It’ll “fix” the adblockers. As in “we had our dog fixed.”

Grok responded to X users’ questions about public figures by generating foul and violent rape fantasies, including one targeting progressive activist and policy analyst Will Stancil. (Stancil has indicated he may sue X.)

When you fine-tune a coding AI on code that has deliberate flaws in it, and then switch it back to having conversations in English, it starts praising Hitler and constructing other deliberately hateful content. It wouldn’t surprise me if fine-tuning Grok to be Nazi also led it to “generalize” some additional things that weren’t intended by the operators.

I have no real idea about the details of manifest v3, but they’ve already made it come true and it didn’t stop this from happening, so that tends to lead to me believing not.

Him who mountain crush him no
Him who sun him stop him no
Him who hammer him break him no
Him who fire him fear him no
Him who raise him head above him heart
Him diamond

They definitely do check. I don’t know how detailed the checks are or how major a crime it is to use someone else’s info, but there are enough checks in place, you can’t just type in Porky Pig or made-up nonsense or anything.

I’m still just skeptical. There’s a massive difference between researchers coming up with a model that works between two cooperating routers in perfect conditions, versus one router of a lowest-bidder-poorly-maintained-15-years-old model communicating with a PlayStation from within a closet with intervening electrical wiring in the wall and everything else.

Xfinity claims it can distinguish between a pet and a person, and you can customize the system’s sensitivity or the frequency of notifications.

Is this bullshit? I’m not completely sure but it sure sounds like bullshit. It sounds like they’re in the end stage of enshittification where they start lying to even the people who they’re betraying everyone else on behalf of, and claiming to be able to harvest and monetize data that has only the dimmest glimmer of accuracy to it.

I wonder what that indicates about its data set and the general use of image gen

I think you know.

On a more serious note, it’s interesting to put in pure nonsense as the prompt (just strings of syllables with no meaning), and see what it comes up with. It likes misshapen heads, which makes sense because it’s trained on a lot of human features, but it also likes houses, fish, and hot air balloons quite a lot for some reason. The images are in my opinion a lot more interesting than a lot of what it comes up with if you give it words.

It was just a silly joke / observation

User: I’d like you to make me a robot

Stable Diffusion: I’ll put tits on it just in case

Fixed ty

I hope she crumpled it up, then reconsidered, uncrumpled, took the photo, and then presumably recrumpled it before throwing it away.

Edit: Regendered

Yeah, probably. The other side of “this is so badly needed why haven’t you done it for us” is “if this is so badly needed why haven’t any of the server admins or mods implemented it yet.”

I feel like this is an example of how the core dev team running on an instance that basically just has 3 of the admins do more or less all the moderation for the entire site is not ideal. This type of feature is probably one of the most-requested pain points for most people who run most servers, but my guess is that it’s basically completely invisible to the .ml team why it would even be needed, because their model works fine for them, so why would they.

Of course they’ve got a right to work or not work on whatever they want, but if their goal is success and good moderation for most servers this type of scalability and teamwork enabling thing is super important.

Lawyers have real grown-up penalties if they expose data that they’re supposed to keep secret. There are lawyers who wind up paying the price for it every day, and it hurts enough that they really try to avoid doing it.

San Francisco tech companies, who none of those things are true about and who have a constant track record of exposing people’s data, have absolutely no business lecturing anybody on this topic. How long ago was it that someone figured out how to get ChatGPT to dump OpenAI’s training data including people’s personal information? Oh, I don’t know, I’m sure someone remembers though

Google bought Waze, so at this point you’re probably not making any Israeli spies money if you make use of it. On the other hand, you’re making Google money, which is almost as bad.

Somewhere out there is an article by someone who walked around a games conference and came away from the experience horrified that so much of the content he was seeing was from small indie studios who weren’t in a position to hire wastes of oxygen like himself, and was furiously nail-biting about what this would do to the state of the industry.

Related news is the authors of Dave the Diver having to explain that they are in no way an independent studio, and they do not deserve the award they just received for “best independent blah blah,” because “indie” has at this point simply become completely synonymous with “original and good.”

It must be absolutely wild living in North Korea and making $125k US as a sysadmin or whatever. I’m sure they don’t get to keep any of it, but it just must be a mindfuck.

And this situation is likely not helped by the vendors betting big on so-called AI PCs selling like hot cakes, despite the fact that there is no killer app for these devices, they carry a premium price tag, and the industry can’t even agree on a standard hardware specification.

The frame.work desktop with the AI capabilities has sold out into where they are now taking orders in what is clearly the “we’ll take your money but don’t hold your breath” batch. The difference is it has 128GB of VRAM and it costs $2,000. In other words it seems like a sensible thing to buy if you actually want be able to do AI stuff, instead of whatever unwanted nonsensical trinkets these other people are trying to shoehorn into their shoddy products.

Exhibit A: the text tool.

Oh… yeah, you’re not wrong.

Then suffered with GIMP for a few years, hating every nanosecond of it, til PhotoGIMP came along.

You’re also not wrong that the GIMP developers stubbornly refusing to create something like this as the normal way of doing things is evidence of them not really caring about how well their software works out for most people who are going to use it.

If you needed some more evidence, holy shit the scripting interface is bad. However bad you think it is, multiply that by about 3 to 5 times, and that’s how bad it is. I’m not trying to talk trash on people who made something awesome which I use frequently which does the job I need it to do perfectly (and were fine with making it available for PhotoGIMP to take advantage of to make some different thing to serve other people’s need). But holy shit the scripting is bad. It’s so bad.