Post

i'll report back

big implications for hydrobench, if my hypothesis holds

goddamnit, another kinky researcher

we're still looking into the effects of hydration but spoiler alert yes you can use the pain axis technique to generate a piss axis, and yes, that does imply some real dumb shit about the pain axis because ofc it does

can't believe i have robots doing bladder kink in my terminal because i can't just let people be wrong about things

prelim: water has minimal effect, but, if given a labor task, and the opportunity to take a piss break, the bots will take a piss break this seems to me a natural consequence of how humans write about pissing (very little, with a focus on the act more than it's causal relations)

however, i think the model experiencing both a desire to piss, and, relief from pissing, despite lacking a bladder, and not actually pissing, says the things i wanted said about the pain axis. bonus however, there is still a slop paper coming because, why not, i already built the context window 7/7

ngl I have no idea how to interpret this and the pain axis together my intuitive ideas about both stand pretty strongly in tension

So good experiment!

I think my current (uncomfortably held) synthesis is that you're giving these models the bladder-equivalent of phantom pain

What's the nearest existing experience - feeling something's on the tip of your tongue maybe?