Britney Muller: The Truth About AI's Roles in Writing and Marketing

Tim Stoddart : Welcome to the Coffee Blogger Podcast. My name is Tim Stoddard. Thank you so much for joining me. My guest this week, the one and only Brittany Muller. What's up Brittany? It's so great to finally meet you.
Britney Muller : So good to finally meet you. I'm excited to chat.

Tim Stoddart : I'm going to start this off in a place where I don't normally start it off with, with a little bit of an open-ended question. I think you're a really interesting person because you are interested in so many different things. When I'm following you online, or even when you used to publish your blog, I don't think you published an article a while ago, I used to follow that. I was always like, man, Brittany's into so much, so many different areas of marketing. I mean, even aside of marketing, you're pretty open with some of your personal life as well. So my question is like, what are you working on right now? What are you excited about right now in the world of the internet?

Britney Muller : Yeah, it's a good question and good observation, Tim. I think Right now, what's really sort of captivated my focus for the last year has been this explosion of AI. You know, I saw the puck going in this direction for so long, but I never in my wildest dreams think it would, thought it would come this fast. And it sort of just exploded, you know, last year with chat GPT and something that. Concerns me a bit, especially with writers and marketers and SEOs is There is a lack of just foundational information when it comes to quote unquote AI, which is really just machine learning. And how these systems are operating under the hood isn't how most people think they are. And so by sort of pulling back the curtain on what's actually happening under the hood of these large language models. equips people with the information necessary to navigate the changing future. Otherwise, you're going to be left to all of this misinformation, all of this AI hype, and using these tools for all the wrong applications. So that's something like near and dear to my heart. I'm sort of on this mission to make this technical stuff more accessible to a wider public, because I think that's the only way we advance as an industry as well, is getting that adoption, getting that buy-in. And quite frankly, it's not people like me, nor is it the people building this technology that are going to come up with the next best application. It's going to be people, you know, listening to this podcast. It's going to be people that have domain expertise in a non-technical position that think, Oh, if it can do this, we could use it to do this. Right. So I get really excited about that.

Tim Stoddart : All right. So let's just get right into it then. Like you just went for it, right? I agree. I mean, look, there's a lot of people in copyblogger, especially in our Academy and our membership site. People are very fearful because they say this is going to steal my job. Um, which maybe it will to some extent with, with some people, that's just a part of life, right? Things change, but I like what you said right there. I don't think people understand enough and it's almost like the definition is wrong. People say AI, artificial intelligence, and that's not actually what it is. Like these aren't sentient beings. You need to feed them with data and you need to give them information. And they're just really good at, well, maybe constructing is the wrong word, like breaking down large amounts of information really, really quickly to provide an output. So, I mean, since we're just going right into it, I think a good place to start is almost at that foundational level where LLM is really what it is, a large language model. And it's, it's almost like a data aggregator on just like a huge, massive level that we've never seen before. So for the people listening that keep hearing AI, AI, like my job is gone. Writing is gone. None of this works anymore. Please just explain the difference. What does it actually do? How does it work?

Britney Muller : Yeah. Oh my gosh. I'll try and keep this high level, but there's so much, uh, to it in terms of like, just even looking at the pipeline of the massive data set required to make these language models sound so human-like is ridiculous. It's large amounts of unconsented biased data that just gets fed into these models. And the way that the models learn is they're word guessing. They're masking different words and they're predicting what they are. And so over a long period of time, like months and sometimes even like a year plus, they're learning the laws of language. They're learning general structure, much like how I'm not speaking in random words right now. There's structure to the way that we speak and that can be mapped out in a high dimensional space. So that's what these models are doing. But what's interesting is when they're generating text, they're basically fancy word guessing machines. They are word guessing. They are predictive engines. And oftentimes, like with all of the content that's fed to it, they grab the median information that they've seen about different things that you're asking it or whatever. And it's AI researchers call it the view from nowhere because it's so middle of the road. You're not going to get edge cases or, um, you know, updated information or all of these things are going to be left out. Um, and so it's, it's a bit scary that we get this kind of median view from nowhere and different things compound on themselves. And so I I've said this for a long time and I still stand by this, that. writing content is one of its worst use cases. It's not good at writing content and that's not what it should be used for. It has no ground truth. It has no experience with the real world. It has no shared experiences like you and I have. Right? So that's what really, in my opinion, is what good writing is all about. And what's always been sort of coveted on the coffee blogger site, in my opinion, is like that high quality writing. And these models can't do that. What they can do is they can help summarize things. They can help explain something to me like I'm five. outline an article based on top ranking content, which is something I like to use it for, you know, they can do the boring stuff for you and set you up for success where you can really shine in your writing and probably write more efficiently.

Tim Stoddart : The best way it was ever explained to me is what they're really good at doing is guessing the next right word. You know, they never have like a cohesive thought. It's literally, and using words is easy just because it's something that as human beings, we totally understand, but the same is true with. With graphics and with numbers and with, with graphs. And if you're asking an LLM about statistics, like it doesn't actually know the answer. It just knows the very next thing to say. And when I figured that out, or when somebody explained it to me that way, it really made a lot of sense because my. um, viewpoint on it is actually that it makes personalized writing more valuable. And like, this is the, the, again, this is something that I heard where it's like, if everybody has it, then no one has it because it's so accessible now that the, the advantages have, have just been washed away. You know, anybody can use it. So now there's really no advantage, which to me. And please tell me if you disagree with this means that your personalized experience, your view on the world, your like unique viewpoint on culture and life and relationships or whatever is even more valuable than it was before because there's only one you and like the one you is like more unique, right? So am I just, am I wishful thinking here or am I on to something?

Britney Muller : No, you said that perfectly. And it's so refreshing to hear someone say that because I think the AI hype tends to cloud people's judgment around this and believe that, Oh, these really are sentient and carry all this stuff. They don't. You're exactly right. They are word guessing engines. And to even add like a layer of texture to that, they are, what they're doing is you put in an input and they take that word or sentence or phrase and they create a probability distribution of the most likely next word. And so with that probability distribution, they are looking at what's most commonly been seen on all of the internet around this context. So again, it's that sort of high volume stuff tends to win out. But it's taking this probability distribution, and it's adding an element of randomness. And this is really important. So an element of randomness, because you might think, why wouldn't they guess? Why wouldn't they generate the most probable next word every time? Researchers have done that and it's incredibly redundant. It's super dry and boring. It doesn't really go anywhere. So that degree of randomness, which can be adjusted is what adds sort of that human sounding creative output that people really resonate with. But it's something to consider, right? Like it's not just a, it's a word guessing machine with random elements attached to it. So. In that aspect, I think that information alone can really teach you a lot about what's possible and what's not. Um, yeah.

Tim Stoddart : Isn't there some kind of, uh, like some measurement tool that predicts, like, isn't there a word for it? Like the, what's coming to my head, like the Pareto principle, right? That's like a definition of some kind of theory we have. Isn't there some kind of, uh, statistical bias where you can measure the probability of if something is, is written by AI or not. I don't, I don't want to put you on the spot. I just, I think there's actually a word for that randomness where people can kind of tell this is too perfect. This isn't human. Do you know what I'm talking about?

Britney Muller : Yeah, I do. And I don't know what the word for it is. I know there have been different efforts around that. And unfortunately, what ends up happening is like, they're sort of okay, but then they get fed something like the Bible, and the Bible on some of these top rated tools of identifying AI versus authentic writing, flagged the entire Bible as AI generated. So it there's flaws, but there's also flaws because people edit the the content, right? So it's not always just going to be, um, copied and paste, but you're, there are, there are some like general clear red flags that something's been written by AI. Um, and yeah, hopefully that starts to get identified on a more regular basis.

Tim Stoddart : Hey there, it's Tim and I need to take a moment to tell you about this show's sponsor. It's a product called Hype Fury. When I was able to speak to Yannick, who is the CMO, one of the founding partners of Hype Fury, and he agreed to sponsor the show, I was so thrilled. And the reason is because I have personally used Hype Fury for the last three years, and it has allowed me to build my social media following and my personal brand to over 70,000 followers. I could not have done it without Hype Fury. And I really, really mean that. I use this product every day. And it's added so much to my business and to my life. So Hypefury is a social media scheduling tool. It has three main features that I think separates it from every other tool. One, it, it allows you to quickly create content and schedule them. So it's a very nuanced feature, but it's so helpful. Basically I sit down at my desk in the morning and I type out my tweet and I type out my LinkedIn posts and then all I do is I hit enter. And Hypefury schedules it at the opportune time on Twitter and on LinkedIn. I don't have to think about it any more than that. All I have to do is sit down and create my tweets, create my posts, hit enter. And Hypefury does all the work for me. Uh, second Hypefury makes it so that you can easily create threads and threads have been the biggest value add for me in growing my following. So threads really helped me grow my following on Twitter and those threads format themselves into longer form LinkedIn posts on LinkedIn. It's actually kind of funny. I made a video about this not too long ago about how, yes, like you want to create threads on Twitter. You want to be a thread boy, because I'd say like 80% of my growth on both Twitter and LinkedIn have been from threads and long form posts. And I wouldn't have been able to format any of this. without using Hypefury. And then third, Hypefury is really good for keeping you inspired. So what it does is it, it shows you some of your most popular tweets and your most popular posts. And it basically gives you information. It gives you inspiration as to what your audience is looking for and what they're most actively engaged in. So you're never sitting at the computer thinking, oh man, like what am I going to say today? What, you know, what kind of content am I going to create today? It's constantly feeding you new ideas, new inspiration, and it allows you to, to quickly create this content so that you can continuously get yourself out there, continuously build your brand. And most importantly, turn that social media following into newsletter subscribers. So through Hype Fury, I've been able to grow my personal email list, timstodds.com to over 30,000 followers. That's turned into a business within itself. It's really helped me grow the copy blogger newsletter. We're at 110,000 followers right now. A whole lot of that is also because of hype fury. So please, this is a product that I use every single day. I personally vouch for it. You can check it out at hypefury.com. H Y P E F U R Y.com. If you have any problems with it, you can send me a DM on Twitter and I'm sure I can convince you as to why it will add value to your life. So hypefury.com. Thank you so much to hypefury for sponsoring the show and let's get back to the episode. My agency website, my agency focuses in a lot of healthcare industries and one of them is autism. And so I wrote an article about marketing agency for autism centers and it's like a relatively low competition keyword. So I got it to number one pretty quickly and, um, I put it in the AI checker and it was 100% written by AI and it's ranked number one. Pretty instantly. I mean, I'm not like the best writer in the world, but I know how to write a 3000 word article and do all of the things. Right. So I wasn't surprised by it and it's personalized and I've had legit statistics in it and it was the whole thing. You know, like when you put it through those scrubs, it highlights it all in red, but like there's been the whole thing was just like red blocks everywhere. I was like, you gotta be kidding me. I wrote this. I know I wrote it because I was the one that did it. So it's, it's a little suspect.

Britney Muller : Yeah. They are certainly not, uh, not foolproof. And I don't know that we'll ever get that right. Because here's the thing is like language is so ambiguous that It's going to be very tough to flag down every single piece of AI-generated content versus not. Also noting that it's been trained on all of the internet, much of which is human-written text. So it does in some aspect write like a human, right? And I know there's different efforts to fingerprint AI-generated content through like Maybe grammar stuff or certain words or layouts. So that's something to think about as well as like, if you're using it, like edit, edit appropriately, make sure it's good for valuable for people. Right. I think that's what it comes down to is like having human centered content that answers people's questions.

Tim Stoddart : Yeah. You just actually touched on the next topic that I want to write about, which is something I've, excuse me, to talk about, which is something that I find super fascinating. Right. You said. They're trained on the internet and the internet is written by people. And I remember thinking during the writer strike a while ago where these, all these writers are mad at the wrong people. Like they're mad at the wrong thing. If they're so scared about AI and chat GPT taking their jobs, they shouldn't be boycotting corporations. I mean, I guess that's a good person to look at, but the actual root of the problem is the fact that all of the content on the internet is just fully indexable by AIs. And when, like, I don't get how it's still that way. How has nobody once said, I spent hours writing this thing. You can't just train your AI on my article and then multiply that by hundreds of millions of articles and millions of authors, right? I guess what I'm asking is I'm presenting the problem first. And then I'll ask you more specifically, do you anticipate people coming together and creating some kind of walls to keep AIs from indexing the open internet? And like, if they do, is that even possible? Right. Are we too far gone for that?

Britney Muller : That's such a good question. And something I struggle with, because I mean, we see it really clearly in the art community, specifically with like, generating styles of specific artists is so evident that it's been trained on. But the same can be true of exactly what you just said of people's writing, and different things that they have available online, it's not consented. People have not consented to their work being used in this way. And there's a couple of lawsuits to sort of keep an eye on right now around that. But, you know, unfortunately, it's sometimes really hard to prove. And yeah, we're just gonna sort of have to see I know that there are ways to block open the open AI crawler from your content, but then you run the risk of not being included or mentioned when someone's asking a question about your product or services or you as a person. So there's, there's a list of risk and rewards to weigh.

Tim Stoddart : So let's talk about the rewards then. Yeah. We've like, uh, we've villainized AI so far, but I'm actually like very optimistic about it. Um, I saw a headline that you, Unfortunately, I didn't buy the article, so don't shoot me. But the, uh, the title was AI is neither good nor bad. It's not going to save us nor rescue us. And you were quoted in that. And I think that's like a good viewpoint because, uh, I mean, I'll give you another example. My, my wife is, she's always wanted her bachelor's degree. I never went to school. I think school is kind of dumb and she wants to do it right. And she's in a statistics class. And she called the teacher's assistant about a question that she had. And the teacher's assistant said, just go to Google. And of course, my wife was like, I'm paying all of this money for it. Like, isn't this what you're here for? And so I say, just use Chatsubt. Like you'll get the answer within a split second. I know it's weird, but this is just the world that we live in right now. And an hour later, she's like, this is the greatest thing ever. I've learned more about statistics with this one program than I have burying my face in a book and asking questions that I don't even necessarily understand what I'm asking. Right. So I see from that viewpoint, I'm like, wow, this is so powerful. I think we have to admit that there's a plus side to it as well. So what kind of plus side do you see?

Britney Muller : Oh, absolutely. I think some of the learning applications are really interesting. Um, but should also be fact-checked, right? Like these aren't perfect. This, it's not information retrieval. It has a good general knowledge of subjects that it's been trained on, but it by no means has like a high expertise on any of those things unless you fine tune it. But to your point, I mean, I am really optimistic as well. I feel like There are lots of applications that haven't even been explored or discussed yet that could help a lot of people, even just like that initial writer's block. Right. Or like starting going from a blank page and just getting like an outline or a brief or I've, I've used it to craft a legal sounding email. and got the response that I wanted so quickly. I was like, it works like this works. This is wild. Um, so things like that can empower people that maybe don't have direct access to legal support, um, or, you know, have a hard time advocating for themselves. I think it can do really powerful things. of that nature, but it's also really good at just like general language skills, natural language processing skills. It's incredible at sentiment analysis. is incredible at categorizing content and keywords and all sorts of things just based on the natural language that it's learned about through this training. And I'm saying that as someone who's such a fan of all of this other technology that already do these things really, really well, right? Like we have information retrieval, we have powerful natural language processing models. But there are different applications where I see ChatGPT blow those, some of those out of the water in terms of the natural language processing. And I do get really excited about that.

Tim Stoddart : Every time you say something, I I'm trying to talk about DataSci, your, your brand that you're working on. And like you came on my podcast, I'm really getting there. I promise. But, uh, every time you say something, I feel like I need to, to riff off it a little bit, because I think we're really aligned here. Like the speed that you mentioned, I think is what makes this a little bit different. I remember I was going for a walk. I was living in Nashville. I was listening to a Seth Godin podcast and he was talking about GPT-3. And he even said, most people don't know about GPT-3, but that's because the exponentiality of this just hasn't quite hit yet. And I think we got to that point where all of a sudden the ability of it If you can measure that in exponents, I think you know what I'm saying? Just shot up so, so high. So I guess. Are we still going to keep going? Like, have we hit a ceiling for this yet? Are we just getting started? Where's all this going?

Britney Muller : Yeah. What I find most fascinating in that conversation about like adoption and use case. So obviously it's like the most explosive software we've ever seen. Reaching like a hundred million users on like what a month or something. What I find most fascinating and why I'm so passionate about this work is I see non-technical people in Um, HR in like teaching, um, graphic design, using this tool in the most. Thoughtful and creative ways, saving them time and effort on some of the more banal tasks so that they can really level up their, either their strategic thinking or they're more, um, like hands-on, right. It comes back to like, just being a human, they, they can spend more quality time with students. Or in a meeting, you're not taking notes because you have Otter recording and transcribing all of the notes on the fly. Small businesses are using this in really incredible ways. I heard, and I do this too, and I've stolen this idea and it's awesome, but I make so many loom videos for, for clients of like, here's how you run this thing that I built you, you know? And so I walked them through it, but you take that and you transcribe it through something like Descript, throw it into chat GPT and say, create a manual with this information. So now you have both like video support and like a user manual to do this thing for new hires. Um. It just makes onboarding more efficient. So those are the ways in which I feel like this gets more and more applied and used is when people know how to use it in the right way and know what to feed it. Again, like content generation isn't one of its strengths. Feeding it the right kind of things and asking it to modify it, or enhance something or, you know, provide some sort of transition on it is really, really valuable. And it's those non technical people that are so scrappy and coming up with the coolest things like JP, I'm gonna butcher his last name, Hocula. He, Holeca, he's coming up with some of the most incredible things. ever for SEOs, for marketers, for writers as medium as jpiddy.com. I can send you a link to it. Kristen Tinsley has some of the most badass applications I've ever seen, and I've used a bunch of her stuff. One of her scripts pulls all of the Google News results, like the top 20 news results for a particular topic, and it analyzes common themes right now, sentiment, and it And then it adds a column, it generates the CSV, it generates a column for how you could potentially news jack this stuff. So what are the things you can write about to get all this visibility in the news right now? I mean, it's just, it's genius and it's using it in the right way to be a more effective writer, to be a more effective marketer and get your stuff seen by more people online.

Tim Stoddart : Uh, I saw that exact script. The first thing I thought of is, um, well, Brian Clark, I bought another one of Brian's websites. It's called your boulder.com. And, uh, I'm really interested in like local media just because Facebook destroyed it. But just because newspapers aren't there, it doesn't mean that people don't want to learn about local media. And so I saw this, like these little micro local media site trend coming along. And I noticed really quickly that. The best thing you can do is get on Google discover for a local news story. But in order for me to do that, I'm not a reporter. So I was like, you really just had to be clicking refresh all the time, seeing a story more or less like copying it, putting your own words into it and then publishing it. And then just hoping that you get picked up. And so when I saw, um, that application of it with, with newsjacking, I think that's like a really good word. That was the first thing I thought of. I was like, you could put any local media site in Google discover, like instantly right away. And. The, the scrappiness I think is a really good word because kind of what you're doing there is you're, you're feeding off the back of like more enterprise organizations that like kind of have the ability to be out there and be on the streets. And you just don't have to do that anymore. And so I think about that and like the scrappy side of me thinks like, that's really fun, you know, but then I also think, so does that mean that everybody is just their own like little scrappy micro media entity these days? Like, doesn't it just long tail the entire thing?

Britney Muller : I don't know. I think, I think people. that have naturally that strategic mind or are creative or driven enough to experiment with this stuff will come out on top. I think oftentimes, a lot of the writers I've worked with and marketers, they take their writing so seriously, which is great. But when it comes to applications like this, having that soft touch of, this is an experiment, Right. And if it's effective, we know it works. Great. We can double down on that. If it doesn't. Great. We can avoid that in the future. The worst part is not knowing, you know, whether something works or not for you. So I do think that a level of, uh, exploration is, is probably required.

Tim Stoddart : Yeah. And that's an easy, that's an easy thing to say, but not a lot of people, there's not a lot of tinkerers in the world, you know, how have you used it?

Britney Muller : Or have you used it for different writing stuff?

Tim Stoddart : Not really for writing. I agree. Um, I've tried, especially with outlines and I just, maybe I've been writing for so long, like I'm a very fast writer. And so it kind of was just like, I can do this faster on my own. And I'm sure you also can relate to this. You do it long enough. You really establish your own voice. Like people know when it's, when it's me, I just, I came up on copy blogger. And so I like trained myself in that. short paragraphs, kind of choppy, right for the internet type thing, almost to the point where I've tried to write a book a couple of times and writing it like complete paragraphs is actually like impossible for me now. So, uh, so I'm working on that. I, I have gotten the most use from it. It was funny when you mentioned loom, um, I do the exact thing. Like we have people all over the world now. And so my mornings is basically spent two or three hours recording room videos and. Um, like creating tasks on, on Todoist. And so now I basically created like my own micro transcription and task management platform all through, through some of the scripts that I wrote. So that's been really sweet. And that saves me like a ton of time. I do think another great place I've found it is I feel like some software applications are just trying to use AI because they think it's, it's cool. And other ones it works really well for like the loom transcription and the auto titling of loom. As soon as that happened, I was like, this is the best thing I've ever seen in my life. Um, another one is on my sales team. Um, HubSpot, actually it's not HubSpot. It's a third party outbound Firefly. I think it's called, um, it, it transcribes the sales calls. So like, I don't have to listen to recorded calls anymore. I can just look at like the real quick transcription. I think that's been. really cool. So the one example I gave to you has been me personally, but the most value I've found for it has actually been when certain tools align with it really well. And so far, like transcriptions and those Loom videos are, it's just the best. Every time I see it, I kind of laugh to myself. I was like, this is great.

Britney Muller : That's a really good point. Yeah, it's insane. Have you used Descript at all?

Tim Stoddart : Um, we tried to when I was trying to transcribe, um, all of the podcasts, but another example, uh, what the hell is the tool? It's like a podcasting transcription tool. I use it every day. So. I, this is, this means a lot to me because I've had a couple of people in the Coffee Blogger Academy that were hearing impaired, like they're deaf, but they love podcasts. And so I felt like it was an obligation to really transcribe those articles. I used to spend a ton of money on Rev and then, um, This tool came out, I can't remember what the hell it's called. I'll, I'll make sure I put it in the show notes, but now every episode that we publish has a perfect written transcription of it. And it's all through this one podcasting tool. And so I think that's another really good example.

Britney Muller : Wow. That's great.

Tim Stoddart : All right. So let me finally bring this full circle to the point. Data sci 101 data sci 101. I can never figure out exactly how to say that word. Um, you say that your mission is to make LLMs as accessible as possible. I think that's a great mission. Um, for a lot of the reasons that we talked about, I think we probably touched on some of the examples, but off the top of your head, can you think of any other examples as to where it's not as like scary as people think it is, where there's some real world, simple applications that people can use in their everyday life?

Britney Muller : Yeah. I mean, I think even in like systems that we encounter every day, like there's so much room for efficiency improvements, whether that be at like your supermarket or like, I think of like real world applications and how different some of this technology could be used more effectively there. I even think of like personal assistance. I think we're getting to a point where that could really help a lot of people. You know, I think the whole, uh, will AI kill us all is like, in my opinion, it's like a rich white dude problem who like likes to think about some of that stuff. And the way to avoid that is quite simple. It was like not connecting it to an action, but there's a lot of. you know, stay at home, single parents who would love this technology to help them do just everyday things, right? Like, um, whether that be scheduling or, uh, just different, like digitally automated tasks, I think. I think it could help a lot of people. I get really excited as well in the, in the medical space. There's, I don't think there's a, uh, a clear use case just yet for LLNs, but image recognition and like, you know, x-rays or cancer identification. All of those things are coming a really long way and are exciting to explore. But something I talk about at different conferences is like, it really comes down to the data set, right? Like, is it representative of skin tones, whether it be like a an image on someone's camera for skin lesions, like there's a lot of dermatology apps that try to accomplish this, and an early model, its single most indicator of cancer was the presence of a ruler, because all of the cancerous images were taken in a doctor's office with a ruler measuring the lesion. So examples like that are kind of what fuel me as well. And I don't know how PG you want to keep it here, but there's- Oh, go for it. Okay. And I think some of these examples make more of an impact of the Tesla stuff. Again, it's A model is only as good as its training data. So garbage in, garbage out. And we are the ones that it's being tested on. So Tesla autopilot, they've known about this for years and the case is coming up next year against this man. who was killed using autopilot, and he was one of a handful that were killed in the same exact way, where autopilot's on, and a semi-truck at a very specific angle, the model identified it as a bridge, and it goes full speed under it, taking off the top of the car, right? Another example is like Uber's early self-driving automation model, was trained on a bunch of people riding bikes and it was trained on a bunch of people walking around the car. Great. It had no idea what to do when a woman carrying all these groceries was walking her bike across a street and she was struck and killed. There was a man in South Korea recently who was smashed by a box smashing machine because it identified him as a box. There, there is real world harm today with these applications. And in the same way that it's important to know how large language models work, it's really important to know how these systems in general are only. As qualified as the level of diversity of the training set, right? There's oftentimes so many, and you could argue all models. are imperfect, but some are valuable. And it's those edge cases where you really have to weigh the risk and reward. So applications like healthcare, I'm super hesitant to, to go there with some of the more predictive models, right? Like there's two types of AI. There's predictive and generative or sorry, discriminative. You want a discriminative model, identifying your cancer that's been trained on all of the, you know, you want something to pinpoint. You don't want to. predictive AI, like an LLM taking your symptoms and saying, yeah, maybe. So again, it's just knowing like which, which is being used and. Is it the right use case? What are the things you have to think about? Do we have access to the training data? That's a huge effort right now that I'm also really passionate about with better access to the training data. At least we know what biases live within a dataset. So Margaret Mitchell was very publicly fired from Google for the stochastic parrots paper. Um, she's been working really hard on getting model cards adopted more readily in the industry. And those are basically nutrition labels for these models. And it tells you what websites, what data went into it so that you can really gauge whether or not it might be suitable for X, Y, Z. That's why I love having this podcast because sometimes it's like

Tim Stoddart : Should we just talk about whatever we're interested in right now? I think this is definitely one of those times where, you know, maybe it's bigger than writing because I'll tell you what excites me most about it. And this is just something that I'm passionate about and it's nutrition. And you talked about diversity in the LLMs and being to understand the difference, right? In my view, the biggest problem, I think any of us in the country, I mean, at this point of the world right now is like, bad nutrition. We're seeing it everywhere. And it's very easy for people to say, you just have to eat better. I'm in a place in my life now where we can go to Whole Foods every day and be really, really picky. And my wife is really into it as well. I grew up in a place where you just had to go to the corner grocery store. And the food is different at those two places. And so I think a lot about what It can't be that hard either. That's the point about it. Because if there's one thing we really, really understand, well, we learn more every day, but like nutrition labels, we pretty much get, and they're all digitized. And so how hard would it be for someone to create a model that makes it so that someone who grew up like in the neighborhood in Northwest Philly can just easily scan some of these things. It spits back out what the macros and what the nutrition elements are. And then that way. You can just totally skip the, the uncertainty of like, Hey, I'm just hungry. Let me buy like a bag of chips right now, or let me do this thing. And that thing, it'll make it so much easier for single moms or just single parents, or even just like young parents where they don't have to stress about it so much. And they just have by this, by this, by this, by this, by this, and takes all the stress away from it. So I know we're veering off course a little bit here, but we're really talking about the same thing. It's like it's aggregation of data. to optimize certain aspects in your life. And that's what I'm excited about.

Britney Muller : Yeah, that's such a good point. I know there was one that came out that you could take a picture of like your plate, and it would try to predict some of the things. It wasn't obviously 100% accurate, but it's getting somewhere. And something as you're explaining that I think that's such a beautiful application of this technology. And the little like industry flag in the back of my head starts going up of unfortunately, and this is something like I really hope we can figure out. Unfortunately, something like that won't necessarily be super aligned with corporate best interest, which is money. And so that's where I get concerned is like, fuck, like how are we going to make sure that this technology is accessible and valuable to underprivileged populations of people in the same way that others can access it. And they do talk about in the industry and like different papers I have read on this topic about geographical racism and how Unfortunately, regions of the world that are going to get the most negative impact from this technology, so the environmental impact, the people that that affects first will be the last people to receive the benefits of it. That is something that I do see. people thinking about more and more, and there's different efforts, like Timnit Gebru, who was also publicly fired from Google, started D.A.R.E. So it's, I think it's distributed AI research, and they are really trying to sort of better democratize and make sure that this technology is being applied in a more thoughtful way that's, again, not just like immediately pertinent to big tech revenue. And so, yeah, that's where like the resources come in that I think about a lot.

Tim Stoddart : Is that like really what you're trying to do? Because you've always been pretty public just about some of your beliefs and trying to make the world a better place. And I read the tagline, make LLMs as accessible as possible. But it seems like there's like an undertone there. Like you can go a lot of different ways with what we have. So, you know, I'll keep it traditional. I have a semi a semi-traditional of this podcast where at the end I always ask like, so with the project you're working on, you know, when you lay down at night and you really think like, what could I do with this? What could this become? Is that, is that sort of your overall mission? Are you trying to just help people?

Britney Muller : I think it's twofold. I think I'm in a really privileged, unique position untied to big tech companies where I can blow the whistle on a lot of this stuff. A lot of people don't know about like the cleaners, right? The people that are paid $8 an hour or $8 a day in Kenya to clean horrific data and have to deal with all the consequences of it. Like it's, there's awful, this is like, I joke how like AI is like, how the hot dog is made, like, but we have no idea, like all the nefarious things it took to put this thing together. So that's one one aspect is like that awareness, that knowledge. And then the other thing I do think about a lot is like, kind of what we've just been talking about is like, how can we use this technology in ways that it does reach those populations of people in a meaningful way. One of the efforts that I've been thinking about and talking to people at the Allen Institute for years now about is like the elephant project where they're using drone and image recognition technology to identify poachers before they get close to a herd. And I think what a cool application, right? And think about all of the, the ways we could maximize resources or figure different things out in developing countries and have more representation of the global South. That is what I get really excited about. So it's like, it's both like the. It's kind of like the head and like the body, the action, like the creating the awareness, the educating people on how to like keep themselves and their families safe and how to be thinking about this stuff. But then also to your point, like what are those applications that could be put on like, they call it edge AI or disposable AI where these models can live. on very inexpensive, like dollar, like raspberry pies or, um, phones. Like there's, there's lots of really cool applications, but again, the funding and the resources aren't readily available for that. So how do we create a space where that's possible? That is something I've always thought about for sure.

Tim Stoddart : Yeah. So it's not just like a profit shill.

SPEAKER_02: Yeah.

Tim Stoddart : Yeah. I'm, I'm so glad that, that we chatted about this. I want to bring more recognition to, to this new company that you started. Datasci101.com. I'll link it in the show notes. Copybloggerpod.com. Everything will be there. Um, you have a free ebook and then a guide that you're working on and we will put all that in there as well. But you know, sometimes you just got to go down the rabbit hole. And, um, I think like you and I are super aligned on this stuff where nothing is good and nothing is bad. It's just. Where are you going to put your energy? And I'm happy that there's people out there like yourself who are like being active about, maybe not active, like being very mindful and intentional about where that energy is going. And I think it's super important. I appreciate it.

Britney Muller : Thank you, Tim. This was awesome.

Tim Stoddart : Yeah, very welcome. Well, like I said, copybloggerpod.com. You'll have all the show notes to everything we talked about. You can go to the website. You'll have all of Brittany's contact information. Anywhere in particular that you think people should go? Just data, data side, one-on-one.com.

Britney Muller : Data side, one-on-one.com. Email's always great. LinkedIn's okay. Yeah. Yeah. LinkedIn's weird. I'm trying to get better on LinkedIn, but I still, I still like Twitter. I know that's like frowned upon these days, but I still like it.

Tim Stoddart : I can't help it. I'm just having a hard time letting go. And even threads like. Now I'm starting to get trolled on, on threads. It took like two weeks and you know, I'm already getting trolled on. I'm like, I'm not that important. What are you trolling on me for? So we'll see. It was really great talking to you. Great to meet you. We'll keep in touch.

Britney Muller : Sounds good. Thanks Tim.