TRANSKRYPCJA 🔗
WEBVTT
00:00:12.640 --> 00:00:15.119
Hey everyone and welcome to the Amplitude of Tech Podcast.
00:00:15.199 --> 00:00:18.320
I'm Sean Kordner, Chief Marketing Officer of Amplix.
00:00:18.879 --> 00:00:22.160
This is the Inside Amplix edition of the podcast.
00:00:22.399 --> 00:00:23.039
What does that mean?
00:00:23.199 --> 00:00:32.079
We're going to bring you thought leadership from Amplix employees, actionable intel and insights that we've learned by doing thousands and thousands of technology transformation projects.
00:00:32.240 --> 00:00:34.079
Hope you enjoy the episode.
00:00:37.920 --> 00:00:40.399
All right, Jeremy Tam, welcome to the podcast.
00:00:40.719 --> 00:00:41.520
It's good to be here.
00:00:41.600 --> 00:00:42.399
Thank you so much.
00:00:42.719 --> 00:00:46.719
And this is an inside Amplix edition of the Amplitude of Tech podcast.
00:00:47.039 --> 00:00:49.280
You are inside Amplix, you're an Amplix employee.
00:00:49.359 --> 00:00:51.679
Why don't you talk a little bit about who you are and what you do here?
00:00:52.079 --> 00:00:52.640
Absolutely.
00:00:52.799 --> 00:00:58.479
So I am a senior implementation engineer and the service delivery team.
00:00:58.719 --> 00:01:01.359
I mainly focus on implementing Zoom products.
00:01:01.520 --> 00:01:05.519
So everything from Zoom Contact Center to Zoom phone.
00:01:05.599 --> 00:01:12.400
And of course, one of my bread and butters is a Zoom virtual agent, the AI product offerings from Zoom.
00:01:12.799 --> 00:01:23.120
So what exactly do you do if some you know one of our clients, let's say, uh wants to implement uh ZVA, uh Zoom virtual agent, uh, what's your role in that process?
00:01:23.599 --> 00:01:38.239
Well, I do everything from discovery, uh so understanding what they actually want the bot for, what kind of business problems they're trying to resolve, because that's actually far more important than what a lot of people uh give it credit for.
00:01:38.480 --> 00:01:45.760
A lot of people might say, Oh, I want an AI agent, but they don't know what exactly they want the AI agent to do.
00:01:45.840 --> 00:01:48.959
They just know that there's a fancy technology out there that they want.
00:01:49.120 --> 00:01:52.159
Uh, but you know, AI is a tool.
00:01:52.480 --> 00:01:54.560
The business problem is what the focus should be.
00:01:54.719 --> 00:02:10.240
So I start from uh understanding what their needs are through discovery, all the way through implementing, conducting testing, QA uh to go live and um even some after go live support afterwards.
00:02:10.560 --> 00:02:17.599
So you know, we've been uh I would say we're three years into the brave new world of AI, right?
00:02:17.759 --> 00:02:23.120
It's obviously been around a lot longer than that, but um I think we're experiencing it in a different way now.
00:02:23.439 --> 00:02:32.319
And uh I remember when it was first really starting to become a ubiquitous conversation amongst enterprises and and really uh amongst
Load full transcript
consumers as well.
00:02:32.400 --> 00:02:34.639
It really happened all kind of at once, it feels like.
00:02:34.800 --> 00:02:39.439
Um the conversations were hey, I need this, I need this AI.
00:02:39.520 --> 00:02:40.479
Well, what do you need it to do?
00:02:40.639 --> 00:02:44.639
Well, I I don't know, I just know I need AI, or my board's telling me I need AI.
00:02:44.800 --> 00:02:46.560
Uh it sounds like you're still experiencing that.
00:02:46.719 --> 00:02:49.439
Like, what where do you think people are at this point?
00:02:49.520 --> 00:02:58.719
Like, are people a little bit more uh focused in what they're trying to accomplish with with AI, or are they really just saying, let's throw some AI at something and see if it sticks?
00:02:58.960 --> 00:03:00.400
I'm seeing a bit of both.
00:03:00.639 --> 00:03:08.319
There are some you know prospects or some clients that come to me having a very strong understanding of AI, right?
00:03:08.479 --> 00:03:23.840
And you know, when when I say AI, I mean in the last few, last couple of years, obviously AI is a very wide term and it has a very long history, but we're talking about like the explosion of large language models that came out of like Chat GPT, right, from a couple of years ago.
00:03:24.080 --> 00:03:39.199
Uh, some prospects, especially the ones that are larger enterprise teams that have you know some very technical people on their team, they understand what the limitations of large language models and they have a good idea of what it can do for their brand.
00:03:39.360 --> 00:03:49.120
But we also have the things on the flip side where you know we have smaller commercial accounts where the people implementing these products are not your technical people.
00:03:49.280 --> 00:03:55.680
They might be the manager of support, managers of you know, um, you know, marketing and sales, right?
00:03:56.319 --> 00:04:02.639
Those conversations tend to require a little bit more um more focus, right?
00:04:02.879 --> 00:04:08.719
Trying to so educating them on what are the limitations, what it can and cannot do, what it's good for.
00:04:08.960 --> 00:04:15.360
Um, and you know, usually that conversation starts with, once again, what kind of business problems you're trying to resolve.
00:04:15.520 --> 00:04:23.120
What are the challenges, what are the friction points that your team is experiencing where AI can uh in you know, inject itself.
00:04:24.879 --> 00:04:38.319
It it seemed like the early promise of AI, at least the perception from uh the boardroom and and maybe the C-suite was that this was gonna be a huge cost saver in the sense that it was gonna save on labor costs primarily.
00:04:38.399 --> 00:04:40.800
I I you know there was the productivity conversation as well.
00:04:41.040 --> 00:05:02.720
But I think originally, and and really a lot of where the the uh the doom and and the current um undercurrent of concern, uh fear, hate of AI that we're experiencing in society right now may
come from that early, you know, those early days of the promise of AI uh replacing human beings in the workplace.
00:05:02.879 --> 00:05:06.959
Uh are you seeing people still trying to take that approach?
00:05:07.600 --> 00:05:08.240
I am.
00:05:08.800 --> 00:05:14.959
And because, you know, at the end of the day, that is one of the value propositions of AI, right?
00:05:15.120 --> 00:05:31.839
If set up correctly, a chatbot, you know, that is you know well trained, uh use cases well defined and properly programmed, it can do a lot of the uh functions that a tier one or tier two uh support agent is capable of doing.
00:05:32.079 --> 00:05:53.040
You know, nowadays, you know, thanks to all the advances with the integrations that these platforms uh can have, can do, are capable of, we can you know uh connect the AI to um the back end of CRMs so that even you know customer data that needs to be updated can be done autonomously by AI.
00:05:53.199 --> 00:06:00.399
So we're seeing um you know a couple years ago where AI can address rather simple questions, right?
00:06:00.480 --> 00:06:02.240
You know, tell me where the store location is.
00:06:02.399 --> 00:06:03.040
Very simple.
00:06:03.199 --> 00:06:04.879
Here's a question, here's an answer.
00:06:05.040 --> 00:06:08.480
But now it's actually capable of doing a lot more than that.
00:06:08.639 --> 00:06:22.240
It can take values from your CRM after asking a question, and based off of what is what information you know uh pertains to that customer or client, um offer out different solutions, right?
00:06:22.399 --> 00:06:43.600
So it is actually getting smarter, and of course, as these products get smarter, there's less need for overseas BPO agents, for example, to fill up that seat and answer that uh to handle that volume because AI, you can have a single bot set up and a hundred calls come in at the same time, and all you need is that one bot, right?
00:06:43.759 --> 00:06:46.879
So simultaneously addressing call volumes, right?
00:06:46.959 --> 00:06:48.800
That's a huge advantage of AI.
00:06:48.959 --> 00:07:00.399
Um resolving these tier one, tier two questions are excellent, huge value, but oftentimes people forget you still need some humans to some extent uh today.
00:07:00.639 --> 00:07:05.360
And AI on the flip side can actually benefit those, uh, that cohort as well.
00:07:05.600 --> 00:07:19.439
So think of um, you know, again, as an example, a support agent handling a very complex issue, like say, for example, if you have a smart light in your background over there and you're having problems with it, there's some back and forth that needs to happen.
00:07:19.680 --> 00:07:23.759
Well, now that AI agent can handle all of that back and forth.
00:07:23.920 --> 00:07:31.439
So by the time the case gets to your human agent, all they need to do is action that replacement, action that refund.
00:07:31.600 --> 00:07:37.120
You know, so it reduces their handling time significantly, if not just
outright resolving the issue.
00:07:37.600 --> 00:07:44.639
And now we have agentic AI, I think, becoming more ubiquitous in uh certainly in the CX stack.
00:07:44.879 --> 00:07:48.959
Uh so that can really complete the action, right?
00:07:49.040 --> 00:07:56.879
So somebody, uh, you know, an AI LLM agent determines that a refund is appropriate.
00:07:57.040 --> 00:08:02.240
And so the agent can act on that refund uh refund, assuming that the integrations are there.
00:08:02.399 --> 00:08:06.079
Uh so it sounds like we can just get rid of our customer service reps.
00:08:06.240 --> 00:08:07.439
Is that what you're here to tell us?
00:08:07.759 --> 00:08:07.839
No.
00:08:08.319 --> 00:08:10.480
Uh well, yes and no.
00:08:10.720 --> 00:08:18.720
So I think you know the size of these you know uh support agents will probably shrink down.
00:08:19.040 --> 00:08:43.600
Uh and I think that's good and bad because you know, uh, I'm sure before um, you know, even before AI, for the most part, I think when I call a business, you know, support online, half the time I get matched with someone who can't really help me all that much because you know their answers are inconsistent, right?
00:08:43.759 --> 00:08:45.200
Experiences are inconsistent.
00:08:45.279 --> 00:08:47.679
I call my credit card, you know, for help.
00:08:47.919 --> 00:08:54.879
It really comes down to who I get matched with with to that you know kind of um determines how satisfied I am with the result.
00:08:55.279 --> 00:09:15.039
But with AI, you know, a lot of those consistency now becomes a lot easier to manage, even though yes, large language models are inconsistent in nature of what how it responds, but what the you know, the core of what it needs to say is still remains, you know, largely the same, right?
00:09:15.600 --> 00:09:17.440
Because it's trained on the on the same model.
00:09:18.080 --> 00:09:34.559
So yes, uh it can replace you know the lower performing agents, but at the same time provide a lot of tools and uh increase in efficiency for your higher performing uh agents as well.
00:09:34.639 --> 00:09:36.799
So there's you know two sides to that point.
00:09:37.519 --> 00:09:46.720
Yeah, you know, we're we're talking uh at least I'm speaking uh primarily from the perspective of uh you know an AI agent, a frontline AI agent, right?
00:09:46.879 --> 00:09:58.000
But there's certainly a ton of use cases on the back end for for helping agents, helping humans uh be more um successful and efficient in the job that they're doing, right?
00:09:58.240 --> 00:10:09.840
But I think what I'm hearing you say is that these agents are good at discrete tasks, um especially when it's a deterministic task.
00:10:10.000 --> 00:10:17.679
So if this then that, that's something that can be determined, um, versus it's not great at judgment.
00:10:18.000 --> 00:10:26.639
So when it's when it's probabilistic, um, it's not really it's not weighing the pros and cons and making a judgment call.
00:10:26.720 --> 00:10:28.879
It's playing the probabilities, right?
00:10:29.200 --> 00:10:54.799
Yes, and you actually bring up a really good point because um managing expectations of what AI, such a broad term, but you know, AI agents um and you know, across all the different vendors that offer uh an AI agent, whether it's level AI, Zoom, Zoom virtual agent, or even you know, um you know, a plethora of other AI out there.
00:10:55.360 --> 00:11:00.720
All of their models right now aren't at a level that you see on the news.
00:11:00.879 --> 00:11:10.080
So when you pull up the news, you're seeing all the headlines of like a gentic AI by these, you know, by the largest players, whether it's Facebook, Google, right?
00:11:10.159 --> 00:11:18.720
They have these, everyone's going for um, you know, a generative AI, uh, I mean a uh general artificial intelligence, right?
00:11:18.799 --> 00:11:21.279
Something that's so smart that it can replace humans.
00:11:21.600 --> 00:11:31.759
The AI models that are available to vendors across the board for you know, creating these chatbots for businesses, they are far from that.
00:11:32.639 --> 00:11:36.320
Those are not uh commercially available AI models.
00:11:36.559 --> 00:11:47.519
And you know, managing the expectation from clients who read those headlines and think, oh, you know what, I can absolutely replace my entire IT support team now uh using this.
00:11:47.840 --> 00:11:55.759
But the models that are currently available, and I stress currently, because who knows in a couple of years what that uh what that would look like.
00:11:56.000 --> 00:12:08.639
Currently, models are very good at retaining a load of information and essentially uh using that as its only source of truth.
00:12:08.879 --> 00:12:17.600
But it's not yet at a place where it can think outside of the box, it's not yet at a place where it can, you know, replace human intuition.
00:12:17.679 --> 00:12:24.240
Uh, you know, someone says that, oh, my computer runs, can turn on, that it runs hot, what does that mean?
00:12:24.480 --> 00:12:24.799
Right.
00:12:24.960 --> 00:12:33.279
So unless you have an article defeat the bot to say explicitly, if the computer doesn't turn on, if it's hot, try these steps.
00:12:33.600 --> 00:12:37.919
If that information doesn't exist, then what it will do is hallucinate, right?
00:12:38.000 --> 00:12:40.799
If we force it to provide a response.
00:12:41.039 --> 00:12:53.200
So I am, you know, encountering some uh wouldn't say issues, some challenges with prospects, uh potential clients who say, you know, I want I've I've saw that it's capable of doing this on the news.
00:12:53.519 --> 00:13:01.200
Write me a bot that can replace my you know um my Wi-Fi um troubleshooting team.
00:13:01.840 --> 00:13:03.039
It's not there yet.
00:13:03.200 --> 00:13:07.120
Uh and a lot of people think, you know, it should be.
00:13:07.840 --> 00:13:19.120
Uh they ask, can your bot just you know, uh ingest all of our
past experiences and be able to therefore afterwards think outside of the box?
00:13:19.279 --> 00:13:19.600
No.
00:13:19.840 --> 00:13:25.919
It will take your previous cases, it will say the same things, but still saying the same things.
00:13:26.080 --> 00:13:32.639
It's not thinking outside its box, it's not being, it's not uh it hasn't reached the human intelligence level just yet.
00:13:32.879 --> 00:13:33.039
Yeah.
00:13:33.360 --> 00:13:35.200
I mean, famously we saw Karna.
00:13:36.320 --> 00:13:49.039
They had, I think, 500, 700 agents, and they decided that the uh the AI was good enough, and so they fired them all overnight, and predictably they had to hire most of them back, if not all of them back, right?
00:13:49.120 --> 00:13:59.600
So there's yeah, there's plenty of stories and evidence out there, and I like the distinction that you're making of you know, enterprise capability versus um frontier model capability, right?
00:13:59.679 --> 00:14:07.120
And and enterprise capability is really good, but it's definitely not the same as frontier model capability.
00:14:07.200 --> 00:14:10.960
Uh and you know, you invoked AGI, which is the holy grail.
00:14:11.039 --> 00:14:16.320
I I think uh Jensen Huang this week said that AGI is here, that we have it.
00:14:16.399 --> 00:14:26.159
And you know, I I tend to not believe any of these guys that have um you know their their income and their uh stock values tied to uh the things that they promise and say.
00:14:26.320 --> 00:14:35.360
So I'm I'm wary of that statement, but um it seems like we're we're getting somewhere, and that's all going to filter down to the enterprise level pretty soon as well, right?
00:14:35.679 --> 00:15:03.679
Yes, I I think that's pretty much inevitable because you know behind the scenes, you know, we have all these large companies, you know, playing uh around models announcing, you know, the uh the coming of AGI, but a lot of people don't realize that there's a massive grass move uh grassroot movement of AI that goes on, you know, not at the highest level, but just of you know your everyday well people who are tech enthusiasts building their own models, right?
00:15:03.840 --> 00:15:14.639
There's um you know, uh I think they're um I think they announced that they were acquiring Hugging Face for a couple billion dollars or you know, something really high.
00:15:14.799 --> 00:15:24.879
But you know, prior to that acquisition, that is sort of like you know, um you know, um a battleground of grass-root, you know, uh model makers.
00:15:25.039 --> 00:15:30.000
Um but you mentioned a point earlier about uh cost cutting, right?
00:15:30.320 --> 00:15:36.879
And how these uh bots can potentially reduce couple uh overall operation costs.
00:15:37.120 --> 00:15:42.960
Well, and you know, they fired all those people and then they realize tokens are expensive, right?
00:15:43.120 --> 00:15:51.759
Every single time you or one of your engineers put in a query to have it uh to
have an AI, you know, code something for them, that costs money.
00:15:52.000 --> 00:15:59.519
And I think that uh cost realization is starting to bubble up to a lot of decision makers and realizing, hey, you know what?
00:15:59.679 --> 00:16:02.240
We're not actually saving all that much.
00:16:02.320 --> 00:16:08.159
If anything, we're running a risk of even uh having it become more expensive, right?
00:16:08.480 --> 00:16:25.600
Because uh there are some challenges that um you know two engineers, two and two, you know, level two engineers can resolve that might take a higher level one, a couple dozen of queries, or if not more.
00:16:25.759 --> 00:16:29.039
That's like a very conservative guess uh to be able to resolve.
00:16:29.120 --> 00:16:29.919
And you know what?
00:16:30.080 --> 00:16:33.759
The latter actually costs more to some to in certain scenarios.
00:16:34.159 --> 00:16:35.679
So I I saw an X thread.
00:16:35.840 --> 00:16:39.759
Now I'm not an X user because I'm an old man and I use Instagram.
00:16:39.840 --> 00:17:10.799
Um not old enough to use Facebook anymore, but um but so uh so I saw screenshots of an X-thread on Instagram, to be specific, and it was something in the fact of you know, some AI leader, uh, you know, company that has an AI product talking about how uh right now they're recommending that people start to look at low-level AI use cases and offshoring them to human beings because it's actually uh less expensive.
00:17:10.960 --> 00:17:14.640
And someone under that was like, oh, look at this guy, he just discovered employees.
00:17:14.799 --> 00:17:20.480
You know, the this whole movement was to try to reduce labor.
00:17:20.640 --> 00:17:23.759
Uh-and I'm gonna take his hand for just a second.
00:17:23.920 --> 00:17:40.559
Um, the idea was to reduce labor, but what I talk to a lot of uh technology leaders because of my job and because of the podcast, and I don't hear any of them on the provider side, on the client side, none of them say I want to eliminate employees.
00:17:40.799 --> 00:17:45.680
What they say is I want to redeploy employees and I want to reduce the need for future employees.
00:17:45.839 --> 00:17:52.559
I want to be able to scale without having to scale linearly my my labor costs.
00:17:52.880 --> 00:18:00.960
So if you can if you can get a little bit more scale with the same number of employees, that's a good economic model for them, right?
00:18:01.119 --> 00:18:10.319
So so the original idea was to save on labor costs, but I think now the idea is how can we make people better?
00:18:10.559 --> 00:18:20.480
How can we pair people with technology um in order to have them be more productive, create better user experiences, or I'm sorry, uh customer experiences.
00:18:20.720 --> 00:18:22.559
Um is that what you're seeing now?
00:18:22.640 --> 00:18:29.519
Do you do you see that shift where the the thought process is more how can we leverage this for our people as opposed to instead of our people?
00:18:30.000 --> 00:18:31.200
I see it all the time.
00:18:31.440 --> 00:18:32.000
Absolutely.
00:18:32.160 --> 00:18:38.640
And you know, I think a lot of the vendors out there are very aware of this aspect of AI, right?
00:18:38.799 --> 00:18:52.559
So, you know, uh I work a lot with uh Zoom products, and so you know, uh while they do have a agentic AI product, like Zoom, Zoom virtual agent, they're also focusing on the human side of it as well.
00:18:52.799 --> 00:19:04.480
How can they uh create tools within the Zoom platform that, uh to your point, improves the efficiency of the humans rather than you know just focusing on uh on the bot, right?
00:19:04.559 --> 00:19:08.559
So there are a lot of bot features that we you know don't really think about.
00:19:08.720 --> 00:19:12.799
For example, well, large language models being such an important thing.
00:19:13.039 --> 00:19:26.640
But why isn't that a why is it that the focus is on building a bot that can answer questions as and not on whether an AI can populate those answers for live humans during a live call, right?
00:19:26.799 --> 00:19:32.000
And that's one thing that uh Zoom, you know, a Zoom product does is called AI Expert Assist.
00:19:32.240 --> 00:19:33.680
Uh, and it's very powerful.
00:19:33.839 --> 00:19:43.519
It can also be hooked up with you know different CRMs, so it pulls the right information uh at the right time uh for your agent, increasing efficiency and uh reducing handling time.
00:19:43.599 --> 00:19:47.759
Um so definitely there's a huge shift towards that um from both fronts.
00:19:47.920 --> 00:19:49.839
I think people are very well aware of that.
00:19:50.000 --> 00:20:00.160
Um and you know, there is some, you know, in the past couple years there's been major shifts in the uh in the overall labor market for support, right?
00:20:00.319 --> 00:20:10.640
Because of AI, a lot of you know, you hear, you read articles about, you know, in the Philippines, they're now seeing a lot of their uh tier one support agents now out of work.
00:20:10.880 --> 00:20:18.240
But you know, at the same time, that means that their labor costs, you know, at heart are pushed down because the demand has gone down.
00:20:18.400 --> 00:20:28.720
Now, as a result, now all of a sudden it might actually be cheaper now because um to go with the BPL route because of that downward uh pressure on on their price.
00:20:29.039 --> 00:20:58.559
So, you know, we live in a very interesting time uh where the smarter AI gets, you know, some um the less actual human agents might be required, but it'll be a shift in focus to hiring quality agents and giving them the tools with AI to make them even better, as opposed to now just blindly, you know, uh we have a you know, uh Christmas is coming, our volume is gonna go up, let's just hire 10 more hit count, something like that.
00:20:58.960 --> 00:21:12.559
Yeah, I I have a um a marketing peer, another CMO that's in the space that I have a weekly
meeting, um, sorry, a monthly meeting with her just to just to talk about how we're using AI and we kind of bounce things off of each other.
00:21:12.640 --> 00:21:13.599
I've learned a lot from her.
00:21:13.680 --> 00:21:19.200
She's um she's definitely ahead of where uh we're at in terms of leveraging AI in the marketing department.
00:21:19.359 --> 00:21:24.000
And you know, she was saying she's got um 11 employees in the marketing department that work under her.
00:21:24.240 --> 00:21:31.599
And um as they grow, the plan is to not add headcount, it's to add capabilities.
00:21:31.759 --> 00:21:42.400
So each one of those 11 employees is gonna become a supervisor or a manager of a team of AI gents and bots that will help them execute and do work for them.
00:21:42.480 --> 00:21:52.880
And they are the human in the loop that manages the process and does the quality control and just just you know basically acts as the manager of you know AI employees.
00:21:53.119 --> 00:21:56.079
Um I think the contact center may end up the same way.
00:21:56.160 --> 00:22:09.519
I've heard a few contact center leaders talk about this of how you know the Eventually some of their agents will uh you know, their their first level agents will become supervisors of multiple agents, and so they just get the escalations.
00:22:09.839 --> 00:22:10.160
Yeah.
00:22:10.400 --> 00:22:11.519
No, absolutely.
00:22:11.759 --> 00:22:21.759
Uh and you know, the amount of tools that are get being created and given to uh agents now are just becoming so incredibly powerful.
00:22:22.000 --> 00:22:25.119
So, you know, languages for for an example, right?
00:22:25.200 --> 00:22:35.359
If the um with large language models, immediate real-time translations are becoming more accurate, faster, and better than it's ever been uh before, right?
00:22:35.519 --> 00:22:46.319
So we used to all use Google Translate, but even nowadays, if I uh were to translate um a larger document or whatever, I would use uh a large language model.
00:22:46.559 --> 00:23:17.920
So, you know, they've Zoom, for example, has applied this to the contact center experience where me, a single uh agent in California, uh and I only speak English, I can now support every region that Zoom supports, you know, even uh Farsi as a random um example, where you know the customer can be chatting with me in Farsi, I see it in English, I respond in English, and they see it in Farsi.
00:23:18.079 --> 00:23:33.519
Now all of a sudden you can just have one small group of um you know good quality support agents located in the States, and now all of a sudden you have a global support team that is capable of um you know handling your global volume.
00:23:33.680 --> 00:23:37.119
It's um actually quite it's quite amazing to see.
00:23:37.440 --> 00:23:38.160
It's incredible.
00:23:38.240 --> 00:23:47.359
I saw a video the other day of a uh body cam footage of a cop pulling someone over and he only spoke Mandarin, and she cannot speak Mandarin.
00:23:47.519 --> 00:23:55.039
And so she was able to through the body cam uh connect to an AI agent that did the translation real time for them.
00:23:55.200 --> 00:24:00.480
So she spoke to it, it spoke out in Mandarin, he responded, and it translated for her.
00:24:00.559 --> 00:24:02.640
And they did the whole uh traffic stop that way.
00:24:02.720 --> 00:24:04.799
And I thought, wow, what a fascinating use case.
00:24:04.960 --> 00:24:17.200
And and you know, and that's the kind of thing that it's not just utilitarian or practical, but it's also that's something that could save a life, it could avoid miscommunication and misunderstanding.
00:24:17.359 --> 00:24:25.200
And so I I think that this will definitely it has the capability to be a very positive uh thing in society if we're gonna be able to do that.
00:24:25.599 --> 00:24:26.000
Absolutely.
00:24:26.160 --> 00:24:36.000
I can't wait to uh so I think it was just yesterday or the day before where we saw the new um airpods, right with you know, the auto-translation.
00:24:36.079 --> 00:24:46.240
I can't wait to go on my next vacation with one of those things and just see what uh you know kind of you know, I think about it, kind of gross to give a stranger your earbud.
00:24:46.640 --> 00:24:49.279
Yeah, but you know, it's getting there.
00:24:49.519 --> 00:24:50.720
Yeah, it's getting there.
00:24:50.880 --> 00:25:01.599
Um I I travel a lot of places um that are Spanish speaking, so I I learned just enough Spanish to order things at the bar, but I I can't get around uh too well other than that.
00:25:01.759 --> 00:25:04.000
So that kind of thing would be very useful for me.
00:25:04.240 --> 00:25:06.079
Well, you got the important part down.
00:25:06.720 --> 00:25:10.640
That's why I started Spanish is you know difficult because of the speed too.
00:25:10.880 --> 00:25:11.200
Yeah.
00:25:11.440 --> 00:25:11.680
Yeah.
00:25:11.839 --> 00:25:16.720
Uh it's not just the language itself, it's just the speed at which people talk.
00:25:16.799 --> 00:25:20.799
Uh well, it you know, it's dialogues too, and that's something that AI solves for as well.
00:25:20.880 --> 00:25:26.160
I I was at an event, I think it was um I it might have been Pali, actually.
00:25:26.400 --> 00:25:37.279
Pali AI, an event that they had, and they were demonstrating the uh regional dialect and accent capabilities of the platform.
00:25:37.440 --> 00:25:41.519
So you could call in to the same AI agent.
00:25:41.599 --> 00:25:45.279
Uh, if you're calling from the south, it'll have a southern accent, but not just southern.
00:25:45.440 --> 00:25:55.599
If you're calling from Georgia, it'll have that kind of accent versus say uh, you know, a Louisiana accent, or or uh if you were calling in from Chicago, it had a Chicago accent.
00:25:55.680 --> 00:25:57.359
And that's incredible to me.
00:25:57.440 --> 00:26:01.920
You you wouldn't even think about it, but that's the kind of thing that puts people at ease a
little bit.
00:26:02.160 --> 00:26:03.359
Yes, definitely.
00:26:03.519 --> 00:26:22.319
Especially, you know, uh, because well, people want to be heard and they want to feel like they're speaking to someone who understands them, even if it's an AI, like off obviously it is an AI, but it's those little things that can really improve the uh customer experience.
00:26:22.480 --> 00:26:34.160
And I'm glad you brought that up because when I first, you know, when um this new large language model version of uh Zoom Virtual Agent got released, I think there were about a dozen of voices, right?
00:26:34.720 --> 00:26:39.119
Now it has gone to upwards of 80 different voices and accents.
00:26:39.279 --> 00:26:49.359
And even more powerful is if your accent is not on the list, let's say you have a deep, deep Louisiana Southern accent that's not uh available as a voice.
00:26:49.599 --> 00:26:55.920
Well, with you know, you can record your own voice now and have that be turned into an AI voice.
00:26:56.000 --> 00:27:02.480
So it will pick up your inflections, your accent, uh, and let that be the voice of your brand.
00:27:02.640 --> 00:27:10.799
Um, assuming you know your bond is only servicing that area of people with those heavy accents, but you know, uh that is already a thing now.
00:27:10.960 --> 00:27:15.279
Um, and it's available in commercial products, even in things like Zoom.
00:27:15.599 --> 00:27:22.559
Yeah, I I would bet that eventually it'll pick up the accent of the person that's calling in and could replicate it real time.
00:27:22.880 --> 00:27:24.960
Um so quick quick shift.
00:27:25.039 --> 00:27:29.680
I'm just gonna share uh uh an embarrassing personal story that happened just last week.
00:27:30.400 --> 00:27:35.119
Uh and and we can dissect this a little bit, and then I think that'll be a good place to wrap up.
00:27:35.519 --> 00:27:43.119
Um so I'm at a point in my life where I am uh I'm doing apartment living, uh luxury apartment living.
00:27:43.200 --> 00:27:46.000
I don't want to have to cut the lawn, I don't want to have to do maintenance.
00:27:46.160 --> 00:27:51.599
Um I sold a house a few years ago, and I'm I'm I'm I'm done with that for right now at this point in my life.
00:27:52.000 --> 00:27:58.799
Um my girlfriend and I, we moved into uh a new apartment uh maybe three weeks ago, something like that.
00:27:59.039 --> 00:28:00.480
Rent was due on the first.
00:28:00.640 --> 00:28:09.759
I tried to log into the portal and I couldn't get in because my last place had the same portal, and so there was some sort of disconnect that I wasn't able to access this account.
00:28:10.319 --> 00:28:18.960
So um I got a uh a text message like the next day because um I was trying to sort it out and then I got distracted.
00:28:19.039 --> 00:28:24.480
And so I got a text message saying, Hey, um, want to know if you needed any help with paying the rent.
00:28:25.119 --> 00:28:32.319
This is this is Jane from, you know, I won't say the
name of the place, but this is Jane from wherever, and wanted to know if you needed any help.
00:28:32.400 --> 00:28:33.599
And I'm like, Yeah, actually I do.
00:28:33.680 --> 00:28:34.240
I can't get in.
00:28:34.319 --> 00:28:35.599
And I explained the whole thing.
00:28:35.839 --> 00:28:37.599
Oh, I'm sorry to hear that, whatever.
00:28:37.839 --> 00:28:40.880
So um Jane said, I'll get right back to you.
00:28:40.960 --> 00:28:52.480
So, long story short, is this thing over the course of eight days, I thought was a human being, was trying to help me get into this portal.
00:28:52.559 --> 00:28:59.680
And I'm thinking that they know that I'm working with them and that's not my fault, so they're not gonna count this as late and charge me a leaf fee or like so.
00:28:59.759 --> 00:29:02.160
I'm not really worried about it, and I'm just engaging.
00:29:02.400 --> 00:29:08.319
And it was an AI, and I had no idea that, and I'm in this space, right?
00:29:08.559 --> 00:29:18.160
But it's we talk about this a lot about like the emotional state that people are in, and and this is uh um true for social engineering from a cybersecurity perspective as well.
00:29:18.240 --> 00:29:25.359
Is you know, if if you're in a state where you're a little anxious, which I was because it's a new place, I don't want to be late on my rent, I'm having trouble with this thing, right?
00:29:26.079 --> 00:29:27.599
Probably should have looked at it beforehand.
00:29:27.680 --> 00:29:30.400
I just assumed that the technology was gonna work, whatever.
00:29:30.559 --> 00:29:34.079
Um, so I I fell for it, and this thing did not help me at all.
00:29:34.160 --> 00:29:41.039
And then I actually got an email from a human being, and I was like, oh, um, but I've been working with Jane in the leasing office.
00:29:41.119 --> 00:29:43.759
And they're like, we don't have a Jane in the leasing office.
00:29:44.079 --> 00:29:57.680
So it turns out it's this platform called ResiDesk, and it and moral of the story of this is the the two things that I think are a lesson for enterprises on this are number one, disclose when you're using AI.
00:29:57.839 --> 00:29:58.079
Yes.
00:29:58.319 --> 00:30:00.480
To avoid any doubt or any confusion.
00:30:00.640 --> 00:30:17.680
And number two, how is this thing not trained that if someone's having trouble paying their rent, especially past a certain point, to kick it off to a live human being, how is there no communication to the actual property that's expecting the rent or that it's a right?
00:30:17.839 --> 00:30:17.920
Right.
00:30:18.240 --> 00:30:29.200
So I think that is a good example of I'm you know, I'm not an idiot and I'm I'm not someone who's unaware of AI bots, right?
00:30:29.359 --> 00:30:36.640
And for people to make that mistake, um, I think what it tells me is that you definitely can't replace human beings yet.
00:30:37.119 --> 00:30:43.200
People for certain things, they want to speak to a human, they need to speak to a human.
00:30:43.279 --> 00:30:52.160
And if you are going to use AI, you really have to architect it in a way that considers the human on the other end of it and think about the experience that they're gonna have with it.
00:30:52.319 --> 00:30:52.960
What are the reasons?
00:30:53.039 --> 00:30:53.920
What are the use cases?
00:30:54.000 --> 00:30:55.200
What are the outcomes?
00:30:55.359 --> 00:31:06.160
Um, and and make sure that you're handling all that appropriately instead of just letting it kind of live within this AI and and never escaping if if it doesn't get resolved.
00:31:06.400 --> 00:31:06.799
What do you think?
00:31:08.000 --> 00:31:20.720
And you know, the remember when at the very beginning of this uh of this chat, I mentioned it's the AI is a tool and it is more of the business problems you're trying to resolve.
00:31:20.880 --> 00:31:22.400
That's should be the main focus.
00:31:22.720 --> 00:31:26.960
It sounds to me like they went the other way with this particular bot.
00:31:27.119 --> 00:31:31.359
Uh, first of all, not announcing, not letting you know that it's an AI.
00:31:31.519 --> 00:31:38.960
Like they pro obviously programmed it to sound as much like a human being, because to your point, you want to talk to a human being.
00:31:39.119 --> 00:31:42.960
They don't want the cost of letting you talk to a human being, so they put in a bot there.
00:31:43.039 --> 00:31:49.359
But it can't do all the business uh resolution things that you that it should be able to do.
00:31:49.519 --> 00:31:52.240
So it's kind of like a stop gap right now.
00:31:52.400 --> 00:31:59.279
Um, first of all, in southern states now is a requirement at the bottom lets you know that it's an AI, right?
00:31:59.680 --> 00:32:09.680
And uh second of all, if it can't do those um carry out those functions and it's just kind of left it in limbo, it means it wasn't properly programmed.
00:32:09.920 --> 00:32:24.000
And so it was the that tells me the focus was just to let someone talk to someone that's they think it's a human without actually providing you know all the tools the bot needs to actually carry out and execute those uh resolutions.
00:32:24.240 --> 00:32:28.799
Sounds like that uh problem from A to Z.
00:32:29.200 --> 00:32:30.079
Yeah, absolutely.
00:32:30.240 --> 00:32:33.759
Well, I'm gonna cold call them and put them in touch with you, Jeremy, so you can set them straight.
00:32:34.160 --> 00:32:37.359
Uh Jeremy Tam, really appreciate your time and expertise today.
00:32:37.440 --> 00:32:37.920
Thanks so much.
00:32:38.160 --> 00:32:39.440
Of course, it's a pleasure.
00:32:39.519 --> 00:32:40.640
Thank you so much.