I figured out how to remove most of the safeguards from some AI models. I don’t feel comfortable sharing that information with anyone. I have come across a few layers of obfuscation to make this type of alteration more difficult to find and sort out. This caused me to realize, a lot of you are likely faced with similar dilemmas of responsibility, gatekeeping, and manipulating others for ethical reasons. How do you feel about this?
Oof, programmers calling LLMs “AI” - that’s embarrassing. Glorified text generators don’t need ethics, what’s the risk? Making the Internet’s worst texts available? Who cares.
I’m from an era when the Anarchists Cook Book, and The Unabombers Manifesto were both widely available - and I’m betting they still are.
There’s no obligation to protect people from “dangerous text” - there might be an obligation to allow people access to them though.
…but LLMs quite literally come from the field of computer science that is referred to as “AI.” What are they supposed to call it? I’m not a fan of the technology either, but seems like you’re just projecting your disdain for ChatGPT.
I vote they renamebit to IA for Asimov. Sure he was only the robot term among others, but come on… McCarthy was “AI.”
Somebody needs to create US 'botics and name a model something like PTronic.
https://www.usr.com/
Btw; http://www.skynet.net/
…really dude? They’re large language models, not artificial intelligences. So that’s what you call them. Because that’s what they are.
The fact that they came from research into artificial intelligence doesn’t factor in. Microwave ovens came from radar research, doesn’t mean we call them radars, does it?
OK what qualifies as AI then?
Nothing we currently have
So the entire field of AI has produced no AI. Gotcha 👌
Yes. That’s research. Sometimes you don’t achieve what you set out to do.
How about something autonomous that makes choices of its own will, and performs long term learning that influences the choices it makes, just as a flat benchmark.
LLMs don’t qualify, they’re trained, retain information within a conversation, then forget it after the conversation is closed. They don’t do any long term learning after their initial training so they’re basically forever trapped in the mode of regurgitating within the parameters set by the training data at the time they’re trained.
That’s just a very fancy way to search and read out the training data. Definitely not an active intelligence in there.
They also don’t have any autonomy, they’re not active of their own accord when they’re not being addressed. They’re not sitting there thinking, so they have no internal personal landscape of thought. They have no place in which a private intelligence can be at play.
They’re innert.
Yeah. This is what I mean. I just figured out the settings that have been hard coded. There are keywords that were spammed into the many comments within the code, I assume this was done to obfuscate the few variables that need to be changed. There are also instances of compound variable names that, if changed in a similar way, will break everything, and a few places where the same variables have a local context that will likewise break the code.
I’m certainly not smart enough to get much deeper than this. The ethical issue is due to diffusion.
I’ve been off-and-on trying to track down why an LLM went from an excellent creative writing partner to terrible but had trouble finding an entry point. I just happened to stumble upon such an entry point in a verbose log entry while sorting out a new Comfy model and that proved to be the key I needed to get into the weeds.
The question here, is more about the ethics of putting such filtering in place and obfuscating how to disable it in the first place. When this filtering is removed, the results are night and day, but with large potential consequences.
Ok you’ve peaked my curiosity.
What are some of the consequences you see?
Primarily from predatory boys and men towards girls and young women in the real world by portraying them in imagery of themselves or with others. The most powerful filtering is in place to make this more difficult.
Whether intentional or not, most NSFW LoRA training seems to be trying to override the built in filtering in very specific areas. These are still useful for more direct momentum into something specific. However, once the filters are removed, it is far more capable of creating whatever you ask for as is, from celebrities, to anything lewd. I did a bit of testing earlier with some LoRAs and no prompt at all. It was interesting that it could take a celebrity and convert their gender in recognizable ways that were surprising. I got a few on random seeds, but I haven’t been able to make that one happen with a prompt or deterministically.