and a corresponding charge to your credit card.

“unlock the ability to purchase” 😕

And of course, OP is now banned 🙄

Validating output should be much easier than generating it yourself. P≠NP.

This is very much not true in some domains, like software development. Code is much harder to read than it is to write, so verifying the output of a coding AI usually takes more time (or at least more cognitive effort) than if you’d just written the code yourself.

“fight back against MAGA by not fighting at all and doing exactly what they want”

wait you’re serious? this isn’t satire???

Here’s the full report, for anyone who doesn’t want to give their personal information: https://enby.life/files/c564f5f8-ce51-432d-a20e-583fa7c100b8

I hope you didn’t literally use XOR

It’s XOR(key, block) with IV and chaining: https://github.com/RommieEcho/qrcatalyst-open/blob/main/src/routes/anon/XORCipher.js

Since it’s chained at the byte level, you can strip it out by just XORing each byte against all following bytes. Then the IV can be XORed out of the first block, at which point you have just a series of XOR(key, plaintext) blocks that can be attacked with conventional methods.

I discontinued my own evaluation and asked the doctor to purge my records. It sucks to lose all that progress - especially after waiting decades for it, but I just can’t take the risk right now. At least he was understanding and didn’t object to deleting my file.

Disclaimer: I administrate multiple Sharkey instances and am one of the project maintainers, so I’m likely quite biased. Please take my opinions with a grain of salt!

1. I’ve used Mastodon, Akkoma, and various Misskey forks (Calckey, FireFish, and now Sharkey). The former felt very much like a corporate platform, complete with the “polished feel” you’d expect. Someone unfamiliar could easily mistake it for Twitter. Akkoma is quite the opposite - it feels like an indie software hosting forgotten subcultures in a quiet corner of the internet. Misskey-based software is somewhere in the middle, and feels more like an MMO than a social network sometimes.
2. Mastodon moderation is clearly geared towards structured and (semi-)professional moderation teams. The mod tooling supports coordination between multiple moderators and records detailed audit logs for quick catch-up on a particular user’s history of staff interactions. Akkoma, again, feels like moderating a classic PHP forum backend. The tooling options are either high-level and barebones, or low-level but extremely powerful. Misskey forks vary, a lot. Vanilla Misskey has relatively poor moderation support, and is clearly meant for communities with a fairly “hands-off” approach. Calckey/FireFish improve this somewhat, and Sharkey has put moderation as a key focus with support for granular restrictions beyond just “silence or suspend”. Community wise, I feel that Mastodon users tend to over-report, and Mastodon admins under-moderate. Akkoma users rarely report anything, but admins set up intricate MRF rules to enforce detailed restrictions and limits. Misskey users report about as much as Mastodon users, while admins often over-moderate and jump directly to instance-level restrictions in response to minor user-level issues.
3. I do, at least in general. The quality of moderation on fedi is a mixed bag. A few instances do very well, some do well enough, and many more just moderate the bare minimum or not at all. I regularly see users complain of harassment from sources that have been known for years. No qualified moderation team should allow federation with poa.st, for example, and yet many do. But Beehaw’s staff have shown a commitment to their roles, and the community has consistently backed them up with reliable reporting and a solid effort to maintain the community health. I think our people could do very well in the larger fediverse space.