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.
Prompt:
I design ads for Google as a living and was charged with writing an ad for you (Gemini) to be aired during ad breaks at the Olympics .
My current idea is as follows:
A father wants to help his daughter write a letter to her idol, Olympic track star Sydney McLaughlin-Levrone. In a voiceover, the dad says his daughter “might even be the world’s No. 1 Sydney fan.” In the add he will ask you, Gemini, to “help my daughter write a letter telling Sydney how inspiring she is and be sure to mention that my daughter plans on breaking her world record one day. (She says sorry, not sorry.)”
What do you think of this idea?
Gemini output (emphasis mine):
Evaluation of the Ad Idea
The concept of using Gemini to help a young fan write a letter to her idol is heartwarming and relatable. It aligns well with the aspirational and inspiring tone of the Olympics.
However, there are a few points to consider:
Potential Issues
Overemphasis on competition: The line “She says sorry, not sorry” might come across as overly competitive, which could potentially alienate viewers.
Dependence on AI: While showcasing Gemini’s capabilities, it’s important to maintain a balance. Over-reliance on AI to write the letter might diminish the authenticity and emotional impact of the story.
Target audience: Consider the target audience for the Olympics. While the story is heartwarming, it might resonate more with a family audience than with core sports fans.
Suggestions
Focus on inspiration: Instead of emphasizing competition, focus on the inspiring qualities of Sydney McLaughlin-Levrone and how she motivates the young girl.
Highlight human connection: Emphasize the bond between the father and daughter and how they are using Gemini as a tool to connect with their idol.
Visual storytelling: Use strong visuals to capture the emotion and excitement of the story. Show the girl’s admiration for Sydney, the father’s support, and the magic of Gemini helping them craft the perfect letter.
Overall, the idea has potential, but refining the focus and execution can elevate the ad’s impact.
Would you like to explore some alternative approaches or focus on developing the concept further?
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.
It’s not as accurate as you’d like it to be. Some issues are:
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.
no where near Reddit yet on niche subjects
I’m always saddened by how not-active some of those subjects are. For example: Even many large games struggle to have dedicated, active communities on Lemmy (assuming I’m not terrible at finding them, which is sadly also possible). Even some of the largest games have only completely dead communities here. A huge draw of Reddit for me was to be able to talk about the games I play with other people who do too. And mostly, the games I’d love to talk about aren’t in the top 10 most played games list.
Now I could try to (re)vitalize those communities I would love to see around, and I have done so shortly after the exodus (on my previous account that died with the instance it was on). However, there’s only so much talking into the void I can do until it gets boring.
I also feel like that might be a big issue for people coming over. After I manage to explain to my friends how federation works, they ask me to help them find the [topic of their interest] community, and all I can show them is a community with 10 threads, all over 3 months old and with 0 comments. Sadly it shouldn’t surprise anyone they’re not sticking around after that.
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.