dalle-2-preview VS laion-datasets

We believe that many decisions about our defaults and hard bounds should be made collectively, and while practical implementation is a challenge, we aim to include as many perspectives as possible. As a starting point, we’ve sought external input on our technology in the form of red teaming. We also recently began soliciting public input on AI in education (one particularly important context in which our technology is being deployed).

From this OpenAI document:

It's also of note that you can't sell a game using this method, as Dalle-2's terms of service prevent use in commercial projects. It's hard to justify rate of return considering you can only ever give it away for free, and even in that case there are some uncertain legal elements regarding copyright and the images that are used to train the dataset.

On their GitHub https://github.com/openai/dalle-2-preview/blob/main/system-card.md I can only see references to v1.

"b) Exploration of the bolded part of OpenAI's comment "Each generated image includes a signature in the lower right corner, with the goal of indicating when DALL·E 2 helped generate a certain image." (source)." (source link: https://github.com/openai/dalle-2-preview/blob/main/system-c...)

I feel the DALL-E 2 watermark signature could be a seed or something.

DALL-E's docs for example mention it can output whole copyrighted logos and characters[1] and understands it's possible to generate human faces that are bear the likeness of those in the training data. We've also seen people recently critique Stable Diffusion's output for attempting to recreate artists' signatures that came from the commercial trained data.

That said by a certain point the kinks will be ironed out and likely skirt around such issues by only incorporating/manipulating just enough to be considered fair use and creative transformation.

[1] "The model can generate known entities including trademarked logos and copyrighted characters." https://github.com/openai/dalle-2-preview/blob/main/system-c...

Some types of content were purposely excluded from the training dataset(s) (source).

Not true, it uses the MIT license, which allows for any use including commercial. According to the license you could even sell the Laion datasets yourself if you wanted.

Datasets can be manually curated to produce more aesthetic results if this becomes a real issue. For example, classifiers can predict whether an image is generated or not. You could adapt the process used to create laion-aesthetic[0] to remove generated images.

[0]: https://github.com/LAION-AI/laion-datasets/blob/main/laion-a...

source: https://github.com/LAION-AI/laion-datasets/blob/main/laion-aesthetic.md

My hunch is that is the result of this: https://github.com/CompVis/stable-diffusion#weights

> 515k steps at resolution 512x512 on "laion-improved-aesthetics" (a subset of laion2B-en, filtered to images with an original size >= 512x512, estimated aesthetics score > 5.0

https://github.com/LAION-AI/laion-datasets/blob/main/laion-a... for more details.

What's remarkable is this: https://github.com/LAION-AI/laion-datasets/blob/main/laion-a...

That aesthetic predictor was apparently trained on only 4000 images. If my thinking is correct, imagine the impact those 4000 ratings have had on all of the output of this model.

You can see samples (some NSFW) of different images from the original training set in different rating buckets here, to get an idea of what was included or not in those training steps. http://3080.rom1504.fr/aesthetic/aesthetic_viz.html