3 ms·
> Is there a risk of spamming racist memes causing non-racist prompts to return racist memes? It is conceivable, if a naive self-improvement feedback system is
by dragonwriter 3y ago
> Is there a risk of spamming racist memes causing non-racist prompts to return racist memes?
It is conceivable, if a naive self-improvement feedback system is implemented. (This wouldn’t have to touch the base model at all, as there are ways of externally imposing qualitative guidance based on feedback.) But it is highly unlikely that any of the major online image generators are implemented in a way which would be vulnerable to this.