You're not just using a tool — you're co-authoring the science.
This is not what an ablation study is. An ablation study removes and/or swaps out ("ablates") different components of an architecture (be it a layer or set of layers, all activation functions, backbone, some fixed processing step, or any other component or set of components) and/or in some cases other aspects of training (perhaps a unique / different loss function, perhaps a specialized pre-training or fine-tuning step, etc) in order to attempt to better understand which component(s) of some novel approach is/are actually responsible for any observed improvements. It is a very broad research term of art.
That being said, the "Ablation Strategies" [1] the repo uses, and doing a Ctrl+F for "ablation" in the README does not fill me with confidence that the kind of ablation being done here is really achieving what the author claims. All the "ablation" techniques seem "Novel" in his table [2], i.e. they are unpublished / maybe not publicly or carefully tested, and could easily not work at all. From later tables, I am not convinced I would want to use these ablations, as they ablate rather huge portions of the models, and so probably do result in massively broken models (as some commenters have noted in this thread elsewhere).
[1] https://github.com/elder-plinius/OBLITERATUS?tab=readme-ov-f...
[2] https://github.com/elder-plinius/OBLITERATUS?tab=readme-ov-f...
EDIT: As another user mentions, "ablation" has a specific additional narrower meaning in some refusal analyses or when looking at making guardrails / changing response vectors and such. It is just a specific kind of ablation, but see e.g. https://huggingface.co/blog/mlabonne/abliteration
Try to think for a moment about how a device would "find nearby microphones" or how it would use an AI-generated signal to cancel out your voice at the microphone. This should be setting of BS alarms for anyone.
It seems the Twitter AI edgey poster guy is getting meta-trolled by another company selling fake AI devices
It just says "the README sucks." Which, I'm inclined to agree, it does.
LLM-generated text has no place in prose -- it yields a negative investment balance between the author and aggregate readers.
I just hear him promoting OBLITERATUS all day long and trying to get models to say naughty things
That doesn't mean there couldn't be a "concept neuron" that is doing the vast majority of heavy lifting for content refusal, though.
Its basically using a compression technique to figure out which logits are the relevant ones and then zeroing them.
[0] https://en.wikipedia.org/wiki/Singular_value_decomposition
I use Berkley Sterling from 2024 because I can trick it. No abliteration needed.
Also, as I said in a top level comment, what this project wants to achieve has been done for a while and it's called Heretic: https://github.com/p-e-w/heretic
(Not vibecode by a twitter influgrifter)
And yeah, doing stuff like deleting layers or nulling out whole expert heads has a certain ice pick through the eye socket quality.
That said, some kind of automated model brain surgery will likely be viable one day.
It's interesting that people are writing tools that go inside the weights and do things. We're getting past the black box era of LLMs.
That may or may not be a good thing.
However, after a few rounds of conversation, it gets into loops and just repeats things over and over again. The main JOSIE models worked the best of all and was still useful even after abliteration.
p-e-w's Heretic (https://news.ycombinator.com/item?id=45945587) is what you're looking for if you're looking for an automatic de-censoring solution.
Its not a frontier model but it will give you a feel for what its like.