Episode Transcript
[00:00:12] Speaker A: Professor Mark.
[00:00:13] Speaker B: Hello, Lomitia.
[00:00:15] Speaker A: Nice to meet you.
[00:00:16] Speaker B: Good to see you again.
[00:00:17] Speaker A: Yeah, I've learned about that. You showed students Chat GPT four with voice version.
[00:00:29] Speaker B: First.
A gentleman named Ethan Moloch has a very informative LinkedIn profile, and he gave an example of how to use chat GPT in negotiations. And as you know, I teach a negotiations class every fall where we partner with Indiana University.
And I thought this was perfect. So I experimented with some prompting and realized that it's a very good teaching tool for negotiations. It does very well. So as background chat GBT is so OpenAI organizes all this, they have another tool called Whisper, which is a voice to text recognition tool. So you can speak. It's very good. So if you have the iOS version on your phone of Chachi BT, you can speak into it. It will transcribe what you're saying, and it does very, very good, in my experience.
And as of the end of September, I forget they added voice. So instead of just responding in the text window, you can pick a voice that it will use to respond to you.
[00:02:04] Speaker A: So you mean it's a version like the original chat GPT is.
The original chat GPT will show text on the window. And now it could speak.
[00:02:17] Speaker B: That's exactly right.
So you literally just turn on the voice function and you start speaking to it naturally, and it will respond to you in voice, in one of the voices it has.
[00:02:34] Speaker A: And how do you think the voice is? Like? The speed or the content is just, like, human?
[00:02:48] Speaker B: My experience is the pace, the cadence, everything about the voice seems very natural. I mean, there's some times where there's a glitch where, and I don't know if it's because it's of the Internet. If the Internet is not that great, sometimes maybe there'll be a pause or it'll go a little bit faster. It's trying to catch up. But I just kind of think of that maybe as an Internet issue rather than the tool having a problem. I could be wrong, but overall, I feel like it's really good is the.
[00:03:26] Speaker A: Pause just like in the original version, when you prompt and GPT will also pause?
[00:03:34] Speaker B: You mean if you tell it to pause?
[00:03:37] Speaker A: No, if I have a long prompt or a complex prompt, GPT will pause.
[00:03:44] Speaker B: It's not that pause, but GPT will do that too. So I've noticed that the cadence, the pace is very good. So it will insert pauses that are kind of normal in normal speech. Right. So somehow it kind of knows how to pace its words.
When I experiment it, I feel like. Yeah, you can kind of tell it's a machine, but it's a very, very good machine.
[00:04:15] Speaker A: Does the machine voice have any accent?
[00:04:18] Speaker B: Not in English.
[00:04:21] Speaker A: You mean? Not in English. So it's an American accent?
[00:04:25] Speaker B: Yeah, I think the voice I use is an American English accent. Yes.
[00:04:33] Speaker A: So can we switch the accent?
[00:04:37] Speaker B: I haven't experimented with that.
There's a few different voices, and I know that we've experimented with Mandarin, too.
So it speaks Mandarin and it speaks Mandarin with an accent.
At least it doesn't sound like a native Taiwanese speaker or Chinese speaker, but it has a slight accent.
[00:05:09] Speaker A: Is that the resource?
[00:05:11] Speaker B: I don't know, but what we've been using it for in negotiation is I've been experimenting with asking it. So on the prompt engineer side, asking it to create a scenario first, telling it that it's going to help me with my negotiation.
[00:05:39] Speaker A: That's the end goal.
[00:05:41] Speaker B: Right. The end goal is you're going to help me with my negotiation, and maybe we're going to focus on Batna or anchoring or Zopa or any of these negotiation concepts. And usually Chachibiti will know those terms, but sometimes I will tell it what those terms mean. Right. So it knows, giving it that asset, a definition of what we're working on.
[00:06:12] Speaker A: Like definition or criteria.
[00:06:14] Speaker B: Yeah, exactly.
And I'm kind of interested in how possibly, like, chain of density or some other prompt engineering idea could help this prompt. But the original prompt is giving it a goal and then telling it to create a scenario, giving it rules, basic rules. Like, you'll tell me the scenario, you'll ask me when I'm ready.
You'll go in and out of character and tell me when you've gone in and out of character, and then at the end, setting it up. So if I give it the keyword, give me a grade, that it will pause, and then it will give me a grade, give me an assessment of how I've been negotiating. So it's quite fascinating. So what will happen is you give it the prompt.
It asks you if you're ready. And this is like, with voice.
And then you say, ready. And then it'll lay out kind of a business deal between two people, something simple.
Like, recently, it gave a scenario where there was an electric car maker and then a battery maker, electric car battery maker, and the electric car maker wanted to buy the batteries from the battery maker.
They had some criteria, and they were trying to find a zone of possible agreement between the two in terms of price. Then you kind of had to play around with will there be extended warranties? What kind of after purchase service? There will be, et cetera. And I found that in experimenting with this and we experimented in class with this, that it does a fantastic job of negotiating. Yeah, I mean, it's really fantastic.
And you can go a long, extended time. So it will give you, let's say an offer. It will wait for your response, and once you give your response, it will then kind of based on that response, it'll give you another counteroffer or something related to the negotiation process.
And so at the end, once you're kind of tired or done with the whole process, you just give me a grade. Yeah, give me a grade. And then it will pause and end the negotiation. Well, you could end or pause in the negotiation however you want to do it.
You could just pause in the negotiation and ask it how you're, you know, give me a grade, and it'll tell you how you're doing so far. And then it'll ask you if you'd like to continue.
So it does a good job of. And you define what are the criteria that you want it to grade you on, defining that in the prompt. So basically, you want it to provide constructive feedback on negotiation skills, something along those lines, and then it does a really good job. Here's what you did. Good. Here's what I heard. Kind of like it's actively listening, right?
[00:10:01] Speaker A: Yeah.
[00:10:02] Speaker B: And here's what you could improve on.
And then usually at the end, it ends on a positive note, and it'll ask you if you want to continue or not. Then you can continue. So then based on the feedback that it's giving you, you could continue and make maybe changes that you want to make or whatever.
[00:10:28] Speaker A: So you mean chat JPT with voice could act like a teacher in negotiation and also a negotiator at the same time.
[00:10:39] Speaker B: Absolutely. Yes. That's perfect. So it's good. As an instructor, a coach, and a negotiator in class, I kind of wear those hats to, I'll say in character and I'll say out of character and try to play those roles. But chat GBT can do that really well.
[00:11:04] Speaker A: If you have two iPads and you ask one iPad to act like the teacher and the second iPad to be the negotiator, will it work?
[00:11:15] Speaker B: Let's do that after this podcast. I don't know, but that's fascinating.
So prompt one iPad like I just set up, and then prompt the other iPad to be the student who is engaging in the negotiation and see what happens if there'll be a crazy feedback loop. I'm kind of curious about that. Maybe we could film that, too.
[00:11:41] Speaker A: Sure.
[00:11:42] Speaker B: Or maybe we can even. Next time, we can bring the iPads into the podcast and ask them to perform. Yeah, that sounds like a great idea.
[00:11:53] Speaker A: Yeah, let's do it.
[00:11:55] Speaker B: Good.
So the final note is. That's all in English, right?
But just briefly, we experimented with Chinese, so in theory, and maybe this can be part two of this podcast.
[00:12:16] Speaker A: Yeah. I think the Chinese response is just, like, human, although it has an accent, but the content is very logical.
[00:12:26] Speaker B: So in theory, we would be able to prompt it to do this negotiation exercise in Mandarin.
[00:12:33] Speaker A: Yeah.
[00:12:35] Speaker B: So let's do that for next time. What do you think?
[00:12:38] Speaker A: Okay, sure.
[00:12:41] Speaker B: Thank you.
[00:12:42] Speaker A: Thanks a lot.
