ideas , given 2023 situation

[dobkeratops]

observations about the current state of AI..

so over the past year there's been these diffusion models & large language models become available to be public

There's the controversy over training data (scrapes)

• e.g. Steam doesn't allow games with AI generated content, or even use of ChatGPT ; they're covering themselves

• That seems to me to validate the need for a efforts such as imagemonkey to produce more CC0 open training data?

Individuals such as ourselves can't currently produce foundation models.

but models like stable-diffusion can be fine-tuned (stability.ai's argument about the scrapes is that a fine-tuned model can remove its ability to recreate copyrighted material in the original scrapes, which sometimes come through)

• I see that you had VGC finetuning integrated in imagemonkey.. have you considered extending your service to finetune diffusion models?

• Also consider that in time multi-modal models may appear.. ones trained simultaneously on images & text

[dobkeratops]

https://github.com/haotian-liu/LLaVA

opensource multimodal LLMs .. text+vision .. imagine being able to finetune these on community curated datasets
another potential usecase for imagemonkey