undefined

points

[-]

I run an abliterated distillation of Qwen 3.8 27B, slightly quantized to fit on my 4090, and I've been evaluating it to use as a worker bee for research directed by a smarter model.

Much like in the article, abliterated Qwen will not obey restrictions on its behavior encoded in the prompt. If you want something not to happen, it better be enforced in the harness or environment (e.g. sandbox). It is much different than the Anthropic models I'm used to, which will, the vast majority of time, follow rules (before auto mode, I used to always run them in "yolo" mode).

I am curious whether there's a connection between abliteration and rule following. These abliterated models are the ones you most want to follow your rules.

by sroussey11 hours ago|

parent|

[-]

Language models have always had an issue with negatives.

A negative like do “not” xyz is just not encoded the same as spelling out what you want vs what you don’t want.

Harder to write though.

by sigmoid108 hours ago|

parent|

[-]

I would say in this case abliteration is the likely culprit. To uncensor a model this way, you literally deactivate the parts that would enact refusals. As in things it was told not to do. But the real process is more like brain surgery performed by a alchemist according to an ancient religious book where noone involved really understands what is actually happening in the model.

by AndyNemmity8 hours ago|

parent|

prev|

[-]

Exactly, I wrote a blog post in what feels like a long time ago on this topic.

https://vexjoy.com/posts/positive-framing-agents-skills/

by PotatoPrime7 hours ago|

parent|

[-]

Interesting read, thanks for re-sharing!

I noticed your joy-check link 404's now... I tried poking around your /skills/ folder but didn't find it easily. Should you still have that available I'd love to check it out.

edit: Found it if others are looking: https://github.com/notque/vexjoy-agent/blob/main/skills/code...

by AndyNemmity3 hours ago|

parent|

[-]

Oh, thanks for letting me know. I need to fix that.

by barbazoo12 hours ago|

prev|

[-]

Efficiently at scale or even as an individual?

by sigmoid1012 hours ago|

parent|

[-]

You'll need tons of tokens for these to churn through heavy reasoning and feedback driven tasks. You might even want to run many agents in parallel. Combine that with GPU memory requirements in the terabyte range just to hold the weights and you've pretty much left local consumer setups entirely.

by Terretta12 hours ago|

prev|

[-]

Note that Mac Studio Ultra M3 (or upcoming M5) with 512GB is effective. You don't have to do this work fast, overnight is fine.

Unless trying to use it interactively and adversarially, in which case it's not fast enough plus would be why those of us without our own datacenters will get told we can't have nice things.