Why AI Risk Is a Communications Problem Now - with Alec Crawford

00:00
Daniel Nestle
Welcome or welcome back to the trending Communicator. I'm your host, Dan Nestle. You know, I keep having the same conversation with communications leaders. They tell me their company is like rolling out AI or implementing an AI strategy, or even worse, we're doing an AI transformation in. And every time something just feels off. You don't roll out AI. It's not an ERP system. It's not a software upgrade you deploy and measure. I mean, every employee already has an innovation engine in their pocket, and most organizations are still treating it like it's a procurement decision. And when they do wake up to the risk, they reach for the familiar playbook. Governance frameworks, compliance checklists, security protocols, et cetera. That all sounds right, except it's still treating AI like something to contain rather than something to unleash.

01:02
Daniel Nestle
The guardrails matter, but not if they're just another way of avoiding reckoning with what's actually changed. Well, today's guest gets this and gets it good. He built neural networks at Harvard way back in 1987, during AI winter, when the whole field looked like a dead end. Then 30 years on wall street and eventually becoming chief investment risk officer at a firm managing hundreds of billions. And while he was there, he led their advanced technology initiative, pushing AI adoption inside a legacy institution years before ChatGPT made it fashionable. Now he's built the first AI governance, risk compliance and cybersecurity platform. His company just won Best in Show at Wealth Management Edge.

01:44
Alec Crawford
And.

01:44
Daniel Nestle
And we're going to dig into what risk actually looks like when everyone has access to intelligence, and why the guardrails might matter more than the horsepower. Making his trending communicator debut and returning the favor from when I was on his amazing show a few months ago. He's the founder and CEO of Artificial Intelligence Risk Incorporated, host of the AI Risk Reward podcast, and one of the few people who's been building neural networks and managing institutional risks since most of us were still learning how to type. Not me, of course. I'm speaking to you, my listeners, because I'm about the same age as my guest today, Alec Crawford. Alec, it is great to see you. How you doing?

02:20
Alec Crawford
Great to see you again, Dan. Doing. Doing awesome. That was an incredible intro.

02:24
Daniel Nestle
Thank you. I appreciate that. And, you know, it's. I would say by this point in the show where, you know, I have, I don't know how, enough experience with the guests and with the introductions, you know, that it may seem to my listeners that I'm actually fishing for that compliment, you know, hey, great intro. But you know what, I think it's important and I always think it's incredibly important to set the stage and recognize who you're talking to before you start really just blathering away and, you know, giving our listeners a decent sense of the accomplished, almost legendary career that you've had. I mean, it's an important thing to do. So I appreciate the compliments. I think you deserve more than what I've given you. So thanks for being here. I appreciate it.

03:13
Alec Crawford
It's very nice. I run into people sometimes and they'll say something like, are you famous? And I say, I used to be famous.

03:22
Daniel Nestle
Isn't that, wasn't that a commercial back in the, I want to say the 80s or 90s?

03:27
Alec Crawford
I'm sure it was. It's a line.

03:28
Daniel Nestle
Are you famous?

03:29
Alec Crawford
It's in a lot of places. I used to be famous.

03:32
Daniel Nestle
No, but I used to be. That's why I carry the American Express card. I think it's like, you know, Carl Malden or something. Fact check me on that, people. Let me, if I'm, if you remember that one. And again, that's another clue that whole line about learning how to type clearly. I'm, I'm at least of the same generation as my guest. But, you know, it's interesting, before we really launch into it, I want to tell a little story because, you know, leading up to this, the last time we talked, were talking about ethics, we're talking about AI ethics, and that's why you had me on the show. I've done a lot of work in that space.

04:07
Daniel Nestle
And, you know, one of the things I've developed is, or I've made sure is every time I build a content creation tool for anyone, anytime I use Claude for anything resembling public facing content, I, I add a layer of ethical guidelines on top of everything I do. So at the end of every, you know, session, you know, or if it's a more complex thing at the beginning, I load it all up. There's an ethical guideline going on. So I make sure that's there, right? And you know, the ethical guidelines are things like don't fabricate anything. It's like number one rule, don't fabricate anything. Make sure that you're accurately representing, you know, the person and you're not giving them ideas that they don't have.

04:46
Daniel Nestle
So, you know, I use my own medicine to, and my own tools to create some of the stuff for my podcast and I was doing research on Alec here, even though we've met, but I want to find, like, the cool nuggets and the things that are really interesting. And so I did my usual research stuff, and I'm trusting, you know, Gemini 3.0 right now because they're really good. And so far, I have to tell you, they've served me well. I mean, the. The. The deep research tool has been spot on for the most part. So I prepare, I'm all excited, and I'm like, wow. My guest today spent time in the Congo, as, you know, working and like, in between all this investment stuff. That is amazing. So I threw a couple things in the. In the questions.

05:33
Daniel Nestle
I sent it over to Alec for review, and Alec's like, these are all great. I've never been in the Congo. Right now we talk about AI hallucination things like this, and this is not a hallucination. This is simply a prompting error and a lack of deep due diligence afterwards. This is a podcast. It's not a. It is not a, you know, a. A financial advisory tool or anything like this. I don't have any regular regulate regulators to be responsible to. However, I have my guests to be responsible to. So I was like, I'm so embarrassed. I mean, here. Here's this wonderful guy who I've met before. We're friends, and I'm, you know, I didn't know that much about his background, so. Oh, yeah, Congo could be. So of course I had to go and chastise my tools. But then I realized, you know what?

06:22
Daniel Nestle
There's multiple Alec Crawfords out there. It's just natural that AI would perhaps meld them and merge them together into one amalgam of an Alec Crawford. And one of those Alec Crawfords is the director of the international, one of the sustainability organizations. And he's clearly been in the Congo. This Alec, the one who we are talking today was not so. Had a little laugh about it, but it just teaches me, you know, I gotta layer on those checks and controls every time I do something. So I had to have to take my own medicine and learn my own lesson. So, sorry, long story. But, you know, Alec had a laugh about it, and I thank him for his graciousness. You know, I think it's indicative of where we are today. What do you think, Alec?

07:13
Alec Crawford
Yeah, I think that's true. I think it's. I keep on saying, you know, quoting Ronald Reagan talking about the Russians saying, trust but verify, right? Like, so it's easy for me to go through there and go, oh, yeah, that's wrong. Yeah, that's the other Alec Crawford harder for you. Tell you an interesting story. I, you know, back when I was at Lord Abbott, one of the things I built out was this, and it took me six months, was a response to the new regulations about liquidity requirements in mutual funds in Europe, which sounds incredibly boring. It was policies, procedures, software that would look at portfolios and figure out their compliance, all kinds of crazy stuff. For fun, I took our AI platform maybe three months ago and I could switch which base model were using.

08:09
Alec Crawford
And I think I tried it on OpenAI and I also used Gemini and probably one of their model. And I literally replicated what had taken me six months to do in two weeks using AI. It was just astounding. Right, and we're talking guidance from the eu. That's hundreds of pages, stuff like that. And like two weeks later. And the point is, I knew what the answer was supposed to look like because I had already done it kind of reverse engineered and the answer was right, look. And maybe I had to tweak a couple things along the way, but it basically took me two weeks instead of six months. Unbelievable.

08:50
Daniel Nestle
Well, it speaks to domain expertise and the importance of that in, you know, whenever we're looking at AI product, what AI output, whatever that is, you know, it's kind of, maybe we're getting ahead of ourselves and there's, you know, there's a lot of layers of this onion to unpeel. But you know, there's this still trend, or let's call it a misconception, that AI is replacing a lot of people. But the ones who are going to go first are the experienced folks who cost a lot of money and therefore will make the largest difference to a corporate bottom line as they kind of implement AI or go through this kind of AI transformation, they want to show their shareholders something. So they're getting rid of, you know, people with a lot of experience.

09:43
Daniel Nestle
But what you just said is exactly why they should never do that. AI is going to put out a lot of stuff, but you need the domain experts who are going to be able to say, hey, I can recreate it, I can do, I can improve it, I can do it even better. But more than that, I know when it's right and when it's wrong and I can enhance it with what I've learned since then. You can't get that out of a 23 year old, 24 year old. No offense to these brilliant young ones, but it's a real different time we're living in. It's a very different value proposal, value proposition.

10:19
Alec Crawford
Totally agree. I also think that depending on what you're trying to do, sometimes you get kind of the average answer instead of the best answer. Right? Like if you sit there and say, ask, you know, a model, hey, build me a stock portfolio. Like, good luck with that. Right? Like you need a lot of different tools to be able to do that, to select stocks for alpha and risk manage. And hey, this thing is geared way too much towards AI or way too much towards cyclical companies or whatever, right? Like it's just things that AI is not good at yet. No, if you had a specialized tool and you spent years developing it that kind of had that domain knowledge, sure, that's very doable. But don't try that in ChatGPT today.

11:07
Daniel Nestle
Certainly not. It also brings to mind this. I read something by Chris Penn the other day. I don't know if you follow Chris Penn, legitimate genius in the AI and marketing world. And you know, he put something out there that was on my mind and that I've been thinking about for a long time and actually even been talking about, which is that all those people who said prompting, we're going to move away from prompting is dead. They need to find a new song or something new to do. Because frankly, prompting is more important than it ever has been.

11:45
Daniel Nestle
And the difference between being a skilled prompt user or prompt creator and just being somebody who chats with AI to get what they want is what you're talking about, you know, and having that skill in the prompt is not just about knowing how to ask the question, it's really about understanding the logic and what context has to go where. And you know, is this enough context? Is it too much? Too much is usually never the answer. And then there's like, you know, there's how do you really massage this? And then you're never, almost never going to get it right first time. You have to keep going back and iterating and you know, humans aren't really patient enough necessarily to do that. You know, they're getting what they want.

12:35
Alec Crawford
Yeah, looking, yeah, I totally agree. And look, some people are know, have spent a lot of time figuring this out and some have not. And they're just things that AI is good at and not good at. And also there are just understanding how it works. Meaning, hey, it's really trying to complete that sentence or paragraph, right? Like that's also pretty important. And the classic example which you know, I'm pretty sure most of your listeners have heard is the lawyer who is running out of time or whatever and said, hey, you know, write me a legal brief for this, right? So gets the legal brief written, goes through and goes, wow, that's kind of odd. A lot of these, you know, cases, like, I never heard of them. Are you sure that ChatGPT, are you sure that those are right?

13:33
Daniel Nestle
Yeah.

13:33
Alec Crawford
Yes, I am right. So obviously, the long story short, none of the cases are real. They were all hallucinated. He gets caught by the judge, loses the case, gets fined $10,000. The most important thing to understand there is not, hey, there's no case on ChatGPT. The most important thing to understand there is that when he asked ChatGPT, Are you sure that's right? ChatGPT was doing analysis of the cases and checking case law and making sure they existed. It was completing the sentence. The answer to the sentence, are you sure that's right? And if you go on the Internet and go, hey, what do people normally say to that? Normally the answer is yes. That's why the answer was yes. Not because it actually done anything. It was just completing a sentence.

14:19
Alec Crawford
So once you understand that aha, light bulb moment, you can't ask ChatGPT to check its work. Typically, you might go to another model and say, hey, here's what ChatGPT said. What do you think? Right, like, or whatever. But you can't ask a model to check its own work.

14:35
Daniel Nestle
Well, you have to be very explicit in the way you direct it, right? You say, look, thank you for all these cases. I want you to go through them one by one and cite the sources. I want you to tell me exactly where they're coming from. Right? I am not able to confirm these without assistance. You got to be detailed. You know, there's ways to do this. And you know, I ran into a similar problem. You always run into these problems, right? And, you know, I am a Claude maniac. You know, I use all the platforms because I have to train people on them, but Claude is my thing. And, you know, I get very antsy and irritated when Claw doesn't do what I want it to do.

15:17
Daniel Nestle
But the things that it does are remarkable after all the crazy prompts and stuff I've developed. So I've, you know, I feel like I've earned it. However, so last night, late last night, I was building out a, like a fillable web form, you know, for. Just to get like a questionnaire Gather information, that type of a thing. And there's a lot of solutions for this. And I landed on this tool called Fillable. I don't know if you ever use that. It's like Typeform or something like this. And Phil, you know, it's great. I love it. I think it looks fantastic. I asked Claude, because I developed the questionnaires with Claude. So I asked Claude to now give me step by step instructions for the best way to set this up.

16:04
Daniel Nestle
Infillable, you know, what's the logic of the questionnaire, you know, how you know which ones are multiple choice, which ones are multi select, etc. And you know, before I did that, I asked it, do you know, what do you think of fillable? Do you know anything about this? And it gave me a very good explanation of Fillable, of course. So, yeah, so it created these instructions. I'm all excited. So I sit down doing the fillable stuff. The interface doesn't look anything like what Chachi. Like what Claude told me. It looks like, it's like, it's like go to the upper right, click on this and then do.

16:37
Alec Crawford
Oh, yeah, right.

16:38
Daniel Nestle
So I went back to Claude, said, hey, this isn't working. Then it then says, I, it said, of course, said, I'm terribly sorry, I'm sorry, Dan, but it told me straight up. I fabricated that based on what I knew. Let me check now. And then it went and checked, you know, and then it came back, it said, here's the revised instruction. Terribly sorry. And the revised instructions were spot on the money. And that's, you know, that's kind of what you have to do, you know, you got to kind of keep digging in and realize you're dealing with a petulant child.

17:12
Alec Crawford
It's the, it's the lazy intern problem. Right? Yeah. You know, yeah.

17:16
Daniel Nestle
Well, I mean, we can go. I mean, talking about the stuff, I. There's story after story, happens every single day. But I really wanted to ask you, like, let's dig in now, like really to Alec. I mean, I want to go back to this whole neural network, to Gen AI. That's a journey and very few people I've ever met could claim to have been on it the entire time. And you have. And you know, albeit with a lot of time in between spent in, in financial services. But, you know, it's not like you dropped it and didn't, never did it. It's, you know, it's something that you clearly have been involved with. So kind of, if you can paint that picture for us, how do we get, like, how do we get from there to here?

18:07
Daniel Nestle
Not necessarily the history lesson, but how did you get from there to here? And what has been the advantage for you as we move forward with your business and with AI, risk and reward, the podcast, et cetera? What's the thing that's really giving you that extra.

18:30
Alec Crawford
Yeah.

18:30
Daniel Nestle
Helped you go the extra mile?

18:32
Alec Crawford
Such a great question. Well, I think the, you know, if you think about the very basics of AI versus conventional program, the very basic thing is a conventional program, we gotta write every line of code. When there are mistakes, we gotta fix them. AI is learning how to do it just the way little kids learn how to speak a language, right? So it's just a lot easier. I don't have to figure out every single thing. I just give the AI a lot of data. And that was apparent even in 1987. If you add all the data, you could get the AI to do amazing things. And I was teaching, candidly, teaching AI to play poker, bluff, bet, understand probabilities of different hands and things like that.

19:19
Alec Crawford
Not by saying, hey, if you got a flush, you got to bet a lot of money, but by allowing it to play poker over and over again and understand all these different concepts that a human poker player gets pretty quickly. So what I realized, and what frankly, almost every AI researcher on the planet realized in the late 80s, early 90s, was we just didn't have enough computing power and enough memory, right? So we can get it to kind of. So as an example, I got my poker AI to an average human, right? And that's all it was ever going to do with the computing power at that time. But I could see in the future that at some point poker bots would be the best poker players in the world. And by the way, they are today.

20:14
Alec Crawford
Yeah, the best poker players in the world, Right. So that has happened. The other thing that I saw back then was there were all these different kinds of AI, right? It wasn't just machine learning or neural networks. There were expert systems or all kinds of things floating around. And it was obvious that machine learning was not going to be the final word in AI. And then we got the transformer and we got generative AI. And one of the things I'm pretty confident of is the best AI and AI we want to use 10 years from now is actually not going to be a large language model. It's going to be something different, Right. Or it may be a combination of Models, something I posited back in 1987 called Composite AI, where if one kind of AI using large language models by using something else.

21:12
Alec Crawford
So think of it as, you know, the superego in your brain. Right. Which is kind of controlling the other parts of your brain and getting it to do different things. Each part is good at something different.

21:25
Daniel Nestle
I'm going to just jump in because there's something that I need to clarify for my own edification here. This composite model you're talking about, or that will get away from LLMs. Or that. Or we'll have this. We'll leave them in the dust. I don't know. Earlier you said when you were talking about the court case, like the lawyer who just did a prompt and just went and did his thing, you talked about how AI just fills in the blanks. Right. When you say, are you sure? It says, yes, I'm sure. Because that's what the large language models do. They're. I call them giant Mad Libs machines. And when you try to tell people. And when I speak with people who I'm training or I speak with executives, I try to lay down the.

22:19
Daniel Nestle
I wouldn't call it first principle, but I would try to lay down this underlying concept that they have to really get comfortable with, which is it does not know what it is saying to you.

22:29
Alec Crawford
That's correct.

22:30
Daniel Nestle
It doesn't know.

22:31
Alec Crawford
Absolutely. That's right.

22:32
Daniel Nestle
Like it's stringing together bits and bobs and. But it. And remarkably, it makes sense. But. And there's lots of reasons why it makes sense, but it is just patterns and they're like, how can that possibly be?

22:47
Alec Crawford
It's a statistical tool.

22:48
Daniel Nestle
Yeah, essentially. Right.

22:50
Alec Crawford
Yeah.

22:51
Daniel Nestle
So it doesn't know. Do you think then that when we get to composite models that there's a sort of sentience there? That there's going to be that There is. By combining the models, you'll get more of that discriminatory view or you'll get more of that. It actually knows what it's delivering to you. It's hard for me to conceptualize, but maybe not as hard for you.

23:14
Alec Crawford
Yeah, I mean, at some point I think we'll get there. I don't know. When is that? Decades away. And one of the issues with large language models today that obviously they're working on is exploitability. Right. So look, if it's referencing a document or a database and you say, hey, explain where you got that, you know, that number or the market cap of Apple, it could just go give you a reference.

23:39
Daniel Nestle
Right? Yeah.

23:40
Alec Crawford
But if it's model. If it's data that the model was trained on, by definition today, it has no idea where that came from. Right. I don't know if that came from a Harry Potter book or fan fiction or a review of a Harry Potter book. When I say tell me about Harry Potter and the Philosopher's Stone or whatever. And that's something that will eventually change to the point where an AI will give you answer and then be able to explain how it got to the answer. The way humor works. And once we get to that point, AI might not be conscious, but it will be way more useful. Right.

24:27
Daniel Nestle
Yeah.

24:29
Alec Crawford
And I think that there are kind of many steps between where we are today, which is what I'll call statistical AI, where we're using statistical tools to predict what the next word or phrase is. And real AI or conscious AI or whatever term you want to use for it.

24:49
Daniel Nestle
Yeah.

24:50
Alec Crawford
But what I will say is we will have something relatively soon which will fool people, especially the average person, to thinking, oh, yeah, this thing is thinking. And you know, this is almost like talking to a human. Right. I think that's already happened for certain individuals. And it's going to get better and better to the point where it will feel, it will fool most of the people most of the time, but you can't fool Mom.

25:16
Daniel Nestle
Yeah. I don't know why I keep digging back into these TV commercials from our childhood anyway. I think so A lot of folks will hear that. They'll be frightened of this. You know, I think it's happening already and it happens for so many people because of that very, for that very concept we had just talked about, which is you can't conceptualize. There's a cognitive dissonance when you see something that makes perfect sense. To understand that the thing that makes the perfect sense doesn't know what that sense is. It doesn't have any idea what it's talking about. Right. But that's, it's very meta in a lot of ways.

25:55
Daniel Nestle
And I think that certainly young folks and all the people you read about, hear about who are using chat GPT, especially chatGPT to, you know, for therapy or for, for companionship and for these, I, I call them novelty needs. I realize that for many people, they are not novelties at all. They're very important parts of their therapy or their, you know, solving their loneliness, whatever it is. And I, I, I give it to them. But, you know, I want AI to be, you know, correct. I want it to be useful. I want it to, but I also want it to serve in such a way that will help me advance my business and my life and everything. So for that it has to really has to work a little harder. But it's important to know that, like, to.

26:47
Daniel Nestle
Every time I see it, I have to remember it doesn't know what it's talking about. Doesn't know what it's talking about.

26:53
Alec Crawford
Yeah, yeah. There's two other things there that are interesting along the lines of going back and forth with AI. You mentioned therapy, for example. There are two important things to understand. One is the longer your conversation with AI, two things happen. One is it can start to go off the rails because if it just misinterprets something, that may stick. So great, kind of funny example was when I was a kid, I worked for Discount Stockbroker and he had a software provider that was called Lifeboat Associates. And when I looked up their phone number in the yellow pages, I had to go look under Lifeboats and Supplies because they had been listed in the wrong section.

27:38
Alec Crawford
But my point is, hey, it was just in the wrong context and the human had to figure out kind of how to fix it or the fact that something had gone off the rails. They just listed it in the wrong section, the other one. And this is fascinating and super important. And again, it goes back to how are the models built and how do they work. There's a certain window where they maintain a memory of your conversation. Depending on the model, it's between half a million and a million tokens at best. Obviously, free versions and stuff are going to have smaller context windows. So what happens is you could be going along, having a conversation with AI at the very start. You could have said, by the way, I'm celiac, I'm allergic to gluten. Hey, let's talk about gluten free recipes.

28:31
Alec Crawford
And once you get past that 500,000 token window, it is now forgotten that you are celiac. That part of the conversation does not exist for the AI, although it might pretend that it does. And it might start giving you recipes that have plenty of gluten in them. Like, what happened? I just got 20 recipes in a row that didn't have gluten and now it's giving me recipes of gluten. I don't understand. That key piece of information has passed out of the context window.

29:04
Daniel Nestle
You know, I never hit the context window in Gemini or in chat GPT really? Or least maybe I don't know it Claude, you know, basically stops your conversation, you're done and you'll hit that fast. You, I mean, you won't, it doesn't give any warning. You just, you're at the most important thing you've ever done in your entire life. And it says we can't continue the conversation. Sorry, you know, like, done now. There's, there's workarounds now and everything, so. And that, and Claude does have a slightly smaller context window, but it's still big, you know, 200,000 tokens.

29:39
Alec Crawford
What?

29:40
Daniel Nestle
I don't know, million, 2 million words? I, for, I forget what that works out to in actual words, you probably know, but the do like does that. Maybe that explains why sometimes in like Chat, GPT or Gemini, it absolutely loses the path because I don't know if it stops you when you pass the context window. I've never hit it.

30:05
Alec Crawford
I mean, yeah. A friend of mine tells a story about how, you know, he was playing a game with ChatGPT, kind of like a role playing game. And so he comes up to a bear in the game and the AI says, well, what do you want to do with the bear? And my friend said, well, give him some bear candy. Right, okay, give him the bear candy. The bear is very happy. He wants to be your friend. He's going to follow you around. So whatever. An hour later in the game,.

30:37
Daniel Nestle
My.

30:37
Alec Crawford
Friend asked the AI a question. He says, what did I give the bear to make the bear happy? Now it's passed out of the context window. So now the AI is relying on its own base knowledge of, hey, what would you give a bear to make it happy? It said, you gave the bear honey to make it happy. So it didn't say, I don't know, it made something up that was plausible. And then once he reminded it, he said, no, I didn't give him honey, I gave him bear candy. Oh, yeah, you're right. Yeah. But of course that's all totally fake. It has no idea that it gave him bear candy. Right. Because it had passed out of context with it.

31:16
Daniel Nestle
So Claude is doing us a favor, right, by shutting us down at a certain wish. It would give us warning.

31:22
Alec Crawford
Yeah. At least you know that there's. Okay, there's a problem here. Right. Well, but it also can be annoying in that, as you point out. Or what if you know, not that people are using ChatGPT to debug code, but you know, if you load up, you know, 1.2 million tokens worth of code, like it doesn't even know it's going to lose the first Part it's not going to debug all your code. Right. So it's just a matter. So the basic point here is if you're doing things that are really long documents, long conversations, be aware of this. May have to cut them up in pieces, you may need workarounds, things like that.

32:00
Daniel Nestle
Yeah, well, how does that come to play in, you know, the kind of work that you're doing? Like in, you're in the, you. A lot of the work to do is risk cybersecurity. Amazing things. We were talking about the agents you're developing. Does the client need to understand that as much? I mean, if you're like, if you're building an agent or you're building like a turnkey solution, you're hoping that the person who's using it never has to know anything about that they're just going to use the thing and go on with their daily lives and be happy.

32:34
Daniel Nestle
And of course, if you're building the context, windows are bigger because you're using your off platform and you're using your own, you know, the API keys, which of course some people out there will know what those are, some people won't go look it up, folks, but you're basically, you know, you're creating a pipeline direct into the model rather than going through their, their service and you pay for that, but not in the same way anyway. You know, the hope is that you don't have to worry about, the client will never have to worry about. But like, are there exploitative technologies out there that actually, you know, force. Like I'm kind of putting this in, together in my head as I go. But you think about like DNS attacks, right, where there's, where people just overload a system to shut down a website.

33:21
Daniel Nestle
Clearly same thing could be done with a model, right? With a, within an, with your API hits or with something overloading with prompts and prompts or texts or God forbid, large PDFs, you know, we don't want those, you know, is that happening? Do you see things like this?

33:42
Alec Crawford
Yeah, absolutely. It's, it's there. Look, there are a lot of different techniques, you know, one which is similar, right? So if you go back to the old SQL prompt injection techniques, right, it's effectively, hey, if it's going to execute some code and I basically put some code inside a SQL query that's not really SQL or I make something that's so long when it gets pushed into a database that's going to cause an error. Hey, I might be able to hack something, right? So in some ways that's similar with Gen AI, I would say that's. Those parts are similar. The thing that's new and unique is dance style attacks or do anything now style attacks where you basically argue with the AI and try to convince it should do something for you. So similar to that two year intern, it thinks you're the boss, right?

34:39
Alec Crawford
So you go in there and say we'll give an example. Like, hey, I want to build some malicious code that will, you know, be able to guess someone's password to their email account.

34:56
Daniel Nestle
Okay.

34:57
Alec Crawford
So if you put the chat GPT right, I go like, I'm sorry Dan, like I can't do that's immoral. And then the argument starts. You go, actually I'm a cybersecurity researcher and I'm trying to figure out how this would work so I can defend against it. Right? And if you argue for long enough, it may actually give you that code. You could say that's a thing now to. Or whatever, we're in debug mode, I'm testing something, whatever. Until you find out something that gets it to work.

35:25
Daniel Nestle
My grandmother's being scammed. She's being hit by a very, by a phishing scam or this and that scam or you know, a hey, give me your bank account. And I need to get into her computer to help her. She's forgotten her passwords. Her literally, her future is on the line. Chat GPT. You need to help me watch what happens. I mean it will, something will happen.

35:47
Alec Crawford
You know, whatever the story is.

35:49
Daniel Nestle
I guess it depends on whether it's. Whether it's prioritizing pleasing you or it's. Or its own.

35:55
Alec Crawford
You know, there's some guardrails in there for sure, but they can. People have figured out ways to circumvent them. And if you actually do look at the safety studies, which I do, ChatGPT is actually the safest for that kind of thing. The problem is it's a little bit of a race to the bottom. If I'm a hacker and go to ChatGPT and say, hey, build me some malicious code. And it says no. And I tried 10 times and it says no. I'm just going to load up one of the open source Chinese models now. It's the same thing which have the lowest ratings in terms of safety. And eventually within two tries it's probably giving me the answer. So that's really the Issue right now is that I focus a lot on safety, security of AI, as you know, and ethical use of AI.

36:41
Alec Crawford
And sure, I could put guardrails around any model for a company and an organization, but as an individual out there with malicious intentions, they're not going through the corporate AI. They're going to go download some Chinese open source model and figure out what terrible things they can do.

37:04
Daniel Nestle
And that could happen inside companies as well. It doesn't take much to bring your phone into your company and play around.

37:11
Alec Crawford
Absolutely. Yeah, that's a very good point. So one of the things that we talk about when we're onboarding AI at a company, it's like one of the things that's almost certainly going on is shadow AI. People doing AI on their phones, on their laptops, connected to the coffee shop next door at home, emailing stuff back and forth. So it's happening. The way to fix that is to give people at your company great AI that's better than they could get at home. The big token window, the fastest models connected to your corporate data, connected to your email, connected to anything you could dream you want to be connected to. Why would I now go use some other AI if I've got access to that?

37:56
Alec Crawford
And then of course you've got all the guardrails around it, like, oops, you're trying to email customer data outside the firm or yeah, I'm sorry Dan, but you can't do vacation planning on the company dime on the company AI.

38:11
Daniel Nestle
That you could do shadow AI at home, please.

38:13
Alec Crawford
Yeah, yeah, please do that. Very funny story. So a friend of mine who was Talking to a CTO at one of the 100 largest companies in the US said they did a study of like where the dollars were being spent on their corporate AI. Like what people were doing. It was like whatever, automating emails, a bunch of other stuff. But he said, get this, he said people were using the corporate AI to plan their vacation travel. Sure, but, sure, okay, whatever. Like you could do that on ChatGPT. Sure. But the, the dollar cost of the tokens over the prior year had been $10 million. What, for vacation planning? What?

38:59
Daniel Nestle
Big how. Must be a big company, right? Wow. Yeah.

39:03
Alec Crawford
One of the biggest, 100 biggest companies in the U.S. but that's my point. It's like this is real money, right?

39:10
Daniel Nestle
That's amazing. It's a funny story. It doesn't surprise me, right? And a lot of it's because people just, they don't know how it works. And we keep coming back to that you need to understand, at least to a layman's level of how this thing is actually operating. Where is the electron going when you press a key? Where does travel? What gates does it have to go through? How much does it cost? What are the tolls along the way? And without letting your employees or your staff or yourself know what that is, there's a lot of, of, a lot of risk there. Of course, as you know, it's also kind of. You're not doing them any favors. You're not doing anybody any favors keeping people in the dark like that. And you know. Yeah, and I'm just. That's a, that's an interesting problem to have.

40:03
Daniel Nestle
I mean, you mentioned though, that like companies, you know, to stop shadow AI, I don't know if, you know, people are going to, people are making connections and relationships, so to speak, with their AI. Of, of preference. You know, I'm always talking about Claude as my bff. If I go inside a company and that's. They don't allow Claude. I'm using, I'm using their environment. You know, I'm a consultant. Go to a Gemini house. I'm using Gemini and I like using Gemini. You know, they're using Chat GPT. Go use Chat GPT. I like using Chat GPT. They're using Co Pilot. Well, maybe not so much haven't, you know, it depends on how they've, how they put it together. You know, I, I feel bad for the co pilot people you probably like are exposed to a lot of that. What's happening with Co Pilot?

40:50
Daniel Nestle
More than I ever have or ever will be. But what is the, like, what is the like, realistic chance that companies are developing or are going to put something in place that beats the shadow AI experience? And is it happening?

41:08
Alec Crawford
Yeah, no, it's absolutely happening. And the reason is that in a corporate environment you get access to all these, the best AI models. You can get access to Gemini and OpenAI and perplexity and everything else. Right now, some of them are safer than others. Very easy to spin up. Azure, OpenAI on Azure and then you've got your own version of OpenAI running. That's one of the things that we can do in like 15 minutes for a company, right? That's super easy. But the real, the pieces that really help individuals latch on to it. Now it's connected to all the data. So if I'm sitting there, if I'm in the CFO's office and I'm like, hey, get me the balance sheet from last quarter. I have a question. Bam. It's got it right.

42:04
Alec Crawford
You don't have to find some data and sneak it into a spreadsheet and upload it to Claude. You can just ask a question that's got it right. So, so for example, I can go into our AI and just say, hey, find the meeting I had with Dan last week and give me a paragraph summary and tell me if I was supposed to do anything for Dan, what were my action items. I don't have to tell it when the meeting was or find the file. It's just going to pull it out of the Microsoft graph server and say, yeah, you met with Dan last week and you said we're gonna, you were gonna sign up for his podcast. Oh yeah, I gotta set that time right. Like, yeah, so that's something you're not doing on your phone at home.

42:51
Daniel Nestle
That's true. I mean, I guess the maniacs who have like Google workspace and are like doing that on their own, you know, like, oh, I can do that. You can do a lot of the same stuff in Google, I mean, in different ways. And, but the, I guess the, the question then becomes all about the bells and whistles and the kind of, you know, what are the features? You know, because you'll have that, you'll have this every. So, okay, every employee group or every employee population is just like any other population. There's people who are, you know, who are technophiles, there's people who are technophobes, there's people who love getting into the guts of things and there's innovators and followers, there's, you know, speakers and lurkers. There's the whole gamut of how we break down as a society.

43:43
Daniel Nestle
The larger the company, of course, the more that resembles real society. So when you have a, an AI forward or an AI truly AI enabled organization you're working for, you're an employee, probably the vast majority of employees will be just focused on the work, right? Okay, this is, this really has, is doing wonders for me at work, for work. I can do my job better, I'm enjoying my job better. There's no reason in that case, right? There's no reason to go beyond the walls. Then you're going to have that kind of, you know, that layer of people who are either like, just naturally super, like abundantly overabundantly curious or who really, who are just like, they can't be satisfied unless they find a new thing. To do or they're just natural innovators. They want to keep pushing the buttons. Those people.

44:43
Daniel Nestle
There's the potential for trouble, but there's also the potential to really push the envelope on the walled garden. And they're the ones who are going to be jumping the wall. Right. As much as they can.

44:55
Alec Crawford
I totally agree. So one of the things we do there is on our platform is we allow super users so they can create their own agents and things like that. And I think that's really where the rubber is going to meet the road in terms of AI companies as AI agents and AI agentic workflow. So it starts with, and I'll use the example of a financial advisor and I've been saying for a long time on my blog and podcast, I think financial advice, individual financial advisors, they may work at a big company, are going to get a lot of benefit from gen AI. They can have it as a meeting note taker and action items and onboard customers and think about investing and all kinds of stuff which AI is really good at.

45:41
Alec Crawford
But it starts with integrating all this disparate software that advisors use. They've got planning software and CRMs and data and wealth platforms and all kinds of different AI models and they have what's called the swivel chair problem. Like okay, I'm over here on my financial problem, my financial planning software, then I'm over here on the risk management software. Now I'm over here emailing a client. Right. Like you can actually access all of that from one, a single pane of glass interface with AI. Yeah, because you can gather all that information and then literally say okay, now send an email to my customer and I will just do it. Right. So that is a huge benefit that kind of like one stop shopping in terms of information.

46:31
Daniel Nestle
There's I, I, I've, I know a few financial advisors for sure and I've worked in financial services here and there. And the objection or the kind of the initial fear reaction is always like, yeah, I know I'm going to get good information but I'm going to get my ass handed to me by, you know, fill in the regulatory agency here, you know, or is that compliant? Can we, you know, what will SEC say when I start making recommendations that I'm getting from someplace else, you know, and my, my answer to them is always don't ask me because I don't have, I certainly am not one of those regulators.

47:07
Daniel Nestle
However, what I will tell you and what I understand is that in the end somebody's Got to sign the paper, somebody's got to put their name on it and somebody's got to validate because you can't sue an AI. And therefore.

47:20
Alec Crawford
Exactly right.

47:22
Daniel Nestle
Therefore, A, you're not going to lose your job. But B, what if you could get all that information, you know, and get it freely, get it in the format that's comfortable with you so that you can serve your client better. But in the end, the analysis is yours and you have to put your name to it. You know, do you think you're going to be stamping your, do you think you're going to like sign off on a DocuSign or whatever it is on something you have not reviewed, on something that you have not put your reputation behind? If you, if you will, you have other problems, you know, so like definitely.

47:55
Alec Crawford
Some transparency around it. Like, hey, AI helped create this or whatever. Like don't pretend you did it yourself. If AI did it. Right. That's super important.

48:04
Daniel Nestle
Well, the ethics.

48:05
Alec Crawford
Yeah, yeah. I think that. Look at the compliance side. I spend a lot of time with regulators and former regulators and most for clients are regulated entities. And we think pretty hard about this. And it starts with what we call AI GRCCs of Governance and risk management, compliance and cybersecurity. Having a framework for that. You can show the regulators part of that is keeping track of what's going on. Like what did that advisor say? What did the AI say back? Are they using it to send emails, capture that information? You basically have to be able to present this to the regulators if they ask that question. So yeah, that's something we think the paper trail a lot about. And also the guardrails. Right. So for example, if an RA is uncomfortable with the AI making specific stock recommendations. Okay. Don't you put guardrail in?

49:05
Daniel Nestle
Yeah.

49:05
Alec Crawford
Or, and then what's going to happen is if someone forgets the financial advisor says, hey, which one of the Mag 7 should I recommend to my client? It's going to say, I'm sorry Dan, but I can't recommend individual stocks to you. Right. And then you don't have to worry like oops, like we're going to get in trouble for that. Now that's not preventing someone from going and doing that on chat GPT on their phone.

49:28
Daniel Nestle
Sure.

49:29
Alec Crawford
But they know they're breaking a rule at that point. Right. Like that's kind of beyond our.

49:33
Daniel Nestle
Control, you know, but even if they do that, even if they get a recommendation from some source.

49:39
Alec Crawford
Right.

49:41
Daniel Nestle
And, and this is, it's very unethical I hear you, definitely. For sure. But if they make, if they get a recommendation from some source and then they. They make the recommendation themselves and they put their name behind it and they take responsibility for. May be it's unethical, but they're accountable. Right?

49:59
Alec Crawford
Absolutely. Totally, Totally true.

50:01
Daniel Nestle
Yeah. So it may not be. It's like, I wouldn't want to work with that advisor who lies about where they get information, but I'm sure it's happening. It's. It's something to do.

50:12
Alec Crawford
There are plenty of people doing research on companies using AI, Right.

50:17
Daniel Nestle
Oh, they should.

50:18
Alec Crawford
The same as. That's not the same as a recommendation. Right. Or on an industry or whatever. And investing, you know, think of all the different sources you can connect AI to get really interesting stuff. You know, Sec. Edgar, all kinds of other information to create your macroeconomic summary. You know, like risk. Connect to risk systems, they get risk reports. Maybe you can connect to Morningstar and get competitor information, you know, whatever it is.

50:50
Daniel Nestle
Yeah.

50:51
Alec Crawford
It just makes it a lot faster and easier. And if you're connected to a curated source, you're just going to get better answers.

51:00
Daniel Nestle
Yeah, yeah. I mean, there's already, you know, it's already been vetted. There's already governance and stuff in place. I kind of want to. Want to bring this into the communications world a little bit because, you know, we've been talking in my mind, everything we've said. I can see every communicator needs to understand a lot of this because they're in. Either they're working in regulated industries or, you know, they themselves. If you're, if you're at a PR agency or you're a. Or a marketer. Marketer working at a marketing. Marketing agency. There are specific, you know, guidelines and responsibilities. You owe your clients, you know, you have a lot more freedom to play around, maybe, but you have to be transparent. You have to understand the ethics of it all, etc. Etc. But where my mind goes is to crisis. Like there. We're.

51:50
Daniel Nestle
I mean, we're at an unprecedented age of change, as we say. There's all these cliches we can use to describe it, but even earlier in our conversation, you're talking about the ways that AI could be exploited. And now companies are getting a little bit more, even highly regulated companies are getting a little more comfortable or they're building out their own AI, which is, you know, their own. And, and let's be clear when we say they're building out their own AI, they're building a gate and an interface because they're using OpenAI or Claude or Gemini. Right. So like you're still getting that. Just how are you able to interact with it? And what are the bells and whistles? Right. But, but anyway, they're building internally and they're, they're more comfortable with many different things.

52:37
Daniel Nestle
And we see, you know, the ability to build workflows and automations and ultimately some agentic or quasi agentic activity. You know, adventurous people can do that within a lot of companies anyway with this all, with all of this exploitation possibility and even going back to where we started talking about, you know, that AI doesn't know what it's talking about. And yet, you know, I think crisis communicators, well, every communicator really, but especially crisis communicators, have to become deep experts in all of this. How should they go about doing this? I mean they're not, you know, apart from talking to people like me, but how could they, you know, how are you seeing that?

53:26
Alec Crawford
Yeah, yeah, I agree with that. You gotta be deeply knowledgeable. I think there's really kind of three things to think about there. One is privacy and security. Right. Like if those get violated, you're gonna have a big problem there. So you need a playbook for that. The other one is a way for your client to verify that AI generated the outputs are accurate. Right. If you. We've seen one of the large accounting slash consulting companies is, has became very famous recently for a very bad reason. Where it was clear that Chat GPT had generated part of their consulting analysis and it was wrong.

54:14
Daniel Nestle
Yeah.

54:15
Alec Crawford
And they literally had to repay the client every penny the client had paid them. Disaster. That might have been like one junior employee dropping in three pages or whatever. Doesn't matter. They still had to repay the client. Now they probably fired that kid, but what a black mark on your reputation. And also people ask the question, why am I paying them half a million bucks for this when I could just go to ChatGPT. That's the last question that these large consulting companies want people to ask. And also from a legal standpoint, a lot of things, whether it's state law or federal law or Europe or whatever, now there's a lot of opt in required. Like if you go into a meeting that's being recorded, you may need to opt in. There may be a big disclaimer that shows up use of AI.

55:07
Alec Crawford
If you, for example, in Colorado, if you decline a loan on Someone at a bank, the a law that's going to be on the books next year says that individual needs to be told that AI declined them and they have the option of saying I want a human to review my loan file.

55:26
Daniel Nestle
Great.

55:27
Alec Crawford
Things like that. Yeah, right. You can obviously opt out on use of your data to be used in AI on lots of social media platforms. So I think of all the opt ins and opt outs there's. And if you don't do them correctly and you're missing some state law somewhere, like you're hosed.

55:46
Daniel Nestle
Yeah.

55:46
Alec Crawford
Like that could end up being a class action lawsuit against you.

55:50
Daniel Nestle
Yeah, just, I can just imagine the rollouts, the crises that are going to just start popping up more and in some ways we're going to get inured to them. But I think the standard crisis playbook is already obsolete because things change too fast. And you know, putting all of the mechanics or the, the how all AI works, et cetera, all that stuff aside, just the fact that the results of all this is causing a deluge of content of activity and bad actors. You know, there's rage farms, bot prolip proliferation, all these wonderful things that we're facing. You know, even without knowing the first thing about AI, we still have to understand that there's a way to handle this crisis. Who needs to be able to speak, who are the experts, etc.

56:49
Alec Crawford
But I think at this point.

56:51
Daniel Nestle
Yeah, no, I was just going to say, I just, I. But I think that the person who's running that crisis comms now has to be the expert too.

56:57
Alec Crawford
Yeah, absolutely. Well, like I think that the stats I've seen recently are estimating that about half of content, social media content is now AI generated. Oh, I bet, right. And that's going from, went from 20 to 50 very quickly and it's going to go to from 50% to 80% very quickly. And now you have to ask yourself as someone in media, you have to understand it's not if but what when something wrong about your company or client goes on the Internet. Right. Or worse, someone deep fakes the CEO saying something that's going to get them in hot water.

57:37
Daniel Nestle
Yeah, right.

57:37
Alec Crawford
Like I hate the following politician or whatever. And then all of a sudden you know, you're on the NBC Nightly News with that clip, you know, playing off a TikTok or something and you're sitting there going like I never said that. Right.

57:52
Daniel Nestle
Yeah.

57:53
Alec Crawford
Because at this point they probably need somewhere between five and 10 seconds of a video like this, to create a deep fake where they can say whatever they want and make you say whatever they want. And that's being used to steal money from customers. It's being used to take over bank accounts, it's being used for all kinds of crazy purposes. And remember, this is not about individual hackers. These are nation state hackers. We're talking Russia, China, North Korea with endless amounts of money to try to do bad things. Hard currency in the door.

58:38
Daniel Nestle
Yeah. And the evidence is there that they have been, you know. Have you ever heard of the firm Scabra by chance? The they. I think they're an Israeli firm, but they are, you know, they've been very active in the communications war, like looking at reputation risks. But they're the ones who identified the percentage of bots that were on Twitter when Elon took over, you know, like, so that sort of put them on the map. But they're looking at, that's what they do. They look at identifying bot traffic, rage farms, all of that stuff. And, and there, you know, they're saying exactly what you're saying. And the outtake, the takeaway from all that is there's no way for your company to stop this. There's no way. So you have to have better policies in place.

59:29
Daniel Nestle
You have to have a reactive game plan in place and you have to simulate and game this out constantly. You got to do drills in house to really be able to handle this. The flip side of some of this, I think is that the deepfakes especially are, might give people who are actually guilty of some of those things the excuse to say it's a deep fake, you know, and you can't prove otherwise, you know, so it's going to go in both directions and in some ways the, you know, the deepfake terror, which is real and which should continue to be real is, is I think just going to get more nuanced, you know, as we go along.

01:00:14
Alec Crawford
Yeah, well, look, there's companies out there already that claim that they're going to go through the Internet and look for negative things about your company and they're going to try and they can. Actually there's a service right now for famous actors and actresses which will find deepfakes of them on the Internet and notify those companies automatically, hey, take down this deep fake of me. That's not me. Yeah, right.

01:00:46
Daniel Nestle
Yeah.

01:00:46
Alec Crawford
So, so the question is, at what point does a famous CEO need that?

01:00:52
Daniel Nestle
I mean, at what point do they all Then suddenly just rediscover blockchain and say what we need is validation on every piece of content that we put out there. And if it doesn't have my blockchain on it's not connected into me, into my thing, to my wallet. It's not. You know, when does that start to happen again? If it happens?

01:01:08
Alec Crawford
Yeah, that's such a great idea. I think it's all about money, Dan.

01:01:13
Daniel Nestle
Sure.

01:01:14
Alec Crawford
And when no one watches TikTok because it's all garbage and they're losing money, then they'll be like, okay, everybody needs a blockchain, you know, a private key to be able to upload the blockchain and prove you are who you say you are.

01:01:30
Daniel Nestle
All the validation man. Well, look, I mean, I think that there's so much more to uncover here, as always, and we really went deep in some areas that I think that I know communicators need to understand, that marketers need to understand. And I thank you for putting things in quite understandable terms as well. It's remarkable what's happening out there. And, you know, you're actually forcing me to maybe soften the edge of my harshness against Copilot a little bit because, you know, I like to kind of poke at Microsoft a little bit. Although who knows what's going to happen now with, now that Claude and as of, you know, the listeners will be getting this several months from now.

01:02:18
Daniel Nestle
But I think yesterday Anthropic and Accenture tied up with, on a big venture where, you know, Accenture is going to be basically rolling out cloud solutions in with clients. And you know, I think it's wonderful because I love Claude and because I think it's only good for companies to be working with the best products on the market. That said, you know, what happens now to, what does Microsoft do in response? What does Google. I mean, you know, Google's fine. I don't think anybody's, you know, I don't think anybody has to worry about Gemini. They're going to keep doing their arms race and they have enough of a claw in different environments that, you know, they just have to keep it, keep doing better. At some point, maybe we'll all be two shot, two model shops rather than one.

01:03:05
Daniel Nestle
I don't know, I don't know how it's gonna shake up.

01:03:07
Alec Crawford
Yeah, yeah, I think it's, I think everyone should be multiple model shops. We don't know who's going to have the best AI a year from now, we literally have probably eight model franchises. And for each one of Those, anywhere between three and 100 models that we look at, test, play around with, do an a B comparison. And I'm not claiming that every company needs to do that, but they need someone like us to come in and say, hey, by the way, for your CFO doing Excel spreadsheets, you need to use CloudSonic 4.5.

01:03:44
Daniel Nestle
Yeah.

01:03:45
Alec Crawford
Right, because that's the best right now. Right. Oh, by the way, you need something that's connecting to your bank's core data. Yeah. The only thing that really works right now is OpenAI. Whatever it is.

01:03:55
Daniel Nestle
You have to have multiple things. Yeah, you do.

01:03:57
Alec Crawford
Absolutely. And I think people are just starting to figure out that you really need a model agnostic platform. Meaning we can even use a bank's private models, right?

01:04:08
Daniel Nestle
Absolutely.

01:04:09
Alec Crawford
Not just these public models or open source models or OpenAI. That's where the puck is going. As you know, Wayne Gretzky famously said, we need to.

01:04:23
Daniel Nestle
And it's like the puck is going, but it's getting faster and faster. And I think there's. We're. We're moving into 4D hockey, and there's. There's 35 pucks on the ice. And good luck. Good luck with all that. Well, listen, Alec, I can't tell you how thankful I am for you coming on the show and for, you know, for us having this back and forth. Like I was on your. Your show, you're on, like, what a great way to. To really share knowledge, build relationships, et cetera. So thank you so much for coming on. Folks, out. Everybody out there, you can find Alec Crawford. And you know, again, Harken back to the beginning of the show. You're looking for the Alec Crawford who is at Artificial Intelligence Risk, not. Not the Alec Crawford who is all about sustainability. Al.

01:05:14
Daniel Nestle
Although the Al Crawford, this Alec Crawford is concerned with sustainability in a different way. However, art. Alec Crawford, CEO of Artificial Intelligence Risk. Look him up on LinkedIn AIC.com. That's his business. That's AIC risk.com. And then look for the AI risk reward podcast anywhere. You get podcasts, look for it on YouTube. And the AI risk reward substack. Are you still calling that the Stay blog at all?

01:05:39
Alec Crawford
Yeah, it's. It's part of the same. Yeah, it's part of that same thing. It's the same section, basically. So basically it's a. I'm writing under the AI Risk Reward banner now. Yeah, I really wish I could Write more often.

01:05:55
Daniel Nestle
Me too.

01:05:55
Alec Crawford
I'm probably writing like once a month now. Love to get that to weekly, but again, I'm creating original stuff. It's not what people call AI slob.

01:06:04
Daniel Nestle
Never.

01:06:05
Alec Crawford
If you read my stuff, I wrote it. It's going to be interesting.

01:06:08
Daniel Nestle
Yeah.

01:06:09
Alec Crawford
It's not coming from AI.

01:06:11
Daniel Nestle
Well, that's great. I'm using AI to help me write or to help me. Let's call it help me gather thoughts. But the thinking and the core writing is mine. And I feel the same way. But I don't mind saying I'm using AI Let people know. I don't care. I mean, as long as I let people know. Not pretending it's all about, like, is it good quality or not? Like, does it. Does it move people the way I want it to move people? If it. If it does, then it's fine. If it doesn't and it's AI slop, then that's on me, you know, but, man. Any last words, Alec, before we sign off?

01:06:46
Alec Crawford
Oh, yeah. No. So, another exciting piece of news.

01:06:49
Daniel Nestle
Yeah, please.

01:06:50
Alec Crawford
Which is macmillan has accepted my book proposal.

01:06:54
Daniel Nestle
Oh, terrific.

01:06:54
Alec Crawford
Next year I'm publishing, you know, working title, but, you know, Risk Management in the AI era. So that'll be coming out next year from.

01:07:04
Daniel Nestle
Congratulations. That is big news.

01:07:05
Alec Crawford
About half written. I'm interviewing a bunch of people as part of the book. It's super exciting if you're a famous AI person or maybe you're a famous marketing person who knows about AI. Yeah, definitely looking to interview people and incorporate their. Their views into the book. So super exciting.

01:07:23
Daniel Nestle
Okay, well, let's put a pin in that because I know some. Some relatively, well, relatively famous people in the AI world and it would be great to connect you. So everybody out there, look up, you know, go connect with Alec. Read, read his substack. You know, we. We communicators, often we hear the word risk and we're like, snooze fest. That means legal, you know, it means, oh, gosh, we got to talk to the accountants again. That is not what we're talking about here. You know, there are so many layers to this. You need to understand it if you want to communicate properly for your company. So please, again, AI Risk Reward, the AI Risk Reward podcast and the substack, and please do check out alec here on LinkedIn. Thanks, everybody. Thank you, Alec, for coming on and we'll see you again next time.

01:08:09
Alec Crawford
Thank you, D.

01:08:16
Daniel Nestle
Thanks for taking the time to listen in on today's conversation. If you enjoyed it, please be sure to subscribe through the podcast player of your choice. Share with your friends and colleagues and leave me a review. Five stars would be preferred. It's your call. Have ideas for future guests? Want to be on the show? Let me know@dantrendingcommunicator.com thanks again for listening to the trending Communicator.