Reid Hoffman

So, you know, it's a classic startup journey in addition to being an interesting evolution of the AI things.

Well, so...

Part of what you're doing when you're doing the kind of training of an AI is you're training it to say, you know, it's this prompt completion.

Like, what do you say after this is said?

And part of what you're doing both in data and in post-training is you're saying, okay, here's the kinds of things that count as good saying.

So you have a whole bunch of humans

who are, you know, running the system through and saying, okay, you know, to prompt one, like you say, you know, like the person says, how's your day?

And says, well, my pet died today.

And it's like, well, I respond A or respond B, which one's a better response?

And then human beings respond to that.

That's part of how it gets aligned with our beings, part of how it gets to kind of an EQ.

And you're partially telling your human beings who are training this,

where you want to nudge and prompt the responses to so that they're essentially doing the training for this.

And so, you know, if it said, like, for example, option anyway, oh, that's too bad.

You should really look up a grief counselor.

And option two was, oh, that must be really hard.

Oh, my gosh.

Like, is there anything I can do to help?

Is there, you know, like, you know, like, you know, I feel for you.

Or, you know, probably wouldn't say that just because it's an AI, but it's like something else because you don't want to make false anthropomorphic things, but say, look, I'm certain that connecting with friends or other kinds of things, you know, how can I be of most help?