Melissa Valdez

Jam Mayer

Resources

Melissa Valdez

Perplexity AI

Hugging Face

Melissa

Experience is everything, right? A lot of times when it comes to new technology, it's easy to start building and not necessarily address a user need with what you're building.

Oftentimes it can make things more frustrating for users. And we were very aware of that being in the chatbot space, knowing that a lot of people have come from experiences where they're just so frustrated. , They wanna speak to a human immediately. Don't make me do this. You know, I've experienced how bad this can be. But some of the best feedback that we would share back at executive meetings were examples of someone saying, this is the best chatbot I've ever worked with.

And the big difference between a frustrating experience and a smooth and delightful experience, which is what we strove to deliver, would be how well the interface allows you to solve your problem. And so it's very important to recognize what the users are trying to do. Like their actual intent when they comment to make that connection with the company.

And so oftentimes, you have a question about one thing and the bot isn't trained for it. So it's just this exercise in frustration. And so making sure that we caught a lot of those fringe cases and then the cases where we didn't have an answer provided, to have really great exception handling.

So if we didn't know what they were saying, or we couldn't answer their question, or we didn't understand them, all of those cases were handled extremely well and you never ran into a loop or a dead end in the conversation. And so that type of thinking through the experience was absolutely critical. The other thing that I think really augmented from kind of a basic bot that, has, you know, three main things that can do well and everything else it does poorly.

And some of the bots that we were building, we were looking at integrations with, like it was a really a full stack system. We had integrations with backend systems and CRMs. We had integrations with things like address validation, APIs and any other type of technology piece that humans are using when we fill out forms and make do our work. We were giving all that same power to the bot and that allowed it to do things a lot faster than humans can often do them. And just deliver that really delightful experience if you could come through and actually get your question answered faster than you thought you were going to.

Like a delightful experience is a 101, right? And so that was a really big transition. So exception handling, I would say, is a big one. Integration with other technologies so that it's not just the chatbot standing alone, that's another one. Another good principle of design was always have an escape route.

So if it was on business hours, being able to escape and talk to a human. If we were outside of business hours where no one would be able to chat with the user, we would offer them the opportunity to get a call back or we could offer them you know, give me your information now, I'll send it to somebody and they'll contact you tomorrow during business hours.

Right? So it was kind of a way to take a number and we'll get back to you rather than you having to come back tomorrow and remember to restart the conversation. And when we did kind of take all that information with the conversation we just had with the user and pass that on to the agent.

Then when they contacted them, they might already have the solution ready. Right? They might have all the information they needed to solve it. And then when they contact the user, it's like, oh, we've solved your problem. Here's what's gonna happen. Does that work for you? Great. Here's the confirmation number.

So it was a much smoother experience. So I think that's the third piece that worked really well. And as I think about generative AI coming forward, I have seen a lot of them are being used. People are testing them out. Things like, help me study or , help me to to think through this breakup.

You know, I'm really having a hard time, help me vent. There's even people who are using it to help them with , drafting. First drafts is a huge use case, but like across industries and whether it's a speech or a standup set, you know, a joke set at a standup for a standup comedian. There's just a lot of back and forth that you can do to play on the bot and that really hasn't been possible before now.

So I think those use cases are new and what I'm not yet seeing is people coming to generative AI for a specific problem. , If I ordered a package of something from a store and it doesn't arrive, I'm probably not going to go to a generative AI tool at this time.

But that may be coming, right? Like I said, they might start , to blend those together. But for right now, the generative AI opens up a really cool opportunity for asking questions. You know, you could imagine a use case where you were a researcher and your lab has a website and people come to your website and check out your FAQs a lot.

Oh, maybe we should build a bot. And if people can use generative AI to talk to the bot who knows about all your research, then maybe they can ask questions to the bot that they would love to ask to you or someone else in your lab. But there's just not always that accessibility between, academia and the public.

So there's some opportunities where that back and forth, that Q and A, having a conversation about things is gonna be, I think, really compelling use case that we haven't seen before. And so the experience will probably get better that way where things that you'd love to have a one-on-one conversation with someone.

Now the bot will be able to actually answer your questions like that. So yeah, I think there's gonna be some neat changes in the user experience that way.

Jam

I think the reason it hasn't really happened yet, or I'm sure large organisations have already started, right, is this big question. And you mentioned it earlier, one challenge is the privacy and personal information and how people need to be comfortable enough, right?

I think this is the first step where people are now not hating chatbots like before and they're starting to accept it. Hopefully not really scared of it, cuz I know there's the other side. We're not gonna talk about that today. And you're right, it's all about the personal experience and that is something that I think is it is the next step. Mm-hmm. People are going to start going, well, hang on, you know, I've used these generative AI tools. They're still very generic. Yes, I understand. It's a first step or a starting point. It will come and I think pretty soon, right? 

I just remembered my conversation with my partner just tonight about that. And he works in a, media company here in New Zealand, so they're a pretty large organisation and now they're starting to think about that privacy and they cannot obviously use ChatGPT and it's, gathering all this data from different people in the world. Is it really safe for you to actually put specific information about your client, for example, some brief and all that?

So that's a big question, right? For organisations, like a media company and they've got all this personal information from clients , would you recommend that they build their own bot in their own servers?

Is that the way to go? Is that what you would recommend them to do instead of using ChatGPT, for example, or similar tools like that?

Melissa

Hmm, absolutely. So when, when we think about privacy and security for business uses of AI, that's always top of mind when you have a consultant come in. If that's not the first thing they're talking to you about, it's your data and where it's stored and how it's gonna be accessed, and whether it has to remain on premises or if it's located in the cloud.

And what kinds of personal identifying information or personal sensitive information are you collecting? Like those should be some of the first questions that you investigate before you start thinking about AI for business. In terms of individuals, like you said, using these tools today, ChatGPT is a great example. It's available for free, meaning that they are collecting the data to improve the tool. And it says that right in it's a free research version, we're collecting information, it's gonna help us make conversation better. And yes, they tell you not to put anything personal in there.

Anything that would be, potentially not just personal, but like, business copyright or business IP, like you wouldn't wanna put that in there either because they may use it to retrain the model. I think ChatGPT has recently come out with an option to opt out of the retraining program.

I think that's part of the pro plan, if I'm not mistaken. But the free version, they want an exchange for you using the technology and their servers, which, you know, it costs money every time you run a query on the large language model. So they wanna get something out of it. And the way I like to think about that is it's like Google, right?

You wouldn't put personal information into a Google search, just like you shouldn't put personal information into a ChatGPT or any other AI tool that's available to you for free. And sometimes even if it's paid. But there will always be terms and conditions that let you know if the data's being used or not.

And if so, hopefully they'll be clear on how it's being used. But for enterprise and for businesses, it's important to think about that data that you're collecting from your users. Even if it's just like name and address or name and email or something like that, that data needs to be protected and depending on your country or the country where your customers are living, there could be various legal ramifications for getting that wrong, right? So it should be top of mind where you're storing the data. The way that the data's being collected, it's all part of it. Now, when it comes to building your own models this is actually the direction that a lot of big firms are encouraging people to go.

There's a general consensus in the industry that there won't be one model to rule them all. There will be multiple models that are being used within each organisation alone. Each organisation will have their own models, trained on their data. It's almost . like a secret sauce, right? There's some proprietary data that you have on your customers that's highly valuable to your organisation.

And so when you build that model, it's, supposed to give you an advantage. You definitely don't wanna be sharing that with your competitors often. That's gonna be the way that it goes. So you can imagine in Canada, we have a, just a couple of banks, like six large banks and a couple of new FinTech competitors.

But all of them are going to be building their own foundation models, almost guaranteed. And so there's libraries that are available that are open source that kind of get you 80% of the way there. So if you are interested in learning more, Hugging Face is a great company to check out. So they're partnering with IBM actually which is how I learned about their vast library of many different models for you to use as a starting point depending on what industry you're in, what use cases you're looking at.

And these are built to be enterprise grade. And so as part of a secure technology stack, it should be no problem using the model s. Tuning it, getting it the last 20% of the way there for your use case, and then deploying it out into production in a way that is secure and respects all the data privacy that you need to consider.

So it's definitely going to be a in-house effort for larger companies and for smaller companies working, you can have conversational AI as a service and working with companies that are well established as leaders in the privacy and data protection space.

Jam

Yeah, that's awesome. This just reminds me of well conversation AI, future of work.

How does this actually play and how can we, look, as individuals, prepare for this shift? Because it is, you've just said it. My partner has already been talking to me about it wherein it looks like a bot can help them in their work, right? Whether it be generating some creative ads, or maybe help their video team make videos faster or easier, whatever it may be. Right? What do you see in the horizon, so to speak? Not too far off. Yeah. Yeah. Go ahead.

Melissa

Yeah, for sure. I think you're, totally on the money that a lot of employees are seeing use cases for generative AI and are kind of impatiently waiting for their employers to figure out what tools can we use and how quickly can we get access.

That's a really big theme that we're seeing across industries right now. So, in the most immediate future conversational AI, augmented, everything, is the immediate future. So already Microsoft and Google for their office suite, their spreadsheets, their documents, their presentation creation tools, all of those are gonna be augmented by some large language model that you can talk to and tell it, build a presentation based on this product brief and you just link it right from your cloud where your files are all stored. Because typically, organisations will be on some cloud environment. And then boom, it just generates out a presentation.

You get to tweak it, change it, modify it. But it's that first draft fast mentality is gonna be across everything. There's project management tools, there's design tools, so you can think a good example might be Notion for project management. They have Notion AI. It can summarise your projects, it can rewrite text for you.

It can find action items out of a call transcript. Some really cool applications there. There's Canva, which helps, you know, there's graphics and social media posts and documents and presentations, and they're offering an AI tool that you can access that again, gets you that first draft fast. You tell it what your goals are, who the audience is.

You have all your logos already uploaded, it can generate all of the first drafts that you'd like it to see based on, your brand voice and what your goals are. So some really cool applications. And so that overlaying of conversational AI on top of the tools you're already using is the immediate future of work.

So when people tell you that, Everyone's gonna be at, have to be comfortable interacting with ai. This is what they mean. It's the same advantage right now that people in the early two thousands who were comfortable writing Google searches had an advantage cuz they could find stuff on the internet better than others, right?

And here it's the same way. If you can prompt AI effectively, you can. Have an advantage because you can get what you want easier, faster, that kind of thing. So it's, it's a bit of a skillset and, and kind of a upskilling of the workforce. But like I said, we've, we've adopted Google, no problem. We've adopted Zoom.

No problem. I think it's just one more, the evolution of technology will continue and it's just one more tool. Perhaps in the very far future, we'll all be doing this with headsets, virtual reality, augmented reality. You know, we could talk more like that's really future, but in the immediate future, like I said, AI is gonna be laid on top of everything that we're already using and we're comfortable using as tools. Longer term, a little more speculation, of course. But I think that we can expect multimodal AI. So instead of just using an image in, in one tool and, and text in another. All of them will include all of them. You can prompt using video, voice, text, images, all of these different things. You'll be able to upload a TikTok video and, and that'll be, use this to help us write a blog post because this is a trend we're seeing on TikTok or something, right?

So that type of prompting will become multimodal. There's also going to be likely more powerful AI algorithms that might open up new use cases for applications beyond conversations. So things like detecting fraud or running optimization for a factory looking to improve its manufacturing processes.

There's just gonna be a ton of use cases in industry that may not necessarily be related to conversation but just, you know, as large language models have progressed, there's other areas of AI that are also progressing that might revolutionise areas of work. So things like computer vision models are a good example.

I love the idea of there's computer vision on and factory to help identify areas where, oh, this might break down soon. Just from, from looking at that would be awesome. The parts way it looks or there's even use cases in medicine where computer vision models are trained on chest x-rays. And they can detect pneumonia or no pneumonia.

They can also look for incidental findings. So things like a mass that shouldn't be there when they're looking for your pneumonia and they notice this mass as well. So there's just a lot of opportunity and different areas of AI beyond conversation that could also affect the future of work.

Jam

That is wonderful. And this reminds me of, I'm gonna put this in here cuz I'm curious, I'm crossing my fingers. You've watched it. I'm a fan of Black Mirror. I don't know if you've, if yep. Do you, do you watch Black Mirror?

Melissa

I haven't seen the episode. Ok. Only

Jam

And, and there's a, there's a point here.

I'm, I'm, I'm getting there, but yeah . So the latest Black Mirror, and I just watched this literally like last night. And we had to watch it again tonight before I went back to the office with our 12 year old. Anyway, so it's basically a story about StreamBerry, which is like similar to Netflix.

And what has happened is that they've got this quantum computer, and it's like magic because it just, what it does, it actually generates, right?

So it's AI, computers, computing everything and it generates and produces end to end. You've mentioned this earlier, end to end shows, and the dark side of it is that because all of us have accepted the terms and conditions and all of a sudden they can use your life.

In this episode, they used the, the character's life. Her name's Joan, and she couldn't do anything about it so everyone could see exactly what she was doing. There was a source Joan, which is the original reality, well, the human being, and then it is being generated by the quantum computer and then there's levels and so on and so forth.

So even the characters or the actresses, deep fake generation as they call it, I think that's what it's, it's called. And, it's all computerised. It's all digitised. Anyway, it's interesting. You should check it out. Yeah. Yeah. So I was just wondering, because of your expertise in quantum computing, is that very science fiction or is that actually happening right now or close?

Melissa

Yeah. So like I said, I haven't seen the episode, but from what I understand, yeah. The, the idea of taking a real human and then using AI to create content that they did not create that's definitely happening already and it's gonna get more convincing and better. So I think there was a news story about a fake Drake and The Weekend song that was generated completely by AI, but it went viral before it was taken down off of, I believe it was on Spotify, but I, I'm not, I'm not a hundred percent on that. But the idea being that, yeah, like the, you can basically have any voice be, mimic the text that you type.

You can have basically any image. It's, still a little off. I find when I watch deep fake videos, I'm a little weirded out by the character, you know? But it's getting better and it will continue to get better. And so this is a big reason why a lot of government organisations around the world are looking at regulation around this.

Like, if something's AI generated, it needs to be identified as such. So that you can't have a sitting president declare war in some video that goes viral. Right? And you know, that could be very detrimental for not just public officials, but individuals like your Black Mirror episode, right?

It, it could be that people's privacy isn't respected or that you know, it, it's kind of the definition of putting words in someone's mouth and it's done in a really convincing way. So this technology exists today, it's gonna get a lot better and a lot more convincing than it is today. And in terms of how quantum plays in.

So just a really quick background. Quantum computing is going to be a whole new paradigm of computing. So right now we have digital computing, which is zeros and ones or bits. When you have quantum computers, you don't just use a transistor that's on or off. You use some quantum system like an atom, and you can have many different states rather than just one or zero.

So with those many states, you can actually represent more information and you represent it fundamentally differently. It's not kind of a one-to-one mapping. This is how it works on digital. So this is how we'll program it on a quantum computer. We have to figure out all new algorithms and all new ways of coding on quantum computers.

So they are in a prototype phase right now. There's many quantum computers that work around the world. A lot of bay organisations and some small startups have made some really exciting progress in the quantum computing space. We're approaching in this next 10 years when businesses will actually start to see value out of quantum computers.

So, like I said, they're very early on. They're still prototypes. It's kind of like in the 1950s when computers took up a whole room. That's the way quantum computers look today. They take up like a whole physics lab. There's lasers and microwave generators and cryo fridges that keep everything colder than outer space vacuums that take out all the air from the system so that it can't mess up the quantum instrumentation.

So it's a very involved process right now. The kind of holy grail is to get to a million qubits. Right now we have hundreds to thousands of qubits, like, three figures to four figures worth of qubits, not a, yep. The, the millions of qubits that were, is kind of what all these organisations are looking for.

And the way we'll get there is by connecting them all together, just like we connect little computers to make a super computer. We're in this decade now where we've got hundreds or thousands of qubits, we can start to see some business value. Oftentimes this is through data science applications where in the past you might have coded a program that calls out to a supercomputer to run your calculation and comes back with the results.

Same exact process, but now you're gonna call out to a quantum computer to do your processing and then come back with the results. So that's the model that we're seeing and, and more and more companies are experimenting with it. A big use case are potentially vehicles that are electric battery powered.

And a lot of companies like BMW and Mercedes and all these large automakers are using Quantum to help simulate battery properties and to see if we can make a battery that you know last longer or charges faster or is better for the environment. There's all these different things that we could tweak to make batteries better and give these automakers a competitive advantage in the market and the reason why they're using Quantum for this is because quantum systems can model quantum phenomena like what's happening inside your battery much better than a digital computer can. And so, quantum computers will in theory be able to do everything digital computers can do. But they will also be able to solve new types of problems that we haven't been able to model accurately on a digital computer.

So that's why it's kind of picked up the mind of science fiction writers the idea that like, we don't know yet exactly what it's gonna be capable of. The same way that in the 1950s, the people building vacuum tube computers had no idea that Netflix would ever be a thing. So it's the same idea with Quantum today.

And so yeah, that excitement around what's possible, like we haven't quite built it yet, but as we build it, we're figuring out new use cases and there's some really exciting activity happening on that front. So in terms of generating AI content with a quantum computer, I could see it. There's kind of this idea that as AI gets better, it'll help us simulate or, crunch through some ideas for how to

better make quantum computers so that once they get better, maybe they can run more advanced AI algorithms. This idea that as one improves, it'll bring the other and then this one will improve and bring the other, they kind of in lock step get better. Both AI and quantum, these technologies might actually help each other to expand in the coming years.

But there's no like definite guarantee that like this is gonna be amazing. And, and here's the use case that's gonna just blow outta the water. Like, we're very still experimenting and there's only a couple of algorithms we know for sure we'll have life-changing consequences. One of them is the ability to factor large numbers and break encryption schemes that we use on the internet today.

So there's already banks way ahead of us looking at making better encryption schemes that are quantum safe. So yeah lots of opportunity and possibility there. But I can almost guarantee no one's using a quantum computer to generate AI videos today.

Jam

Well, apparently the Black Mirror Universe. There is one. Yeah. Right. That is, yeah. No, no, no. That, that's, that's awesome. It's just amazing. And isn't it, I guess the humans are just in general just trying to find out whether we can keep up with the technology. But it is definitely exciting.

And I have this term called I just had a nerdgasm. So all of that, I just had a nerdgasm, so there you go. Love it. Yeah, no, that's awesome. Just to wrap up, if there was one thing that you'd like to tell the listeners cuz there's just so much valuable insights and all of your stories and stuff.

I, I mean, I, we can keep on going. I'd love to keep on going. But in terms of conversational AI, what's one thing that you'd want them to remember? Just that bite-sized chunk on what they should expect in conversational AI. Is it work? Personal? What can they do? Wrapped in a little bit of ribbon for them to take away?

Melissa

Yeah, I would say that if you're listening to this episode, you're obviously very interested in artificial intelligence and how it interplays with human interaction. And so I think you as a listener represent some of the very few people on the planet who have this knowledge and who can do something with it.

So I definitely encourage you to start conversations in your workplaces. Start conversations with your family. There's definitely benefit to experimenting with these tools yourself. If you haven't already, I highly encourage that you're playing around with tools like ChatGPT, or there's Perplexity, P E R P L E X I T Y.ai and that's a great back and forth source that actually cites its sources that's connected to the internet. So if you ask it questions about news that happened yesterday, it'll pull up news articles from yesterday. So it's right on the, the cutting edge of being up to date and it, it cites its sources, which I think is a really important factor that we should be looking at.

When it comes to explainable AI, how did it come to this conclusion? How did it make this decision? So playing around with these tools, you'll start being exposed to features. You're like, ooh, I really like that feature. I think that should be implemented in the AI that my government is using, and I'm gonna vote for that.

Or, this should be implemented in my children's schooling if they're using these systems for how they're interacting with their teachers or how they're learning the material. This guardrail or this feature should be there so that we can see transcripts or we can get alerted of, of issues, or we can understand how it came to this decision, how it came to this conclusion.

So as a listener of this podcast, you're probably at the absolute cutting edge of thinking about these things in the workplace and in society today. And so your exposure to those tools and your use of those tools makes you even better equipped to have these conversations and be the leader in your space to start adopting these tools responsibly.

And making sure that we're seeing as many benefits as we can with as few downsides as possible.

Jam

They are definitely ahead of the pack. I always tell this to my team. I was like, be proud of yourselves team. Seriously, you are still a very small percentage of people around the world using this. So yes.

Spot on. Melissa. Thank you. Thank you, thank you so much. I enjoyed this conversation, this episode for taking the time to share your knowledge. You're very, very generous in sharing your knowledge, and I hope we can do another episode. Not necessarily conversational AI. Maybe we can do quantum computing and see what's going on, or other examples.

I would love to have you back in the show if you. If you want, you're open to doing that.

Melissa

Sounds great. I'm definitely open to chatting about this stuff. I think my husband's sick of me talking about it at the dinner table. So good to have another outlet.

Jam

No, that's good. That's good. Cool. And for those who's listening or if you catch this, a video or a YouTube short somewhere, we're not on TikTok yet, we're still still thinking about that. But hey, but we would really want to know your thoughts. Look, it's not just Melissa and I and the small percentage knowing about AI, we'd love to know what you think.

We'd love to share this knowledge to as many people as possible. Leave a message on Spotify, or if you found this on social, please leave a comment so we can start that discussion and of course hit that follow or bell to be notified of the next episode on your favorite podcast app. And if there's anyone in your network that might benefit from the show, please go ahead and share.

I would love you for it and thanks for listening and remember to keep the conversation going.