> What makes this build different is the word before FP8: uncensored. We applied abliteration — orthogonalizing the refusal direction out of the residual stream — to remove the model's safety-alignment refusals. The result is a model that will comply with requests the original would refuse.

Surely this has unintended side effects on output quality?

▲

andsoitis 7 hours ago | parent | next [-]

> > What makes this build different is the word before FP8: uncensored. We applied abliteration — orthogonalizing the refusal direction out of the residual stream — to remove the model's safety-alignment refusals. The result is a model that will comply with requests the original would refuse.

> Surely this has unintended side effects on output quality?

Can you help me understand why that's the case?

▲

willy_k 5 hours ago | parent [-]

Because deleting model weights after training is likely to cause knock-on effects in model knowledge and/or behavior. Targetting it might mitigate this but it’s

a) not guaranteed that only censor-ey parameters get removed, and b) likely that removing those parameters still has effects on the effectiveness of related parameters.

	▲	jszymborski 4 hours ago \| parent [-]
		The weights aren't deleted, it's just additional fine tuning, is my understanding.

▲

DiabloD3 7 hours ago | parent | prev | next [-]

It does depending on the technique.

▲

timmmmmmay 4 hours ago | parent | prev | next [-]

Early attempts at this sort of thing definitely did, but these days the impact is minimal

▲

miroljub 7 hours ago | parent | prev [-]

A bit worse quality is a fine trade off when the alternative is no output (zero quality).

	▲	radlad 7 hours ago \| parent [-]
		On censored inputs only.