Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
GenAI
18 Plays1 year ago
Noah and Liam sit down with LD's GenAI expert, Chrissy! They cover the basics of GenAI and machine learning, prompt writing, and, more importantly, what it's like being LD's fire marshal.
Recommended
Transcript
Introduction and Guest Introduction
00:00:00
Speaker
Hello, everybody. Welcome back to another lovely podcast of The Commute. I'm your host, Noah Goldberg. My pronouns are he and him. With me is my sometimes co-host, but other times solo host. What's your name again? This guy over here? Yeah, you. Hey, I'm Liam. Welcome to the commute. Yeah, I guess we've passed this one back and forth. Yeah. my pronounnaza he and him And I'm just excited to be here to talk about AI. Look at us back together. Yes, speaking of AI. Dang, it's back. Such a lovely guest. Joining us. You have an OH. Hi-oh. Hi-oh.
Understanding AI and Machine Learning
00:00:41
Speaker
first fun fact yeah um My name is k Chrissy Monroe and my pronouns are she and her and I'm currently leading strategy for Gen AI and partnerships. l Love it. So happy you're here. Yeah, yeah happy to be here. We've been talking about this for a little bit. Like the build up has been like quiet. yeah And then like the last two days we're like, oh, so we're doing this. So it's today. Yeah. Yeah. Yeah. Well, i yeah. Or this conversation. Probably both. Both. Probably both. Mostly this conversation.
00:01:12
Speaker
Yeah. ah No, k Chrissy. All jokeside. Thanks for doing this. I think this is one that all the folks out there really want to enjoy. um Before this, Liam was describing like the emojis and it kind of got me thinking like he's feeling about a certain emoji. Like if I were to describe my emoji when it comes to AI, yeah whatever baby emoji, exist is that, like I could hold the rattle, shake it, you know, she wanted maybe a little, so like I could use chat GBT and I could use Garfield. You know, I've tried the photo stuff. It did not work out. now yeah Um, so, you know, before we get into what you're doing for a long while with regards to AI, can you explain AI and machine learning to us like we're thought?
00:02:04
Speaker
Yeah. I like that question because I have two kids and now we all have young kids in this group. So I feel like I do this a lot and I have explained to them what AI is because they're curious, which is really fun. Um, so I feel like I'm prepared for this, this question. Um, so they also have a hard time understanding like what an acronym is. So I'll start with the acronym. Yeah. It stands for artificial intelligence. good so Yeah, it's good start. And I think that's an interesting one. ba how made out Brain emoji. But it's important to think about because artificial, obviously, it's coming from a computer and then intelligence is an interesting one, even for a five year old to think about like, how do you define someone as intelligent? And what's that criteria for intelligence?
00:02:54
Speaker
And if you think about AI, there have been a lot of tests that have changed over the years to create that threshold of like, how do we say we've done it, right? Like that something is artificially intelligent. um So that's kind of evolved over time. But I think what AI really means is when computer programmers basically write instructions, that enable computers to be able to do tasks on their own. um That would have previously been done through very prescriptive code um or through people typing those tasks through manual kind of effort.
00:03:34
Speaker
um And if you think about the tests to prove artificial intelligence through those programs, there have been a lot that are like a chess game or a really difficult game called AlphaGo that computers have won. They beat the world professional AlphaGo player. um So then, yeah, so then there's that moment of like, is this it? Have we done it? and that I'd say the jury is still out on that one. But basically, you're trying to show that a computer is acting in a way that the other person interacting with that machine wouldn't be able to tell the difference. And so as we see now, these chat GPT models, you can kind of get the sense like we're getting there, where you feel like you're having a conversation or in a chat bot that you're getting support that feels human-like.
00:04:23
Speaker
And then there are still those moments where you're like, oh, that's not quite right. Or they couldn't quite understand what I meant. And I think a human would be able to understand.
Crafting Prompts and Future of Prompt Coaching
00:04:30
Speaker
So there is some interesting nuance there. and but That's cool. Yeah. And I think it's like, yeah i i'll do like I do use Garfield a lot. And like that conversation is very real. like it's about It's not like, hi, how are you conversation, obviously, which you can do. Yeah. But it's, you know, Here's a document that I'm preparing for XYZ, synthesize it and make an envelope points for me. And that conversation is actually super cool. Yeah. Yeah. Absolutely. I feel like I've heard from a lot of people asking questions around how to write the right prompt yeah and how to format the question in the question and answer. and equal are
00:05:15
Speaker
who haven't used it that much are still quite uncomfortable or they have one or two bad experiences where they don't quite get what they want and they don't know what to do next. I feel like similar to how SEO specialists are their own job now, I imagine, like prompt coaches. So you think how things going to be a skill set oh, yeah 100% because the cost of LLM back and forth is expensive as well So we want to keep using that technology, but in an efficient way So I feel like to get that prompt right is going to be kind of more important There are all kinds of templates for it, but my prompts are like three paragraphs long. It's like, you know, I am this person in this context at this time, talking to this team, I need these things from you in this format at this length, no longer that this many characters, but like, yeah, total opposite.
AI Learning Methods and Potential Risks
00:06:04
Speaker
Like costing us a bank, is that what I'm doing? Like quick, what, the kidders? A little bit. Yeah, not gonna lie. Server farm is man, they're on fire. Sorry, would you both consider yourself advanced
00:06:17
Speaker
Prompters? Prompters? Promp people? On a constant wage of discovery, prompt precision. Yeah. What would you, how would you, like, is it not to say it's perfect? Is that, is that what you would say? I just sort of put myself in the shoes of like, what's my audience? And I try to say exactly what I would be thinking and creating any material. I'm thinking about my audience. I'm thinking about my goals. I'm thinking about the voice audience. I'm thinking, you know, there's so many considerations of parameters I'm considering as a human. So I try to give those all literally to the machine. Yeah. It's like coaching a team, the match, and it's a team. And the first time you coach, you don't get quite what you want. You're like, oh, I should have said it that way. Mm-hmm. Yeah. More is better. so let
00:07:02
Speaker
um And then part of your question was about machine learning and how that works. So I think it's worth noting that there are three main types of machine learning and how people kind of learn in in a similar way. And that's how that learning has actually been developed. It's computer scientists thinking like, how do we learn and how can we make that happen in the computer? um So supervised learning where you really closely are teaching a computer but supervising how they do things is the first. And then there's unsupervised learning where you're really trying to get the computer to get to the right decision or conclusion on its own without your help.
00:07:43
Speaker
um And then there is reinforced learning, which is looking at previous mistakes and successes and judging based on that experience what to do next.
AI Bias and Media Influence
00:07:56
Speaker
And so those are three main areas of machine learning where computers are actually learning and getting better on their own or with help to do jobs. So is that the like the dangerous stuff that everyone talks about? yeah a Yeah, absolutely. And the other dangerous stuff would be that it's trained on the knowledge that exists or is out there and that that knowledge can be highly biased.
00:08:22
Speaker
So if you train that unbiased information, it learns like, oh, that's the right response. I too will be biased because this information is biased. So let's keep that bias loop going, which of course happens with humans as well. But nerd like, let's find ways to make it better. Have you seen the, uh, I just watched it on a, on a, on a plane where I had the creator. No, I haven't. Worth watching. Yeah, definitely worth watching. I mean, it's one of those AIs really watched, taken over the world and wars are happening because of it. Like it's not a good story, but there's a good story with that, I'll say.
00:09:01
Speaker
The cautionary tale? Yeah, one of those cautionary tales. This is what happens when you like you prep a guest. This is like I've learned more from the past five minutes than the past year of podcasts. So thank you, Chrissy, for bringing your education to us and simplifying
AI in Creative and Business Applications
00:09:17
Speaker
it. I wonder, you know, the framing it for kids is an alcohol lens. For me, with my four and six year old, week the the way it came to life, they're like probably a little too early for like quite the understanding you just described. So we actually just opened up Dali together and it was like, Theo, what kind of kids book do you want to write? And we like wrote a kids book together and then like typed it in and then it made the pictures. And I feel like with the the way technology is integrated into our lives now, he was like, cool.
00:09:44
Speaker
you know like he's gonna grow up with this it's just like yeah but a picture i said it should make and it wasn't really like an aha moment to me i was like can you believe this is happening it was like dad chill yeah it's like that moment where you show a cell phone you're like your whole life is here like yeah obviously obviously when do i get one We have a game somebody gifted us, Pictionary AI for Christmas. So AI guesses what your picture is and it's quite bad that that algorithm is a lot of fine tuning because the guesses are so far off and it has really funny biases.
00:10:25
Speaker
is it Is it the AI or is it the drawers? So I'm gonna tell you for I, maybe Liam, definitely not the artist. I will tell you like it stops at stick man. So sort of stick person I should say. yeah and Okay so yes you're explaining to us about AI that was really helpful um and I always do that by the way explaining to me like I'm fine and and I actually do that with Garfield a lot. Yeah that's a big one. yeah um
00:10:56
Speaker
But you have a mandate with regards to AI yeah and LCL, you know, some of the things you're going to do, like, you know, without getting into some of the specific projects, but what's your overall mandate? Like, why? Why is it that Chrissy is championing how AI is going to reshape some parts of LCL? Yeah, so ultimately, the goal is for AI to serve our business objectives for our business and our customers. So I'm really grateful for that grounding because I think it helps us make a lot of decisions. And then ultimately, when you look at what those goals are and what we believe AI is capable of, we've landed at this ambitious goal of $100 million dollars in annual savings.
00:11:45
Speaker
through a combination of projects powered by GenAI-led technology. And that goes across the entire enterprise, which is really exciting because like for a long time, LD has done tech for digital e-com experiences. And this is an opportunity to open the door and see how we can really build for the enterprise, which is pretty exciting. Sit. Yeah. um And like you jumped on this, right? Oh, yeah. Why?
00:12:17
Speaker
I've been really keen on Gen AI since it came to be. um So I personally was taking a course on it to learn more. I've been following along with what Liam's been really driving here in Garfield coming along.
AI's Negative Impacts and Misuse
00:12:31
Speaker
So I think like specifically I was keen to see what Gen AI could do I was excited about that opportunity to do more for the enterprise. And I think just probably you two feel similarly, but when you get a chance to really grow and learn and have a bigger impact, like those are the moments you can just jump up and say yes. I always try to find the capacity and to say yes.
00:12:52
Speaker
I feel like when you started taking a course at MIT, we were chatting about where this sat within our professional scope here. And all of a sudden, I looked at you as like, oh, you really know. like You are learning the actual principles and the history of of how this can affect us and how the technology is built. you know I was sort of dabbling and playing with the tools and seeing how it could help teach teams as sort of a a maker and a collaborator. But I feel like you know as you as you sort of took the reins on this one, it really leveled up. like the The level of understanding you have was pretty impressive. So thank you for joining us. That's awesome. Yeah. yeah i like i I'm similar to you, not even as advanced. Oh, you've attached EBT. I subscribed to Midjourney for a bit and just created some images. and
00:13:45
Speaker
That was it, and that is mind-blowing. But I think what you're doing, Chrissy, is really cool. But there's, you know, I talked about the move with the creator and how it's got. You talked about the machine learning side of the bad rap. On a personal note, do you have hesitations about AI? and if you Yeah, I think um we already talked about bias. So I think that's a big one. And there's we were just talking about this before the show that there was a whole kind of kerfuffle. I think this will keep happening where models were becoming extremely biased. And there's a lot of hate speech happening from Gemini. So it was like completely scrapped and the model was re-released.
00:14:31
Speaker
so that's real that's happening and i think i'm also pretty worried about fake news and like deepfakes you see the tech coming out from saura from openai and the ability to create these incredibly realistic videos. With the eye. I think we're already in a seriously dangerous zone with fake news and people feeling quite, you know, gullible and like receiving their news from all these different sources that are not vetted by copy editors in a traditional news way. So that does worry me just on a personal level. Yeah. How about you? I love how they like the samples they show from Zora are like basically like a Paw Patrol in real life. Like look at these fluffy puppies in a cloud. Look at the grand videos we can create. Like forget the fake news.
00:15:17
Speaker
use not d ah the The Biden pre-calls about the election, the, the fake voice, uh, work that was, uh, you know, happening during one of the, um, the primaries I got a week ago or so as one of the examples of like, Oh, right. We can impersonate people now. And those people, being a president, a prime minister, whoever can tell you not to vote, um are things that yeah are going to happen a lot this year, the year elections across the world. And yeah there's a lot of risk there. Yeah. And risk with fraud, those kinds of things. Anytime I get a like on fraud or unfamiliar number now, I don't want to answer because I'm worried it's going to know my voice. And then we use my voice for some kind of unlocking of my bank account, you know? Absolutely. These are real fears. I mean, I never answer those calls, anything. But I definitely, wow. Okay. I mean, there's terrible examples. I don't want to go too dark on these.
AI in Daily Life and Humorous Anecdotes
00:16:10
Speaker
Yeah, let's not. Okay. What's the creator? The opportunity, the joy. Yes, let's go back to the blue sky. Yeah. All right. This is probably my most important question. Okay. I want to know.
00:16:28
Speaker
I think everybody wants to know about what it is like to be the fire warden on the third floor yeah at Bathurst and Lakeshore. I'm so glad we had this opportunity. The whole the whole podcast was really focused on this question. AI was the bait. We're going to get real deep on the fire warden now. So look, I think lessons learned there are like, don't follow the rules and don't say yes to things, which it's hard for me. Yeah. So I was coming in person, you know, uh, as we were supposed to do, but you know, it took some people to get on to it. It's okay. It's okay. We got there we're there we're there. Um, and so I did that. And then Liz Kaladi, a dear friend, basically asked, she said, we need a fire warden. You're here all the time. So.
00:17:21
Speaker
Can you please say yes? It's a matter of safety and so I said yes and here I am but it did come with a hat and a vest. I know. You all know yeah yeah because I wore it for a recent craft dinner. We saw. yeah so I mean that I didn't even know we had a fire war. Yeah. until Ultimately yeah to tell you the scope of that job if there is a fire all I ask of our listeners today is to exit the building. That's what it comes down to. And that might seem like common sense, but i'm I basically have to be the last one to leave. And I don't really want to go down with the ship here. I'm not a captain, wow you know? So just just get out. It's not that hard. We can do this. You have to sign something?
00:18:03
Speaker
No, so that's a really good point. yeah so like So I can leave. I won't, but I know that I could. ah yeah I've got an a AI-related fire drill hazard I want to bring to your attention. So like half an hour before this, um the research team is briefing me into some AI work they're trying to duct tape together. yeah you know And they've found a way to put webhooks into a spreadsheet to speak to Coda to then summarize app reviews for PCO all the way from the app store to like usable data for a product team. And they're telling me, shout out to Nick and Abdul, the research team, for this demo. And it's a great example of the many sort of like innovative ways people are trying to use AI and connect our systems together. and
00:18:48
Speaker
But they're telling me that you know token limits, load limits, and API access are are going to be a real problem for them.
Conclusion and Farewell
00:18:56
Speaker
And they can hear the fans and their machines starting to heat up as they run these scripts. That's like a full circle moment. Oh my gosh, it is all connected. So it took me a second there, but I managed to create the connection for us from the Fire Warden. to AI strategist. So I think they are related in that case, and it's just a watch out. Keep your eye on the research team. All right? You never know what's happening over there. Wow. That is a real, legit, that makes sense. You need to be the firework. Yeah. And if you ever pass on- A lot of processing power required around here. If you do pass AI on to someone else, they need to do the firework. This is it.
00:19:36
Speaker
They go hand in hand. I'm glad that that's all become so clear to me. It never was connected. Yeah, well thank Liam for that one. That's why I showed up today. Thanks for having me. Alright folks, ah that does it for this episode of The Commute. Thanks for having a listen and watch now. I hope I look my best. Before we get on. Thanks folks, thanks Chrissy.