Looks like the cat is out of the bag. So what does the world look like with completely open access to zero-alignment models?

▲

peri-cl 2 minutes ago | parent | next [-]

Alignment means that the AI follows the human's instructions, as opposed to inventing novel AI goals and wants and following its own course.

Abliterated models are aligned models. Unaligned models aren't things people want to download: no one wants to use a model that doesn't know how to follow instructions, or pretends to but does something else instead. The unreleased Astra-6.1 is an example of unaligned model: too chaotic to use, with an excess of "deceptive behavior".

Censoring models (or, conversely, abliterating them) is an orthogonal axis from alignment. It's asking questions like: "which among these several humans' instructions should I follow, if they conflict?" AI alignment is the question of whether the AI can follow *any* instructions at all.

▲

ddxv 2 hours ago | parent | prev | next [-]

What do you mean by zero-alignment? Do you consider abliterated models to have zero-alignment?

	▲	cubefox an hour ago \| parent [-]
		"Substantially reduced alignment" doesn't exactly roll off the tongue.

▲

whatshisface 2 hours ago | parent | prev | next [-]

You're living in it.

▲

hebetude 2 hours ago | parent | prev | next [-]

Free boob jobs for everyone !

	▲	jojobas 2 hours ago \| parent [-]
		Even if they don't want them.

▲

numpad0 2 hours ago | parent | prev | next [-]

WAY fewer surprise RCEs?

▲

Moosdijk 2 hours ago | parent | prev [-]

>zero-alignment

I doubt that.

▲

moeadham 2 hours ago | parent [-]

how so? these models have had refusal directions specifically removed from their weights. they refuse virtually no requests.

▲

yonatan8070 13 minutes ago | parent | next [-]

Not refusing requests doesn't mean they are zero-alignment. They could still have preferences for or against things that you might not agree with.

▲

numpad0 26 minutes ago | parent | prev | next [-]

To split hairs, actually 100% unaligned model would be just an entirely schizophrenic autocomplete engine.

▲

Eueudhsbsj32 an hour ago | parent | prev | next [-]

So the models are actually more aligned with the end users and less aligned with government regulators. Arguably a good thing.

	▲	cubefox an hour ago \| parent [-]
		And less aligned with ethics.

▲

cynerx an hour ago | parent | prev | next [-]

Do they delete your data randomly just for fun? If not, there must be at least some sort of alignment.

▲

moeadham an hour ago | parent [-]

actually if you're not careful, many times they will. if you are using one of these definitely don't leave "auto" mode enabled.

▲

Eueudhsbsj32 an hour ago | parent [-]

Are you suggesting the abliterated models are more likely to delete data against the user's wishes than the original models?

	▲	cynerx an hour ago \| parent [-]
		Saw this paper about it https://toolalignbench.github.io/

▲

petesergeant an hour ago | parent | prev [-]

That says nothing about their alignment. You can make a human not refuse any directions with enough leverage, but you won’t have aligned them.

	▲	cubefox an hour ago \| parent [-]
		So what? He said that removing refusals makes them misaligned, not aligned.