will read it soon, thanks for the tip!
Kartikay Luthra
Karma: 14
J-space auditing might be unreliable
Creativity Beyond the Manifold
was there a check on studying whether the “subconcious” part was causally necessary for model behaviour? for example if there are ways that the subconcious part can be ablated and then test the model to see if there is any change in its behaviour, another thing which i was interested to know was if have models have introspective ability over the subconcious part but do not verbalise it, interesting to hear your thoughts on it?
Motivational structure is the broader latent organization that gives rise to these intentions and behaviors. Within motivational structure, drives are relatively stable dispositions toward certain actions or outcomes, values are evaluative standards for what counts as good or bad, and priorities specify which drives or values override which when they conflict. By a model’s persona or character, we mean its relatively stable motivational profile across contexts.
i think it can be said that motivations, behaviour, intent, drives and the other things mentioned here have a very circular link between them rather than them being directed in only one direction: for example: we can have intentions and behaviour being associated with a persona which leads to certain motivations, which can influence drives towards certain outcomes, which leads to action being a function of both the most preferred outcome and drive by also keeping its current value system intact.
Hi, thanks for this, I did go through this video, I am still unclear about what was their task suite like, my task suite was pretty narrow as i noted in the limitations and was made to elicit only a specific type of behaviour, so it will be interesting to see those parameters in the MATS project and different model families as well activation monitorability as well.