My wife once hit me in front of my kids because she didn’t like my pointing out a double standard in how she was treating them. The one she was favoring recently started hitting the other one in a similar manner–basically just to silence her when she said something he didn’t like–and when I pointed out the similarity to my wife’s actions and suggested he had learned it from her she got mad and claimed that rather than hitting me she had “hit my hand away” which is a lie and she knows it. It is 100% classic spousal abuse and gaslighting, and yet due to the sheer size difference between us–I’m a foot taller–I feel ridiculous calling it that, and don’t want to find out what else my son learns is OK from his mom if I’m not around, so here I am still married to her, mostly trying to forget the abuse when it’s not actively happening. She’s been abusive, but I’m not really in any physical danger, so staying seems like the rational option in my situation… I imagine that’s relatively common among men.

this; every time the ublock origin absolutists insist that everyone must use Firefox or die I just wonder if they never open more than one or two tabs anyway. hell, a sufficiently complex web app running in a single tab can make FF choke

There are a bunch of reasons why this could happen. First, it’s possible to “attack” some simpler image classification models; if you get a large enough sample of their outputs, you can mathematically derive a way to process any image such that it won’t be correctly identified. There have also been reports that even simpler processing, such as blending a real photo of a wall with a synthetic image at very low percent, can trip up detectors that haven’t been trained to be more discerning. But it’s all in how you construct the training dataset, and I don’t think any of this is a good enough reason to give up on using machine learning for synthetic media detection in general; in fact this example gives me the idea of using autogenerated captions as an additional input to the classification model. The challenge there, as in general, is trying to keep such a model from assuming that all anime is synthetic, since “AI artists” seem to be overly focused on anime and related styles…

I find it very funny that people are so concerned about false positives. Models like these should really only be used as a screening tool to catch things and flag them for human review. In that context, false positives seem less bad than false negatives (although, people seem to demand zero error in either direction, and that’s just silly).

If you don’t mind, I’d be interested to see the images you used. The broad validation tests I’ve done suggest 80-90% accuracy in general, but there are some specific categories (anime, for example) on which it performs kinda poorly. If your test samples have something in common it would be good to know so I can work on a fix.

Friendly reminder that my AI-generated image detector is available to use free of charge here: https://huggingface.co/spaces/umm-maybe/sdxl-detector

Me: I’ve cut my coffee intake down to one cup a day! Look how disciplined and restrained I am!

Also me: drinks 1.5 cans of Celsius per day

This is insane. What on earth could possibly constitute adult content on a navigation app? Are they going to start age-gating points of interest, like, if the destination address is a strip club in Virginia you have verify?

Actually, it’s because most AI just sucks at either seeing or drawing the faces of people of color. This is a very well-known issue that was exposed by Timnit Gebru, one of the most formidable critics of the generative AI craze. Worth a follow on Mastodon: https://dair-community.social/@timnitGebru

The problem is, that would limit my own option to make a version of the software and sell it under a more limited license in the future. Whomever I sell it to then has the right to go ahead and redistribute it, competing with me. Sure, my current, highly niche code already carries that risk, but the MIT license doesn’t stop me from releasing a modified version I may write that is more valuable as software, and then protecting that release with other licensing terms.

I am a consultant who sometimes writes code to do certain useful things as part of larger systems (parts of which may be commercial or GPL) but my clients always try to impose terms in their contracts with me which say that anything I develop immediately becomes theirs, which limits my ability to use it in my next project. I can to some extent circumvent this if I find a way to publish the work, or some essential part of it, under an MIT license. I’m never going to make money off of my code directly; at best it’s middleware, and my competitors don’t use the same stack, so I’m not giving them any real advantage… I don’t see how I’m sabotaging myself in this situation; if anything the MIT license is a way of securing my freedom and it benefits my future customers as well since I don’t have to rebuild from scratch every time.