From Computational Chemistry to AI at Scale

Transcript generated automatically from the episode audio and lightly cleaned up (filler words removed, misheard names corrected where clear). Speaker labels were added afterward and may contain errors.

Rob May: Hi everyone, I'm Rob. I'm the co-erric and you're watching the AI in NYC show. You just have me today. Both my co-hosts were not able to make it. So, I'm here with our guest, Alexa Griffith, who is a senior principal software engineer at Red Hat. Alexa, welcome to the podcast.

Alexa Griffith: Thank you. Thanks for having me.

Rob May: I actually want to start with a non-technical thing, which is I'm from Kentucky and you're from Tennessee. How did you get to New York and how do how do you growing up at a place like that? How do you how do you feel about the city?

Alexa Griffith: So, how did I get to New York? I'll answer that one first. I was actually a computational chemistry major in college and I was working at Oak Ridge National Lab which is an amazing place and their scientific computing department and I'd done some research at University of Tennessee Knoxville as well and I was at a point where I didn't know if I wanted to continue and go get my PhD in chemistry and but what I did know is that I really liked the coding part of it. I really liked creating scripts. I like the problem solving problem solving that coding entails.

And so I reached out to a VP of engineering that was working at a startup here called Bluecore. And he was a UT alum. I met I got connected with him through my friend who was also at UT and I basically just somehow convinced him to give me an internship and I just came like I didn't really have a goal to move to New York City. Maybe I wanted to move somewhere bigger like we were just talking about. I'm from a very small town in Tennessee. We have farms and hay bells and I grew up on a farm. So, it's going to Nashville.

Rob May: Nashville's a good city. It's a big city.

Alexa Griffith: Yeah. I'm I'm quite close to Nashville. So, my family lives there and it is. You're right. So, it's not like I'm out in the wilderness alone, but the day-to-day living is quite different. Yeah. But that's how I got to New York and I've had an amazing time here ever since. I love New York. I think it's great.

Rob May: Yeah. Yeah. Same. I like being at the center of everything that's going on. And I actually So, it's funny. I have a connection to Bluecore which is I know Michelle Mcome the CFO.

Alexa Griffith: Awesome.

Rob May: She was the CFO at data which acquired my first startup. So that's how I got to know her.

Alexa Griffith: That's amazing. Bluecore in 2019 was such a fun time. It really shaped me as an engineer and I had great mentors there as well. They're like incredibly smart people at Bluecore. I met one of my best friends there. So it was really a great time for me.

Rob May: It's funny how that happens. So now you're working at Red Hat, but you have at least one, maybe more open source projects that you're contributing to. Talk a little bit about the Envoy AI gateway, which I is that your main project that you work on on the side or not as much now.

Alexa Griffith: I'm still in open source, but more a little bit lower in the stack, which we can get into, but I can talk a bit about the on gateway. That's what I was working on mostly when I was at Bloomberg for the last part of me being at Bloomberg. I was there four and a half years working on the inference platform.

Rob May: So for people that aren't very technical, explain to them what an what an what a gateway is for an AI system.

Alexa Griffith: Yeah. So I mean the name gateway is a little is a good name for it as well. It's what we were doing with the envoy gateway and the problem that we were solving is that when GenAI came into play and our AI engineers wanted to build products on LLMs to get more resources and to get experimentation going and building those products, we need to have access to them. So that means that we were using different model providers. So GCP, Azure, AWS, but all of this traffic is going outside of Bloomberg's system and that needs to be secure. We need to manage it and we need to be able to have obser observability, know everything that's going on. And as a platform team, ideally it needs to be easy to use. And if you use all of these different providers just as they are, none of those things are true.

So what you need is a gateway in front of it. So we run this instance and all the traffic passes in and out from us so we can control it and we have like a unified API for example. So like I said all of these different providers GCP, Azure etc. They all have a different way in which to communicate with the API. So we created a central unified API and then we translate it within the gateway. So that's some of the logic takes care of. Same thing with like authentication tokens, authorization, things like that. There are a few different features that it's nice for and observability. It's like a central place to track how many tokens you use and rate limiting things like that.

Rob May: Yeah. And so you developed this when you were Bloomberg?

Alexa Griffith: I my manager Dan Sun co-founded it with Tetrate, but I was on the team that was working on it non-stop.

Rob May: Yeah. Okay. Very cool. So talk a little bit about you the woman that you gave the presentation with at the event that I saw. Was a little bit like anti-Python let's say for AI stuff and I guess you guys have written envoy in Go and C++.

Alexa Griffith: Yes. Or I wrote it mostly in Go there might be some layers in C.

Rob May: Okay. How do you feel about the fact that like despite these problems like most of the machine learning ecosystem is in Python? And so have you guys got any push back on any stuff like that or do people do people care or do the issues that caused you not to use Python to build envoy? Do they do they perpetuate into other AI products because people use so much Python?

Alexa Griffith: It's funny because a lot of the AI tooling is written in Python because AI engineers write in Python. A lot of the infrastructure code is written in Go. Go is really great with concurrency and speed and it's generally preferred for infrastructure products. For example, we had an internal tool at Bloomberg that was actually doing the same thing the gateway was and it was running into scaling issues and I think with a lot of Python products typically at scale running Kubernetes native you run into these things. So Go is usually preferred for that scale cloudnative distributed systems logic.

Rob May: Yeah. So you mentioned earlier that you were a chemist but you like the coding side of chemistry. And so it makes sense why you'd become an engineer. How do you feel now? How much have you adopted the automated coding tools? And do you feel like it's changed your view of coding? I like I've heard several people say like I actually enjoyed writing code and I don't like software engineering as much now because I don't really do it. I'm more like a product manager. What are your feelings towards this?

Alexa Griffith: I don't feel that way actually. I enjoy it. I think at the time what I liked about I liked coding because I like creating things. I like building things. I think that's fundamentally the thread that I have. It's if it's not coding then it's using the agent or agents or an AI tool or a coding tool to help me build things quicker and at a larger scale. So I think it's a really great tool that enables you and I think the ability to adapt and change and make use of the best tools you have is really important. So I think it's just changed. I love systems problems and I think that I've lended in exactly the right space for that. So there's still a lot of very challenging things that I enjoy.

Rob May: Oh yeah, for sure. I mean AI is in the first half of the first inning. So it's still a lot a lot of stuff to do. So talk a little bit about what what are the things that you work on generally at at Red Hat and sort of what made you sort of shift into that position? Do do you have an AI position there? Do you work on AI stuff?

Alexa Griffith: Yeah, I recently moved. I just joined Red Hat like two months ago.

Rob May: Okay.

Alexa Griffith: Yes. This is new. So I'm actually a Ford deploy engineer and I know that's a hot topic. Yeah, I'm really excited about it and I was looking for other possible IC roles and Tanim which is my manager. And one thing I'll say, first of all, I always think about the people first. I think following people is so important. And one thing I was struck with at Red Hat was not only that they're smart, but they're also kind. And so I definitely that was one thing that drew me there, but also they came to me with this opportunity of a for deploy engineer to work with customers. So as you know, I have a podcast and I've also done quite a few different talking engagements and content creation things in the tech sphere. And I wanted a way to use that more internally to create levage or like I wanted to challenge myself in a different way. So I feel like this is the best opportunity that I could have been presented with because I actually like talking to customers. It's that puzzle of figuring things out and why I like having a podcast as well. I like I'm very curious about how people are running things, what their problems are and what pain points they're running into. So I find it really fun so far and it's a great new challenge.

Rob May: Yeah. You're the first podcaster that we've had on our podcast. So, what what made you decide to start your own podcast?

Alexa Griffith: Yeah, good question. I think it comes from that nontraditional background that I had when I joined Bluecore. I was painfully aware of all the things I did not know and like I said, I had great mentors in there, great teachers, but I always felt like a lack of knowledge that I there's so much I need to catch up on. And I think actually that worked out well for me because it pushed me to actively learn so much so quickly. But in order to learn, I started to write. So that was my first content creation. And I was actually writing blogs on Airflow, which I don't know if you know Airflow, but at the time it was a new open-source like ETL DAG pipeline, just like a step-by-step process where you probably transform some data that our data scientists were using. And I wrote about that and a lot of people were looking up looking it up because they were looking for advice or like help on it as well. So that was a great feedback loop where I just gained momentum and kept writing. But then I started to get writer's block a bit. I mean writing is writing you need a fullformed idea and I was thinking how can I create content and also have an excuse to talk to smart people and learn from them and so the podcast started.

Rob May: Yeah. What's been what what have you learned from doing the podcast about a you know a well are all your guests in New York or do you see people anywhere or what's the

Alexa Griffith: Yeah, I do it remotely. I would love to do something like this but for now I just do it on like Riverside.

Rob May: Yeah. Have Have most of your guests been local in New York or have you just has there been a have they most been in Silicon Valley like

Alexa Griffith: I think they're everywhere generally. Yeah. I've had guests from in Europe quite they're everywhere. Yeah.

Rob May: What's your what have been some of the takeaways? What are some of the things you've learned about AI that have surprised you or you know from from interviewing all these guests?

Alexa Griffith: About AI specifically, I love asking people how they use it every day. I think that we don't share enough of that, like how to use it to be productive, how to use it outside of work. I think I have learned a lot about CEOs or how people have started companies or how I wouldn't even say CEOs like I had well Joe Beda was a CEO but just a builder or someone who creates something that's highly impactful. Their stories are not always the same and mostly I would say I don't know if there is a single story that is even so similar but what's common in all of them is they kind of just did it more or less you know they they build they act they don't wait for everything to be perfect they ship before it's perfect and they iterate quickly and they're very surprisingly humble about what they don't know and I think I appreciate I appreciate that a lot about them as well I take it as a learning about people and their stories. I'm really interested in that.

Rob May: Has it has it shaped your views on where AI is going? And do you have a point of view around you know like Silicon Valley is very obsessed with AGI and are we going to get there? And part of the reason I'm in New York and what I love about New York and part of the reason we do this show is I think this is the hub of like applied AI where people actually think about what can you do with these things and if we get to AGI or not it it matters but it's not as important maybe as people think it is. And so what what what are your points of view on that? And has has anything from the podcast shaped your point of view?

Alexa Griffith: Yeah, I think my podcast has expanded my network to meet people and be in contact and have these conversations with people that directly shape how I view AI and the future of AI. There's a couple that I've had recently, like one with Rob Shaw, and a previous one from the CEO, Wafer AI, that were extremely insightful and informational about the direction in which AI is going. And I feel like that's changing all the time, you know, and so I'm really appreciative for those conversations and those insights. And it's really amazing to see all like the technological advances and the algorithms or the routing of AI and all these things that are coming together from different levels of the stack to create a more optimized AI imprint stack. And I think it's fascinating all the different edges of innovation.

Rob May: Yeah. That yeah it's definitely a hard area to work in because you do have like you have you have new chips coming out you have new tools coming out you have new models coming out you have new new ideas and infrastructure coming out and that's all on top of you know the applications that are coming out and you know I've been in tech for 25 years and normally you have like pieces of the stack changing but not the whole stack at one time so I think it makes it a really complicated problem.

Alexa Griffith: I agree and it's one reason that I love this role I'm in so much as well because it forces me to need to stay up to date on all of that and I think it's important just to be honest about what you don't know because anyone who expects someone to be up to date on everything all the time that's just like crazy. You spend your whole time just reading about AI because everything changes so quickly and we're in New York there are so many other amazing things to do and fill your life with as well. But I try to learn at least one thing every day. Okay, I think I'm learning way more than that. But as I'm talking to customers, I need to know the language, you know, I need to know when they're asking about one product versus another in any part of the inference stack. I need to have some context on that. So, I'm really happy for that challenge to be able to, you know, need to know all these things all the time. But I agree with you that's overwhelming and like I don't know all of it.

Rob May: You know, you could read model cards all day, right? Like that would be it. That would be a whole day. And you can

Alexa Griffith: It's interesting though. I love doing deep dives into the different different things like the different types of GPUs or the different types of algorithms that people are using now, but it's a lot.

Rob May: Yeah. So, you're you're dealing with customers, you're helping them implement Red Hat tools in their organizations. What are some of the things that have surprised you about Well, let me ask this question. I is their level is the average level of AI expertise deeper or more shallow than you expected when you got out there? Are people overwhelmed and they don't know what's going on or they have a pretty good handle on things?

Alexa Griffith: I think generally people are overwhelmed. I think there's two types of people. One is the person who is so overwhelmed because you're throwing acronyms at them and they think they need everything and they're just overwhelmed. And the other is someone who might completely underestimate what they need because they just take some things for granted and maybe think it'll just work. So I think a big part of my job is bridging that gap, finding the use case and helping people to understand exactly what it is they need. And I think that is a very a very significant problem right now. I think people don't clearly understand what they need and how they need to run it. And there are so many different knobs. So I think simplifying that, simplifying that process, making it easier to understand and easier to use, I think that if I can do a little bit of that, that's a huge win for me.

Rob May: Yeah. We made this mistake at the company I run where we because we're all a bunch of engineers. We're like people want all the n knobs and dials, you know, and they they want to do all this stuff and then we find out was people really want kind of like batteries included, right? Like make a recommendation. Let me check. Yes, I want to do that, you know, or or something. It is it is pretty overwhelming. Are you are you seeing amongst the customers is it still a pretty stable common use case you see across customers is it mostly like support chat bots and stuff like that or are you starting to see a lot more variability in the ways people are applying AI

Alexa Griffith: I think there's obviously there's common themes but there is quite a bit of variability and I think that what's interesting about that is then that entails a different set of configs for each of those like batch you want to probably for example maximize throughput if it's a chat a lot you probably really care about the time to first token because that's how the user perceives how snappy you are, how quick the UI is, that's the user experience. So I think always starting just like any software system starting from what's your use case, what are your requirements, what's your user story, what do you want your user experience to be like is important, but I think there are so many knobs to turn as you just mentioned that helping people figure out what those are clearly as they're just as a lot of them are developing in real time is important.

Rob May: Yeah. So in terms of time to first token, have you have you deployed a language diffusion model anywhere? Aren't they faster to first token? I slower overall or maybe it's vice versa.

Alexa Griffith: I have not personally deployed one, but I know that they have a slightly different algorithm that people are very excited about. Yeah. That's supposed to be optimized.

Rob May: Seems to me like they shouldn't work when you like I understand how they work for images where you're filling in some of the you know pixels as you go but doing that for words seems yeah surprising

Alexa Griffith: yeah it's crazy what they're coming up with I think also like some other optimizations like speculative decoding is interesting but this has a way of you know trying to guess those words in a more optimized way that's inherent to that LM or to that language model algorithm algorithm which is very interesting interesting. I think there's a lot of excitement over it and I think that's one thing that maybe surprised me. I thought all of these different knobs that were turning outside of the model would be really important, but it turns out still that the algorithm within the model itself is quite important and I think you're seeing a few different ones pop up that are people are really excited about. So that's really cool.

Rob May: Yeah. What are you seeing on the cultural side of these implementations? Are these implementations working in organizations or is there a lot of push back or a lot of confusion because it's a new way of doing things?

Alexa Griffith: I think there is quite a bit of confusion. I think we're still in that like slightly after that exploratory phase trying to make it a product market fit and trying to run it optimally and what does that even mean? And I think what's interesting about that is that you can have the best model, you know, have the best benchmarks for that model, but that doesn't always relate to having the best product in production. So you not only need a good model that's good for your use case, but you need to make sure you have everything around it tuned, you have the routing correctly, it's running efficiently. I mean, when it comes to latency, it's not just the model, it's all of the steps in between as well. And I think we're seeing a lot of that routing logic be surprisingly important because if you can't route to the cluster that has the KV cache that you need to use, well, you're going to miss the KV cache and that's going to be like a cost and a latency head. So, yeah.

Rob May: Yeah. And we've seen that a lot there. You know, there's a there's a series of use cases that are related to like AI search. And those are very difficult things to do because these LLMs, particularly the big ones, are so slow. And search is one of those things where people are like, you know, give me an answer. You have literally hundreds of milliseconds before they they leave. And so we've seen a lot of different there's a lot of different strategies to deal with that. You know, I think some people are using these bigger chips like Cerebras and that's crazy, right? Yeah. Running stuff on that. Some people are offloading some of the steps to like small language models that run quicker trying to some people trying to parallelize some of it. But we've seen some interesting latency focused AI applications I think are one of the more interesting areas of engineering right now.

Alexa Griffith: Yeah. And I think what's interesting is with every optimization obviously there's a trade-off. So speculative decoding you need a model that's predicting the next few sequences of tokens or whatnot. And with with each one it's like if you parallelize the model well you might run into a memory issue. So yeah, all of these things, managing them and balancing them, I think balancing them is the most important thing to get the best product output you can, not specifically honing into just one. And I it's not it's not easy. It's not intuitive. And I think there's a lot of room, like I said, for helping people figure that out.

Rob May: Yeah. Well, it's I mean, we're really just now in sort of the last year moving into the engineering phase of AI, right? It's been about this. It's been about capabilities. Can you make a model do this task? And now people are moving out of pilots into kind of the reason I would call it the engineering phase is what you said making trade-offs, right? What do we care about? Accuracy versus speed versus cost versus latency versus time to first token versus overall throughput. And yeah, and then you have network latency in there if you run models in different data centers and all these kinds of things.

Alexa Griffith: So yeah, and evaluating it is there's nuance, right? Because you could be evaluating a model in AWS running bedrock model X and on GCP same model X it doesn't perform the same. Why is that? The interconnect might be different. Now we're dealing with like GPU interconnects and how how data is transferred and there's there's a lot of different things. I mean it's a distributed systems problem which makes it so challenging and fun at the same time. But it's not exactly easy. I mean like every platform I think that we're building it out and all of these optimizations and advancements are all going to start come to coming together and I'm hoping at some point that you know we'll have a more mature AI platform but that process is really fun right now I think

Rob May: but it is by nature probabilistic and we may be stuck with that forever. Yeah. So it'll it'll be it'll be interesting to see but we've seen we've seen similar things. We've seen if you run the same model, even a relatively reliable model, maybe like Llama 370B, like running it in two different places on different systems, like it might even give you slightly different answers sometimes the same prompt. Yeah. So, which is crazy, but welcome to the world of probabilities.

Alexa Griffith: Yeah. And then when you put an agentic in there, now you're giving a whole other set of complexities as well.

Rob May: Yeah, that's definitely challenging. So what's your what's your take been on the New York City AI ecosystem? You know how do you feel what do you feel about some of the strengths? What would you know what would you like to see more of? Yeah what's your overall take?

Alexa Griffith: I think AI itself, I mean, I think New York City itself is such a great place because, well, I think diversity in different forms is really useful and great, especially when it comes to thought. So, one reason I love New York is because there's so many different types of people, a lot of amb ambitious people working on a lot of different verticals, different career paths, and I love taking advantage of the city and all the different things they offer like going to the museums, being a part of the young fellow museum clubs or doing art classes because you meet so many different people with different perspectives. And in tech, we're always built, we are in tech, but we're building things for people to use. And so I always think that New York is such a great place to innovate because you can learn from so many different types of people with so many different perspectives. So that's one thing I love about it.

Rob May: Yeah. San Francisco has a little bit of a monoculture and in one way it's awesome because everybody's in tech and other ways it's terrible because everybody's in tech. It's like the same flip side of the same coin, right?

Alexa Griffith: Yeah. I was talking to a CEO that's there and they said, "The greatest thing about that is that you're really focused and I bet that's amazing. Sometimes I wish I had more focus here honestly." Yeah. But the great thing about New York is you have such a full life. There's so many things to do. And I think it goes back to coming from Tennessee. I didn't really have a lot of opportunity to do a lot of things in Tennessee. Like if I wanted to be on a volleyball team, a club volleyball team, I went I drove an hour to school and then I would have to drive a different hour to be on that team. And so that when it comes to opportunity, I think that I really want to take advantage of it living here. And I've had so much fun doing that. Like I take random classes all the time. It's fun.

Rob May: Yeah. Well, even I even contrast it to I was in Boston before this for 14 years. And Boston is a fantastic town, but in Boston, I never had problems like figuring out what I was going to do on like a Monday or Tuesday or Wednesday night, right? Because all the stuff was like later in the week or weekend loaded. There wasn't much going on. But like this is a city where you're like it's like a Monday night and you're like there's a Knicks game and there's a math lecture and there's a concert and there's a Broadway show and there's museum opening. It's like it's wild how many things there can be. So it's definitely pretty overwhelming.

Alexa Griffith: Yeah. But it's great. And also I do think when it comes to thinking about systems, for example, I just saw something this girl was making a video about how knitting is like coding. And I love these interdisciplinary systems thinking and ways of thought. And that's what I think is really great about having hobbies or just what New York has to offer is that it opens you up to think about how people operate or how systems operate in a different way. And I think that broader view is really great for bringing it back to your own career and your own life.

Rob May: Yeah. I definitely agree. Let's hop into a little bit of some some of the news that's come up this week. There's some interesting things and I'm curious if you have an opinion on some of these. If you don't, just say so and we can move on. But the most shocking I think piece of news to me this week is that one of the companies that looks at credit card data and what people are paying for has said that Anthropic is winning a lot of business against OpenAI and ChatGPT on the consumer side which is really surprising to me. You know which of those models do you use and do you have a perspective on why that might be happening?

Alexa Griffith: I love Claude. Yeah. I started off with ChatGPT. I think ChatGPT was first to gain people. Yeah, Claude has been really good about keeping people in my opinion just from what I've seen and what I've heard. I think Claude did a great thing with Claude Code coming out when they did and the adoption. I think adoption is important especially in enterprise and speaking from just the people I know here and like the companies that I interact with in New York and the people that I know that are engineers here, it's been the same. I mean, I use Claude code for things I would use ChatGPT for. I use it as just like a chat and like to keep up with my to-dos and like all these workloads. I love skills and I think they were so good in bringing that early and being focused on their values. And I think that translates being really focused on your values is something people respond to, I think.

Rob May: Yeah. I think you know what I've said on the show before I think the mistake that OpenAI made is being first and having to spend so much money to train these models therefore needing to raise so much money. They had to go and convince investors, you know, when you have to raise $500 billion dollars or trillion dollars, whatever, you have to convince investors, well, here's how we're going to return five trillion dollars to you or something like that. And when you when you have to do that, you have to sell the story that you're going to be in every TAM, right? Because I think if you had even told investors what the coding TAM alone was going to be, they would have been like, "No, no way." Right? Right? Because they would have just equated it to like engineers, but they wouldn't have factored in the idea that people got to code more stuff. You're going to make more software if it's faster and easier. Things you wouldn't have coded up before. And so I think the thing that OpenAI struggled with was spreading themselves too thin because that was the only way they could raise money was to say like we're going to be in all these TAMs and have all these tools and again profit came along and didn't have to do that.

Alexa Griffith: So yeah, I'm also really impressed with Cursor. I think cursor is great as well. I mean, I think they're all great, right? Like I haven't used Codex as much admittedly because it's not just available and it wasn't available when I for work, but I also really I think I underestimated the IDE. I thought, oh, I don't need anymore. But actually, it's pretty nice to have one. So, I'm also a cursor fan.

Rob May: Yeah. Are you a CLI person in general, though, or what do you

Alexa Griffith: Yes. Yeah. I mean, yes and no. I think I go I go between both quite a bit.

Rob May: Okay.

Alexa Griffith: I used to be really big on JetBrains. I don't use JetBrains much anymore. JetBrains tools were awesome like GoLand, PyCharm, and then there's also one for investigating databases that was amazing. I forget the name of it, but I don't use it anymore that much.

Rob May: Well, it'll be interesting to see what happens to Cursor now that they've been acquired by SpaceX. Yeah. So, I'm sure I, you know, I think I think Grok had a Grok with a K, Elon Musk's Grok, not the not the chip company. I think they had a moment where they were briefly maybe earlier this year, February, March, where their models were in the lead, but I think outside of that, they've really really struggled to follow what everybody else is doing. And so I love the model race because you really don't know what's going to happen now.

Alexa Griffith: GLM all here. That's all you heard about this week. GLM 5.2.

Rob May: Yeah. Yeah. Well, and it's interesting in that perspective because it's you know, there have been Scott Bessent was on CNBC I think yesterday saying that, you know, look, America's way in the lead on AI and all that kind of stuff, but the models would say maybe not. And it's you know, we talked on the on the last show about the fact that there's some legislation coming. Some people think the model should be legislated. Some people think that the United States government should regulate who has access. But like, you know, I guess the question is like, do you think they can stop it? Like, like even if they come down and they're like, look, anything that tests over this level, tests at like mythos level capability or more, you can't sell. I mean, with with the internet the way it is and all the people working on it and the number of people that know AI and the dark web, like I mean, how would you do do you think the government can or should even stop people from using these things?

Alexa Griffith: I don't know. I think that you should be cautious about what you build and I think that people should always try to do good in the world. And you know, I think that there are a lot of great use cases for AI and it can enable a lot of great creation and help us move forward as a species, as a tool. At the same time, it's a bit of a, you know, a double-edged sword because it can also be used for bad. And I think that we just need to be mindful of that. I don't really know what the right answer is. I think it's hard to stop it at the government level, but I really don't know.

Rob May: Yeah, it's kind of out there and I think it's going to be tough. I want to talk about a rights issue, IP issue that I think is pretty interesting. There's a show called Peppa Pig that I don't know if you're a fan, but yeah, it's a show targeting you know, early school age children. And the voices on the show, the children that were on the show doing the voices have been asked to sign a contract that they can use the voices to generate AI voices sort of going forward. And the industry is very upset about this because you don't get much money on those AI generated voices sort of, you know, going forward. And I'm curious like do you think we're going to see stuff like this in other fields and like does it concern you like what do you think that you know you're talking about like hey we should use AI to do good like how does it you know what would you do you have a policy in mind that you would craft for these kinds of things you're like well AI is going to be able to take your likeness or take your code or take your voice and do some stuff and how we should approach that.

Alexa Griffith: Wasn't this a Black Mirror episode?

Rob May: Maybe. I don't know. I don't watch the show. Everybody tells me I would love that show, but I haven't had time to watch it.

Alexa Griffith: You would like the show. I think you should watch it. It's a good episode, but I think it's a great question and yes, I definitely think it's already happening. So, my friend is creating a platform for models, for example. She was showing me it's great. And it's part of it is about using their likeness for advertising. And I think that's a great tool because it's giving them the ability to say yes and be compensated for that. And I think they should be compensated for it as well. What do the rules look like? What do the contracts look like? I'm not really sure, but I do think people should get something for their likeness, you know? I would hope so. There's so much free free content they've been trained on. So, I don't know.

Rob May: Yeah. Well, so related to that, sort of the last article for today is from the New York Times. So, Gina Raimondo, who used to be the secretary of state under the Biden presidency, has pulled together OpenAI Anthropic, Amazon, and Microsoft, and they've signed onto an effort to try to ease the AI transition for American workers through, you know, training them to use these new tools and all this kind of stuff. So, I'm very excited about this. I think this is the kind of thing that like politically I'm very libertarian, but I feel like this is a good use case of government time and energy, right? Which is to there's a big disruption coming from technology. And you know, we're going to try to help the American people work through it in a way that they come out the other side more positive.

And I prefer this to something like universal basic income where it's like, hey, we're just going to pay you and figure out what you want to do. I actually think this training could be good for the United States if it can help us transition to sort of an AI focused economy. You know, what what have you seen around job loss at the places that you've worked or the people that you know, have you seen any hints of AI sort of, you know, you've seen Coinbase and Square and some of these places like lay off a lot of people? Like, have you been involved in any of that? And does it worry you?

Alexa Griffith: I haven't like I don't know many people who have been impacted by that but I know that it's happening. Does it worry me? Not really because I feel like I have more work than ever right now. But generally does it worry me that like as a for a population? Yes. Although I don't I think that we're just not there yet for the adoption for that to be a real thing. I think right now we're just creating more and there's so many more things to create. At some point it may get to that and that would be unfortunate. I hope it doesn't. But on my podcast I've actually had the chance to talk to a couple different people. One's working with elementary school students trying to lobby for different different teaching mechanisms and different ways to use AI like for the teachers to learn and implement them. And that was an interesting conversation and also with someone else who was working with different businesses like different enterprises and having helping them to change the culture and have like AI champions. My point is I think that it's definitely a cultural problem and so the integration and the adoption of AI is really important for it to be useful and that itself is a cultural problem problem. So I think that having a institutionalized learning initiative is super helpful for Americans as well.

Rob May: Well, I think that's a great point to end on as we wrap up. So, Alexa, thanks being on the for being on the podcast. And if somebody wants to reach out to you or they want to be on your podcast or whatever, what's the best way to do that?

Alexa Griffith: Yeah, I mean, I'm on I'm on LinkedIn. You can find me on my website w.alexagriffith.com and there you can find my link tree which has all my links to my podcast which is Alexa's Input AI podcast and yeah, I'm just reach out to me on LinkedIn or something. I'm there.

Rob May: Awesome. Well, thanks for being on the podcast. Everybody watching, thank you for watching the show. Like us, leave a comment, suggest a guest. Anything that you want us to cover, or anything you'd like to see us do, please send us an email and, we'll see you next week. Close the door, please. The next