i tried to explain that in various ways and repeating myself and doesn’t sure it useful to do that again, and i get the feeling from that replay that you simply not understand my claim, despite me already made it in previous comment, and it doesn’t feel you engage with any of my claims, so here is one last attempt:
if someone who err 1 time out of 100 make the conclusion is time was off, and change his algorithm to one that does not tell his joke, instead of updating his probability to tell this joke slightly less in a way that make him still tell the joke, he is doing the wrong update.
this is not what you do when you program simple neuron network and update in with backpropagation. the mathematically right update is “i have slightly less pobability to commit the same error, but i would have behave the same way unless i get this input a lot of more times”.
so you advocate doing the mathematically-wrong thing. we should learn what we can, but what you advocate is literally over-correction.
i tried to explain that in various ways and repeating myself and doesn’t sure it useful to do that again, and i get the feeling from that replay that you simply not understand my claim, despite me already made it in previous comment, and it doesn’t feel you engage with any of my claims, so here is one last attempt:
if someone who err 1 time out of 100 make the conclusion is time was off, and change his algorithm to one that does not tell his joke, instead of updating his probability to tell this joke slightly less in a way that make him still tell the joke, he is doing the wrong update.
this is not what you do when you program simple neuron network and update in with backpropagation. the mathematically right update is “i have slightly less pobability to commit the same error, but i would have behave the same way unless i get this input a lot of more times”.
so you advocate doing the mathematically-wrong thing. we should learn what we can, but what you advocate is literally over-correction.
it seems your model is that telling a joke is like rolling a die. sometimes you get a 1, and the joke falls flat.
this is not my model at all. each joke is different. you can learn which ones work and which ones don’t.
maybe it helps to ask: how does the person get good enough at jokes that they succeed 99⁄100 times? they have been learning, no? so do more of that.
no, this is not what i said. this is not at all what i said, and i actually said something that is very close to the opposite of that.
i don’t know who to communicate if i say something and you come to the conclusion that i believe the opposite of that, so i’m bowing out.