"Your text batch moved the weights away from the final values. Your contribution is negative."
replyWhere do I collect the $0.00000012 antidollars owed to me by OpenAI for my valuable inputs?
replySlightly more seriously, you could perhaps make an argument that, just like weight decay, an apparent "anti-contribution" moves the learning trajectory along, and helps the network settle into a more optimal basin eventually.
That way, my contribution is still valuable on the net, and I'm owed $0.00000003 positive dollars instead.
>you could perhaps make an argument that, just like weight decay, an apparent "anti-contribution" moves the learning trajectory along
replyWas that not the joke?