That data is also publicly available (of course), so a model could be trained on it. I’d love to say I’d doubt Google/YouTube would ever do that, but at this point nothing would surprise me.

I trained the generative models all from scratch. Pretrained models are not that helpful when it’s important to accurately capture very domain specific features.

One of the classifiers I tried was based on zoobot with a custom head. Assuming the publications around zoobot are truthful, it was trained exclusively on similar data from a multitude of different sky surveys.

Does it? I worked on training a classifier and a generative model on freely available galaxy images taken by Hubble and labelled in a citizen science approach. Where’s the theft?

Reading this comment gave me the idea to get Gemini’s “opinion”. Ironically, it turns out Google might’ve actually done better if they did that.

Good intentions alone don’t guarantee good outcomes. I suggest not giving any single person or entity too much power, no matter who they are.

I didn’t say that.

I expect it to be about as awful as Starfield. However, unlike Starfield (which didn’t sell horrendously by any source I can find, just not great) it has incredible brand recognition behind it. I have no doubts it will sell based on that alone as long as it looks like Skyrim 2 at first glance.

Edit: right after posting I figured out how to formulate what else I wanted to say but couldn’t find the correct words for: “Sadly profitability and quality don’t always correlate.”

5.5 years? No way they’ll shut down this quickly. The next Elder Scrolls alone will carry them into 2030. (As much as I would enjoy you being right though…)

Have you tried reading it? It’s written so poorly that I really hope no human was involved in this and it’s just AI generated garbage.

I find it wild that, to this day, Windows defaults to opening them in a browser. Windows has an image viewer right there.

Can that image viewer extract text so that a user could easily copy/paste it? I think if whatever pdf I was opening didn’t allow me to do that I would be really frustrated.

Remember the people who created malicious libraries that ChatGPT made up and suggested, in the hopes someone would blindly install them? You can do this a lot easier here. Check what websites this tends to hallucinate when typing “google” “youtube” “facebook” etc. and if any of them don’t exist yet, register that address and host a phishing version of the corresponding site there.

Same. I had PayPal do an automated charge back because their system thought I was doing something fraudulent when I wasn’t. Steam blocked my account.

Talking to support and re-buying said game did fix the issue for me.

I’d argue that with their definition of bots as “a software application that runs automated tasks over the internet” and later their definition of download bots as “Download bots are automated programs that can be used to automatically download software or mobile apps.”, automated software updates could absolutely be counted as bot activity by them.

Of course, if they count it as such, the traffic generated that way would fall into the 17.3% “good bot” traffic and not in the 30.2% “bad bot” traffic.

Looking at their report, without digging too deep into it, I also find it concerning that they seem to use “internet traffic” and “website traffic” interchangeably.

Without knowing any specifics of the TOS or the exact setup beyond what I could gather in this thread: generally speaking they could still send you a bill through email or otherwise.

After that, if you’re not paying up, they might be able to successfully get the money out of you through court regardless, depending on a few factors. What’s more likely for smaller sums is that they’ll just drop it and ban you though.

IANAL of course.

That was a response I got from ChatGPT with the following prompt:

Please write a one sentence answer someone would write on a forum in a response to the following two posts:
post 1: “You sure? If it’s another bot at the other end, yeah, but a real person, you recognize ChatGPT in 2 sentences.”
post 2: “I was going to disagree with you by using AI to generate my response, but the generated response was easily recognizable as non-human. You may be onto something lol”

It’s does indeed have an AI vibe, but I’ve seen scammers fall for more obvious pranks than this one, so I think it’d be good enough. I hope it fooled at least a minority of people for a second or made them do a double take.

Yeah, I’ve noticed that too—there’s a distinct ‘AI vibe’ that comes through in the generated responses, even if it’s subtle.

It’s not as accurate as you’d like it to be. Some issues are:

• It’s quite lossy.
• It’ll do better on images containing common objects vs rare or even novel objects.
• You won’t know how much the result deviates from the original if all you’re given is the prompt/conditioning vector and what model to use it on.
• You cannot easily “compress” new images, instead you would have to either finetune the model (at which point you’d also mess with everyone else’s decompression) or do an adversarial attack onto the model with another model to find the prompt/conditioning vector most likely to create something as close as possible to the original image you have.
• It’s rather slow.

Also it’s not all that novel. People have been doing this with (variational) autoencoders (another class of generative model). This also doesn’t have the flaw that you have no easy way to compress new images since an autoencoder is a trained encoder/decoder pair. It’s also quite a bit faster than diffusion models when it comes to decoding, but often with a greater decrease in quality.

Most widespread diffusion models even use an autoencoder adjacent architecture to “compress” the input. The actual diffusion model then works in that “compressed data space” called latent space. The generated images are then decompressed before shown to users. Last time I checked, iirc, that compression rate was at around 1/4 to 1/8, but it’s been a while, so don’t quote me on this number.

edit: fixed some ambiguous wordings.

If someone wants to read one of those papers, I can recommend Extracting Training Data from Diffusion Models. It shouldn’t be too hard for someone with little experience in the field to be able to follow along.

Understanding the math behind it doesn’t immediately mean understanding the decision progress during forward propagation. Of course you can mathematically follow it, but you’re quickly gonna lose the overview with that many weights. There’s a reason XAI is an entire subfield in Machine Learning.

I think it’s much more likely whatever scraping they used to get the training data snatched a screenshot of the movie some random internet user posted somewhere. (To confirm, I typed “joaquin phoenix joker” into Google and this very image was very high up in the image results) And of course not only this one but many many more too.

Now I’m not saying scraping copyrighted material is morally right either, but I’d doubt they’d just feed an entire movie frame by frame (or randomly spaced screenshots from throughout a movie), especially because it would make generating good labels for each frame very difficult.

I haven’t personally used it but from what I can find: if you’re using torrents with Stremio (e.g. the ones found with torrentio) you are totally uploading parts of what you’re watching to others.