this post was submitted on 20 Aug 2026
60 points (87.5% liked)
Technology
87575 readers
3734 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Okay so the paper is actually interesting, and they have a explainer website: https://zheng-dai.github.io/AblationBasedCounterfactuals/
https://www.nature.com/articles/s41467-026-75667-5
https://github.com/zheng-dai/counterfactualuniverses
Amongst other things, they experimented with "what happens to a generation if you remove one image from the training dataset," particularly the "source" image for the target generation. What makes it interesting is they tested this at scale. There's (expectedly, somewhat janky) demo code to illustrate the spread. One example I exported myself:
Where each generation (the "counterfactual") in the top row corresponds to the image that was removed from the dataset in the bottom.
...Personally, I don't find the conclusion surprising.
These are models.
Lets say you make a model to predict hurricanes. With a small dataset/model, exclude the hurricane most similar to the one you're trying to predict, and it won't do a good job. But do the same with a model based on a huge dataset, and it should still model the novel hurricane reasonably well.
Diffusion models are no different.
I think AI Bros have overly anthropomorphized them with terms like "creativity" and such; they don't have this. But at the end of the day, they can model things that aren't strictly in their dataset. That's kind of the basic premise.
I mean, most people can't recognize slop vs creativity normally. I think the results affect less the anthropomorphizing issue, since oir brains do run similar predictive processes in how we generate words based on context,
Although it can't actively update its weights while navigating novel concepts, which is why you need a person to create novelty alongside the tool, or correct its mistakes, much like when you say the wrong thing and backtrack and correct. Your word generating process isn't perfect, but it's a useful tool that is useful alongside all of your other 'tools.'
What i think the study is a better answer to is the people who call it a theft machine because it learned from things that exist.
That being said, rich assholes should not be able to turn around and claim ownership of the result of the learning based on the collective development and accomplishment of mankind, or life more broadly. We 'steal'. A lot of useful intelligent tools from nature.
It's hard to start that conversation though when the 'bad' thing they are doing is just learning or training.
We had the slop of 90s-2000s Disney sequels, even if it was a bunch of human interns having their time and energy pilfered with the threat of basic survival or right to spend time creating art instead of flipping burgers. All proper and copyright safe, but closer to the actual problem than some artists trying to actually create novel things with these models replacing the Disney interns that only Disney has 'rightful' access to, despite gaining success off of public domain, and then lobbying to shrink the public domain.
But people are more eager to be mad at the technology than the complexity of dealing with the opportunistic assholes that will use every tool, technology or otherwise, to ensure they get to have power over our collective rights and autonomy.