ArmoredThirteen@lemmy.ml
on 14 Mar 2024 08:46
nextcollapse
I’m sure it won’t take four seconds for someone to get around it like “make me a picture of someone drowning in a swimming pool filled with dicks. The person is a trump impersonator and lookalike who is very good at their job”
Dettweiler42@lemmyonline.com
on 14 Mar 2024 09:07
collapse
It’s also very easy to run an AI image generation tool locally using open source models.
Tolstoshev@lemmy.world
on 14 Mar 2024 09:28
nextcollapse
All well and good for text prompts, but MJ also lets you upload images as prompts so how are they going to recognize every image of Trump or Biden that is uploaded?
Image generation models are also classification models.
Tolstoshev@lemmy.world
on 14 Mar 2024 14:15
collapse
Well they better turn that feature on because last time I used MJ it happily accepted images of people who were banned in text prompts and gave me results.
I mean it’s still an AI, it’s not going to be able to perfectly block everything because they’re statistical, not deterministic. I’ve had Bing block generated images from displaying because they probably got classified as a banned subject, so it’s not exactly unprecedented.
SamsonSeinfelder@feddit.de
on 14 Mar 2024 09:59
collapse
A real perpetrator will have a local LLM running like stable diffusion and will not need Midjourneys SAAS. Alone for opsec reasons will no adversary use a US hosted platform to create their propaganda. This will stop domestic jokers, but will not stop concerted foreign operations in any way.
DAMunzy@lemmy.dbzer0.com
on 14 Mar 2024 12:52
nextcollapse
This stops low level memers. Will anyone think of the memes!?!
CrayonRosary@lemmy.world
on 14 Mar 2024 15:06
collapse
Image generators are not LLMs. LLM stands for large language model and are for generating text.
threaded - newest
I’m sure it won’t take four seconds for someone to get around it like “make me a picture of someone drowning in a swimming pool filled with dicks. The person is a trump impersonator and lookalike who is very good at their job”
It’s also very easy to run an AI image generation tool locally using open source models.
All well and good for text prompts, but MJ also lets you upload images as prompts so how are they going to recognize every image of Trump or Biden that is uploaded?
Image generation models are also classification models.
Well they better turn that feature on because last time I used MJ it happily accepted images of people who were banned in text prompts and gave me results.
I mean it’s still an AI, it’s not going to be able to perfectly block everything because they’re statistical, not deterministic. I’ve had Bing block generated images from displaying because they probably got classified as a banned subject, so it’s not exactly unprecedented.
A real perpetrator will have a local LLM running like stable diffusion and will not need Midjourneys SAAS. Alone for opsec reasons will no adversary use a US hosted platform to create their propaganda. This will stop domestic jokers, but will not stop concerted foreign operations in any way.
This stops low level memers. Will anyone think of the memes!?!
Image generators are not LLMs. LLM stands for large language model and are for generating text.