Transcript for "What AI Still Can't Do": Welcome, everyone. Thank you so much for joining us for our session today, what AI still can't do. We are so excited to hear helpful insights from our partner, Griff Bohm, vice president of growth at Momentum today. Just as a quick reminder, you will receive a recording to the link to this webinar after the session. And with without further ado, I'm gonna go ahead and let Chris get into it. Awesome. Thanks so much, Abby. And if anybody needs, obviously, beyond the the link to the webinar, if anybody would like, a link to the slides or anything like that, of course, feel free to get in touch with me. But as Abby mentioned, I am Griff. I lead growth here at Momentum. Our background and sort of why I'm gonna be talking about what AI still can't do and all of these things today. This is our this is our leadership here at Momentum. We were all basically former researchers, and we focus mostly on the behavioral psychology of giving and and the psychology of philanthropy. And you will see that that basically ties in really cleanly to, what we're gonna go through today. If anybody has questions as we go along or anything like that, please feel free to ask them in the chat. I think it's a much better, kind of more fun, webinar if people are asking as we go along, and I can, I'll do my best to pause it a couple times, and answer those as they come up. So, before we can talk about what is AI and what it can't do or what AI can't do rather, we should talk about what AI even is. And AI, of course, is basically an acronym for artificial intelligence. And one of the things that that sort of implies is that there is a natural intelligence out in the world. Right? I think, all of us kind of understand that we think of humans as being intelligent. We I think many of us would think of dogs, but I think it could it starts to get pretty interesting when you ask yourself, like, well, is a honeybee intelligent? Is an ant intelligent? And so, you know, over the last, really, over the last sort of 50 or 60 years, people have been working to develop systems that are, trying to basically emulate the intelligence mostly of humans. And the reason that I led with the introduction around our my background as a psychologist is that most of the, I think, best researchers and the leading leadership of, AI development has largely been, psychologists and people who study the human brain. I think a lot of folks in the space would say, that the human brain is sort of the most intelligent, organism or thing that we've ever observed in the world, in the natural space. And so it sort of makes sense that if you're gonna try and build a artificially intelligent, system that you would emulate the human mind. So these are 3 of my personal favorite researchers in the space. There are many, many, many more than this. Allison Gotnick, she's a professor at California Berkeley. She mostly is focused on the payers or, the psychology of infants and of children. And her research has really been, I think, one of the most interesting ways of sort of going in and examining the different types of, you you know, the way that minds work, especially in the very early developmental stages. She is the person who came up with, the concept of explore and exploit. Oh, sorry. And I may have lost my screen there. Did we did we lose the slides? There we go. There. We're back. So, she she studies children. Joshua Tennenbaum, he is, a professor of computational cognitive science at MIT, which just the title sort of explains it to you. And he's, he's really, really been focused on sort of, like, building something called that that emulates the human mind. And Gary Marcus, I think, if if folks like the talk that I give today and and wanna read more or just continue on with the the conversation, he would be a really good person, I think, to to look up. He has a Substack and some other things that he publishes that are really, really interesting. So, yeah, generally, there's there's a lot of stuff that comes out of that comes into the AI research field out of, the psycho psychology and neurocognitive science, domains. So I think probably most folks who are attending this webinar are at least familiar with ChatGPT, as sort of, you know, like, the the the thing that really brought artificial intelligence and the cognitive into the zeitgeist, over the last, what, it's been 2 years or so. But there's actually a lot of other types of AI that have been developed over time that I think would be worth sort of looking at and just sharing a little bit. So I chose sort of 4 examples here that are my personal favorites, AlphaGo, AlphaFold, Tesla, and Facebook. AlphaGo and AlphaFold, as the name suggests, are actually both developed by the same lab. That lab is called DeepMind, which now, is basically a subsidiary of Google. AlphaGo was actually just, like, sort of one of the first big AI breakthroughs and that it was just a robot that would play the game of Go, which is an ancient Chinese game that's sort of notoriously complicated. And it, basically, we started considering it artificially intelligent when the robot that they developed was basic was was good enough that it could beat the world champion in Go. And it could win games against high quality players. Obviously, if you have a computer that's an artificial entity playing against a naturally intelligent human being, and doing better, then you can start to call that intelligence. AlphaFold is one that is probably I was saying this to a friend actually before, before this. It's probably the most unequivocally positive example of artificial intelligence that's ever been developed. What AlphaFold did is it actually sequenced the human genome. It it is what made possible the the human genomics project, which has really unleashed unleashed a ton of, medical research and and breakthroughs in the field there. They they are all about sort of, like, mapping and diagramming how proteins fold, along a DNA sequence. Stuff that I really frankly don't even understand, but it's really, really, really, very interesting and really cool. If you wanna learn more, I would I would just Google AlphaFold. There's tons of good stuff there. Tesla and their self driving cars. You know, any car that can drive itself on the roads, I think I I qualify that as AI, certainly. And then Facebook. And I think a lot of people don't think of something like Facebook as being conventionally AI, but, there are really two parts of Facebook's, product and software that they make that I, I certainly would. And those are their news feed, which is really, really great at knowing sort of what the next thing that you wanna see is. It's an excellent predictor of sort of understanding what is coming next, what is the next most important post that somebody should see. And then, of course, their advertising algorithms, which are a little bit more controversial, but certainly something that, I think anybody who is sort of evaluating it just strictly from the standpoint of technology, what they've built is really very impressive and very, very powerful. So there's basically, there's lots and lots of AI that goes beyond just Chat GPT. And I figured I would spend a second just sort of going through a little bit of the taxonomy of it, of of the the field. I think people if you've heard much about AI in Chargept, you may have heard about generative AI, which I have in the middle here. But but what predated generative AI was just predictive AI, which is basically predicting the next thing in a sequence. Right? So, when we when we talk about that example of AlphaFold, which was, all about, sort of sequencing the human genome, really what that was was just a system got very, very, very good at predicting the next thing that would happen in the human biology and the human anatomy. And really, I guess it's just it's not even human. It was the whole anatomy. Then you had generative AI that came next. That is the Chachi the ChachiPTs of the world. You sometimes hear about large language models. Again, those are guessing the next thing in a sequence, but now they're really specifically focused on guessing the next thing in a sequence of text or of an image or something like that. Actually, the image on the screen here I had, made by an AI for me. I just sort of typed in, hey. Give me the taxonomy of AI and the 3 main buckets, and it spit this out for me and did an okay job. I I don't think I would have done much better myself. And then lastly, agentic AI. And this is probably something that unless you're, like me, you're fairly hardcore and in in the space, you you may not have heard much about. Agentic AI is, again, guessing the next thing in a sequence, but now it's starting to guess the next thing in a sequence of activities or or of actions. So when I say to Siri, hey, Siri. Can you, you know, can you fill my calendar up with all of the tasks that I need to do next week? Right now, Siri can't do that kind of work. If I asked her to sort of, like, book me out or or schedule stuff for me, it would she wouldn't be very good at it. But I think the next big development that many of us are are looking to the Frontier Labs for, is the development of this AgenTic AI where you can basically enlist an agent to go and and do things for you in the real world. And so that is on the horizon. I think people have long been speculating, but that's sort of the next big release that OpenAI will, will come out with, now to follow on on the the tales of Chechipti. Cool. So one thing that I figured would we would dwell on a little bit is just this predictive nature of how all of artificial intelligent works intelligence works. So imagine here that you have this giant circle and this circle, you know, I I have a few of things just like funny little examples labeled. But imagine that that is the entirety of the Internet. You have all of the text that's ever been published on the whole of the Internet, the whole corpus of of, you know, the the digital world, really. And what they did is they basically built a giant map where they put all the words they that appeared together or similarly near one another on this map. So in the little sort of insert, you have sofa, chaise, love seat, chair, all in the furniture section of the, of the map. Right? And that's really the core thing that underlies a lot of the technology. I think it's it's a little bit convoluted to think about, but just basically understand that, the way that any of the AI models that you see today work is that they're trying to build a giant map like this and then understand the distances between any two things. Right? And and they're doing all sorts of fancy probabilities and statistical modeling and all of that to to achieve that outcome. But the core thing is just that they're measuring the distance between sofa and chaise and low seat and chair. And that's how they that's how they really can are able to see sort of how, that's how they're able to predict sort of the next thing in a sequence of text. And so because they are predictive, AI is a huge massive prediction engine. It is necessarily not always correct. I think this is a thing, you know, like, people, who have had a lot of problems with chat gpt hallucinating. I found an example here online of somebody asking, you know, chat, hey. What's 2 +2? And it says 5. And it's one of those things that's sort of it's, like, kind of funny, because it's just so flagrantly incorrect, but, it it it goes to how sort of core prediction is to the whole thing. All an AI really is is a massive prediction engine. And it's very, very good, really, at a lot of things, but it is far from perfect. And it will actually there are folks I mentioned Gary Marcus at the top. Folks like Gary Marcus who feel like that is actually sort of a limit to the development of the field, that it can't we actually can't overcome this prediction problem, and that we won't ever be able to make make systems that that are perfect because we're fundamentally always making a prediction, and predictions always have some likelihood of being wrong. And I'll get into some of that a little bit later. But, one of the primary things here is basically that there's no such thing as a linear rule in an artificial intelligence model. And so, you know, if you type into a calculator, a calculator I mean, the one I have pictured here is probably from, like, 30 or 40, maybe even 50 years ago at this point. And you can type in 2+2, and that calculator will actually give you the correct answer. And you can type 2+2 into Chattopbt, which is, you know, at this point, one of the most sophisticated and most complicated, most cutting edge piece of technology ever produced. And it will give you 2 plus 2 equals 5. And so, there's something to be said for the fact that, you know, AI is is certainly not able to understand, the core rules and some of the core things that are just true in the world. It can only just make guesses, and sometimes those guesses will be wrong. And so that leads me to the number one thing, the the first thing that AI cannot yet do. AI cannot make decisions about something it hasn't yet seen before. That is a core thing. If you're asking an AI to to predict something that is new, brand new in the world, it won't be very good because all it can do is look back. And just to go back here, all it can do is look at its giant map of everything that it's ever known, that's ever been written, and try to guess basically what's within this, right, within the bounds of this. If you go outside of the circle, it won't really work. Oh, sorry. I guess I'm realizing you guys can't see my mouse, but I'm just I'm here drawing all over the screen with my mouse, and nobody can can read it. Anyways cool. So AI, you know, as I mentioned, it's it's it's a it's prediction engine, and that means that it's fundamentally about probabilities. Now I promise you I'm not gonna go too deep on on the bell curve or any of the statistics here. I won't bore you with that. All that's important to know is that, like, one of the oldest things that has been true in statistics since day 1 is that if you put all of the observations of anything in the world onto a bell curve like this, 95% of them appear within, 2 standard deviations. And anything outside of that is typically considered an outlier. Now as I said, AI is not good at guessing things that it hasn't yet seen before, and so it's very bad at dealing with these outliers. And so the problem is, basically, if 95% of all observations fall within the band, then one out of every 20 will be an outlier. And one out of 20 is just actually not that good of odds. If you think about, like, you know, if every single thought that you had every day or every single idea that you had, if 1 in 20 of them was going to be very wrong, that would actually be not great. Statistically, you know, that would be a a tough way to live a life. And so there's this core problem that is that exists in the field, that is basically that, no artificial intelligence can overcome the fact that, it's kind of batted out at atout virus. And here is one of my favorite examples of this. So this is, published in the journal, a few I think it was a couple years ago now, where Tesla basically came Tesla was was drive you know, it was in autopilot, and it came upon a tipped over tanker truck in the middle of a road. Now I can tell you confidently that if I were driving my car and I came on to a tip came on to a tipped over tanker truck, I would think that I would slow down. I would maybe pull over to the side, but instead, the Tesla confidently drove directly into the crashed, the crashed truck. And the reason for that is basically that they they had never had a tipped over tanker truck in any of the data that the Tesla autopilot had seen before. It was such a rare occurrence, so bizarre that it sort of basically sat outside of what the AI knew. And because it was something that was outside of what it knew, it didn't know how to handle it. And so it sort of went to its default, which was to continue driving forward. Now and so that's that that basically goes to the core thing, which is that, AI models cannot abstract out and think, okay. Here's something I haven't seen before. Here's a big giant thing in the middle of the road. It looks kind of weird. I'm gonna slow down. I'm gonna pull my car back. Instead, they just kind of continue doing whatever they're programmed to do, you know, ex ante. So that's that's sort of the the obvious example to point to. And, you know, sometimes this can have very real effects. This is, there there was a study that was done that, Tesla's semi autonomous driving, technology had killed 11 people in the US in 12 months. And that was you know, obviously, that's very bad and and, not exactly a testament to what we want out of our technology. And so I actually have a poll question. I shouldn't have loaded the deck like that, but, I'm I was gonna ask people, in the poll, and Abby, if you will, you know, how do people feel that, self driving AI has killed 11 people in 12 months? Should be a fairly straightforward question to answer here. Yeah. Exactly. Most people are answering very bad or very bad. It's sort of overwhelming. Cool. That was obviously me loading it up for for everybody to to, to see. I I appreciate the few sarcastic good votes in there. At least I hope those are sarcastic. Cool. Then if we basically go back along, what I'll say is, you know, obviously, it's it's a it's sort of an objectively bad outcome to have, any anybody, you know, dying as a result of of technology. And then if we can get the slides, but yeah. Sorry. It's very bad that anybody would die at the hands of technology, but, of course, 1,200,000 people die, from human driven cars every year. That's just a very standard statistic that's been pretty consistent. It's actually been trending up over time. And so it's very, very, you know, when when you think about it, it's sort of there's this almost double standard that we have, which is and and, actually, I was looking into this just because I I was, you know, doing research ahead of this. 50,000,000 you know, 20 to 50,000,000. If you take the 50,000,000 number of people who who suffer a non fatal injury, You know, globally, that's one out of every 160 people in the world will be hurt, you know, non fatally, but hurt all the same by a car in a given year, which is sort of a crazy, actually, statistic to think about 1 in 160 when you think about the entirety of the planet. But, anyways, there is sort of this double standard that, you see basically coming up where, there's sort of one what we think of our what we hold our AIs to and what we think of of AI being and and then, you know, really the realities of the world on the ground, which is that, on a per capita basis, I think it's fairly well played out at this point that, you know, the self driving cars are in fact safer than the median driver in the US. But this is the global risk perception. This was done from a study in that Oxam did where it was looking at sort of how people think about AI globally. And despite the fact that, you know, obviously, many more people die from human striving and stuff like that, in any given year. Generally, this what this is saying is that the the the warmer the color, sort of the more risky people think, that artificial intelligence is. And and you can see it's a pretty warm map. One interesting sort of data point here is that China is the coolest. You know, like, they're down at 1 point 0.1 here. You would assume that there's at least some problem with, surveying people on their risk perceptions in an authoritarian country, but, I did think that was sort of an interesting data point anyways. So, some of the long term limitations of AI, things that I think nobody really has a really good answer to at this point, things that are really, really hard, and and there's not really been any technology that has made progress on these. Anything having to do with complex decision making, anything having to do with understanding of cause and effect, and then maybe most importantly, ethical reasoning, and the ability to sort of evaluate risks on the ground, in real time, which is sort of a short way of saying all of the things that make us human. Right? I I mentioned Allison Gavnick at the, at the top, and her work is really focused on infants. And what she basically was able to do is is starting at, you know, really as young as 3 to a lot of her work is almost a little bit older, but as as young as 3 months old, a lot of the the basic understanding of cause and effect, object permanence, ethical reasoning, start to emerge in in children. And and it's one of those things that I think a lot of people feel may be innate. I think Alison Gottman's work has been, very, very important in arguing that it is, in fact, something that is not innate. It develops very, very, very young. And so there is this core basic thing to say, which is that one of the things that AI can't do is it will not be able to replace humans. There is all the all sorts of sort of anxiety or or something like that around job loss and and anything like that. And I think, one of the the core most things that I think has been has become very apparent to anybody who's in the field is that it is it may replace some of the most rote, some of the most boring tasks that people do day to day, But the ability to reason, at a sophisticated level, the ability to understand cause and effect, the ability to think ethically, all of those things are still, there's really no no technology that has made a lot of progress there, no matter how impressive chat gpt can be at trying to convince you, otherwise. And so for this audience, I figured I would just make a comment that, is sort of perhaps obvious, but sometimes I feel needs reiterating, which is that an AI fundraiser is not the same as AI for fundraisers. There's a lot of good tools and good technology out on the market. We make a product that we're very proud of that that deals with a lot of AI, but, there's really it's it's really important to say that when we look at at studies of donors and studies of fundraisers, which we sort of conduct on our own periodically, it's very, very apparent that people continue to give to people. People continue to give when there's positive relationships in the play. Of the most important things that any organization can do is to make sure that the that the donors that that their donors and supporters have positive associations with them. That's usually sort of part and parcel to them being a donor, at all. But, yeah, there there's something to be said. There you you will occasionally see stuff, people trying to basically build an AI fundraiser or something like that, an an AI x y z x AI job title. And it is my fervent position that that sort of technology is both not ethical, but also just actually not possible. Like, I I don't think it can be done in some sort of a core way. And I think that that will be a really challenging thing for anybody who tries to do it, and I wouldn't be foolish enough to try myself. Rolling along here. Checking the QA. No. Nothing. Great. The last thing is I have this metaphor that I've been kind of working on with friends, which is that AI is a little bit like, water, and that it's only valuable if it is in the appropriate container. So if you pour water all over the ground, what you have is in fact a spill. What you need instead is you know, if you're going on a run, you would want a water bottle. If you're sitting at a table, you might want a glass. You need something that, sort of creates a container around, the technology in order for it to be useful. Otherwise, all you have is just some chatbot that can sort of spit language back at you, and and it sort of makes sense, but it's not really doing anything of value. And I think that's one of the the core most things to think about with AI that, how what are the sort of what are the boundaries around it that you're building, or that you're putting inside of so that it can continue to serve the the means to to your ends, to to what you're trying to to ask of it. And so where we're at today, you have basically the chatbot, which is in fact a very good container. Right? You do need something so that you don't have to ask it. Like, you you don't have to know code or or know how to write, sophisticated software programming languages in order to be able to interact with something like Chat GPT. It it's very, very user friendly, actually, and it's really, I think, a a great success story. But there's a lot of other good containers out there as well. If anybody hasn't played with, Perplexity, I think that's a pretty good one. I believe the URL is just perplexity dotai. I'll actually put it in the chat. That's a really good one that's useful for research. They do a much better job of citing their sources and some of the things that I think have been have been levied against chat, over time. And then I think there's an opinion that is emerging, which is that, you know, using artificial intelligence and certainly sort of these LLMs, it's more of a feature than a platform. Right? If you think back to most of the examples that I gave earlier, Tesla builds a car, and a feature of that car is that it can, it can drive itself. It has an artificial intelligence that can drive itself, but it's not, it's still a car, 1st and foremost. Right? Facebook is a social network, and it has sort of an AI that can, you know, predict the next thing that you wanna see. But, fundamentally, it's only valuable if my friends are on Facebook. And, likewise, I think in in the fundraising context, you know, AI can do lots of things that are helpful for fundraisers, but, fundamentally, it it, it is not going to be able to replace sort of the the donor and organization or donor fundraiser relationship. And I think that's the thing that I really wanna stress and drive home for for people because I do hear that a lot, from folks. Cool. So wrapping up, I will just mention, you know, like, this is basically what our container looks like. We have a lot of all sorts of artificial intelligence built into this, but we do fundamentally believe and we work with Blackbaud very closely, that, you know, what fundraisers need, what professionals in the field make is, or what professionals in the field need is is a good container that is dedicated for fundraisers to be able to, you know, focus on their work, focus on their donors, and and spend more time interacting with their supporters and less time sort of having to deal with admin work and doing some of those road tasks. And so I think there's there's been space that has opened up to really apply some of the technologies to drive towards, you know, giving people a little bit more of a focus, focus time and focus energy and and less having to spend less time on on logging data and doing all of that sort of stuff. Cool. That is all I had to share. I was trying to keep it under half an hour, and I successfully did. I know that Abby booked this for 45 minutes, but, if anybody has any questions or anything like that, I am more than happy to answer them. Yeah. Let's see. There's a couple in the chat here. Thank you so much. Oh, sorry. Go ahead, Addie. Oh, no. I was just gonna say thanks so much. We really appreciate it. I know this is such a buzzy topic right now. And we do have a question, that came in. And just a reminder, anyone can put in questions in the q and a section right next to the chat. So we have someone asking, what about non frontline staff back end tasks like gift processing and etcetera? Yeah. So, that's a really good question. There are some things where AI might be really, really valuable and some things where probably won't be. So, like, gift processing. Right? Like, going back to, you know, like, my calculator slide. See if I can get there this. Gift processing, you kinda want, like, a something more like a calculator to be doing that. Right? You wanna know, hey. The dollars came in the door. They went to this, you know, maybe restricted fund or maybe they're unrestricted or whatever. And then, you know, we're gonna deploy them effectively, and I'm gonna write a receipt. Right? And so there's a lot of stuff that happens with gift processing that, that you actually wouldn't AI wouldn't be an appropriate application for. Some stuff that is really valuable though is we have a project that we're doing right now actually with, I guess I'm not I shouldn't say their name out loud, but a large state university in the Pacific Northwest where they have something like 80,000 old, you know, bequest records sort of plan giving donation, commitments. And they need to read those documents, but there's 80,000. There's a lot. And so we're gonna use basically AI to go in and read a lot of these documents since kind of collate some of the data for us and do some of that kind of aggregate aggregating work. And, you know, so I think there are definitely applications for back of house staff. I think there's tons and tons of stuff really where, you can use a mix of AI and automation to do a lot of these kinds of projects for you, but it's not going to replace good old fashioned, like, accounting. Right? QuickBooks is is a great software actually, and, you know, financial edge and all of those. And, there's there's going to be use cases for it, and then there's gonna be some spaces that I think are actually better left untouched. Thanks, Greg. We also have a question, just referring to momentum. Does this work with Raiser's Edge NXT? Yes. We definitely work with RE. We are very, very, very deeply integrated with it. And we everything that we develop, we basically develop under the pretense that, you know, you need your CRM to be your single source of truth. You need that to be the core most sort of data store. And then on top of that, you can build sort of the container, you know, going back to my metaphor, where you can can use artificial intelligence to to great effect. So, yes, if if you're interested in learning more about us or anything that we do, please, just reach out, give momentum.com, or my email is griffithgivemomentum. You're welcome to get in touch with me directly. Awesome. And then we have a question around, what do you recommend as best practices for prompt development? I've noticed that using AI chats is very similar to understanding how to use a search engine effectively. What are some things everyone should keep in mind when working within an AI chat platform that will help us get what we want or need out of it? Example, writing an introductory letter for donor cultivation, writing campaign content, etcetera. I love that question. That's that's that's definitely, like, a 200 class level question. But, prompt development I mean, my take on prompt development is you wanna have, you wanna be able to give it as much context as possible. Right? Going back to this one, the more you can sort of indicate what part of the map you're gonna occupy, the better it's going to be, the more effective it is. What we do personally is we basically will containerize all of our client's data so that we can put, you know, identifiable data up into the up into the AI systems without them being, you know, used to train future models or or taken by sort of the of the world. But I would say, if you wanna learn more about prompt development, send me a note after, and I can send you a bunch of examples. Or, actually, maybe I'll send a follow-up email that has a few examples of some really good, you know, some good examples of prompts that that we can do that that are, good good in the fundraising context specifically. It's difficult to talk about prompts, you know, like, because it's written, it's difficult to speak to them. Thanks, Greg. And another question probably a little bit along those lines, but just, someone mentioned one of the other limitations that we have come up is bias. How do we account for bias when interacting with AI, you know, ensuring inclusive language, preventing triggering content, etcetera? Yeah. Super good question. And, actually, that goes back directly to the comment that I made about sort of what AIs are trained on is what we already know. And, unfortunately, the bias, you know, reflects what what we already do. Right? Like, most of most of the Internet doesn't actually have inclusive language and and all of those things. And and so when you when you are using AI, one thing definitely is to just say explicitly, you know, like, you should use inclusive language like, and then give examples of, you know, respecting gender, you know, like, whatever whatever the examples are. You also can I mean, one thing that is said, and and this is basically sort of ubiquitous in the field is that it's very important that you have a human in the loop, a human who reviews everything that an AI, puts out there? And so you should never basically just take something that it gives you and copy and paste it blindly. You see all sorts of scandals around this. Lawyers who sort of, like, submit things to the court that say this was written by an AI bot. You should always be reviewing them. I think that's one of the most obvious. It it should go without saying, but it unfortunately doesn't happen often enough. And you should just be reading when you review stuff. You should be reading for, you know, for for bias, for inclusivity, for all of those sorts of things. There are they are now sort of working on tech that that will be more proactive, around addressing bias sort of in the models themselves. But, assuming that, you know, nobody on this call, including myself, is going to be doing anything directly developing sort of the new the newest stage of of LLM models, It it it's worth it's basically just what you should do is basically ask for inclusive language, ask for non biased stuff, and then review it after you once you have a response to make sure that you've gotten what you need. Thanks, Griff. So we have another one here in the chat just talking about AI adoption. Someone wanted to know, is AI adoption growing, and how do we know if our organization is ready to start truly adopting AI? Great question. So, AI adoption is definitely growing. There's there's really no question about that. If your organization is truly ready, that's always a very tough one. We actually publish a series of, basically, readiness assessments that you can take, and you can sort of ask around your organization. Hey. Are we ready to do this? In general, I would say, if you're if you're thinking about it, if you think you might be ready, you probably are. There's a lot of people who are just sort of fully you know, they're they're not ready to try it. And like I say, we we basically build it into our products. We believe very, very strongly that, you know, nothing should be done that isn't sort of reviewed by a human first. Nothing should be put out there. And so, we always make sure that that that is honored. And so you can always try it and then not not not sort of, like, deploy it, so to speak. You can adopt it in the sense that you're playing with the language, you're playing with the model, you're playing with, you know, like, one of the chat box or something like that. But you decide you're actually not quite you don't like what it's putting out there, and so you're not gonna sort of deploy it to something that would be donor facing, for example, would be, I think, the the use case for most of the people on this call. Great. And then we have one last question. What are the considerations around implementing AI safely? Another great question. Safe. So, I encourage everybody to check out, fundraising dotai. This is an organ an industry organization that we are actually on board of. Blackbaud is a part of as well. I'll just put that in the chat here. Or did I put that in the wrong place? Yes. I did. Here. I'll put it in the chat here. Now I should be in the right place. They again, we publish a series of, of, basically, guidelines to make sure that people are able to adopt AI responsibly and and that they can sort of deploy it without having to, you know, feel too concerned or anything like that. There should always basically be an adoption period where you start playing with it internally before you start playing with it externally. And then, of course, you wanna just make sure that there's this human in the loop. I think that's really one of the most important things as well. Great. Thank you so much, Griff. And I'm just going to, just a just a reminder for anyone, if you wanna drop any more questions into the chat, please feel free. And I'm just gonna share, a poll if you're interested and would like to learn more about our solutions here at Blackbaud and at Momentum. You can go ahead and share your response there. But if not, I think we are getting ready to wrap up. So thank you so much, Griff. We really appreciate you taking the time to walk us through that. I know that was really helpful information for a lot of us on the line. Yeah. Cheers. Thank you all for attending. Like I say, if anybody has any questions, feel free to to shoot me a note, and we'll talk soon. Alright. Thank you so much. Hope you all have a wonderful day.