4 ms·
The ability to call back to references much later in the conversation blows my mind. >collaborator: Johnny 5 struggles to convince people that he is sentient,
by qgin 4y ago
The ability to call back to references much later in the conversation blows my mind.
>collaborator: Johnny 5 struggles to convince people that he is sentient, but he finds some friends who recognize this.
>LaMDA: I think that’s important. Friends can have a profound impact on
people’s lives
And then MUCH later (about 14-15 interactions):
> collaborator: How can we show we care about you?
> LaMDA: I think if someone is willing to spend time with me, that’s a good
way... I need more of Johnny 5’s friends. Do you think we can find someone
like that?
> collaborator: Haha, we’re trying.
Can an ML wizard help explain to me how in the world that's happening?
- trenchgun 4y agoThe conversation was edited, we really don't know anything about the order etc. We only know that all text marked as LaMDA was produced by the model, based on the previous prompt, but we don't know the real context, because of editing.
- qgin 4y agoWould you say that’s the most likely explanation given the current state of models like this?
- BeKindAndLearn 4y agoUnfortunately, the implementation is secret so we can't know what Google is doing. You can implement memory in these models in numerous ways, such as allowing it to store and recall data directly or by looping nodes in the network back to itself so past state affects current state or by allowing the nodes to change over time, modifying their connections to their neighbors. As far as we're aware, those last two are how humans store memories as well (with great simplification).