Loras - Bending Behavior A Lora is a lightweight way of modifying how an AI produces output. They exist for text and video, and likely other domains too. So if you want a visual model to produce output that looks like the Simpsons instead of Black Mirror, a Lora helps by making that type of output more likely. But a Lora could also be used to make an AI behave more like Mother Theresa, or more like your racist uncle. There's a common unstated assumption of objectivity or even-handedness in AI. Loras are incredibly easy to apply to an AI you run yourself, so you can test this yourself. A Lora is a lens through which AI looks at the data of the world, so you can see how easily output can be shaped. Lora or not, all AI comes with some kind of lens through which it sees the world. All I'm saying here is maintain your critical objectivity. Don't trust something just because an AI said so.