Prompt Madness

>> Anouk: Welcome to fusion talk with anouk and steve. Well, here we are, our first series.

>> Steve: Is it the first?

>> Anouk: Well, what other series did we do?

>> Steve: We did a little series about something,

>> Anouk: uh, when we did the breaking down of the ten tips. No, the, uh, um, I don't know, PowerApps governance thing.

>> Steve: Maybe. I don't know anymore.

>> Anouk: Let's assume this is our first real series.

>> Steve: Yes.

>> Anouk: All right, guys. Hey, welcome to Fusion, uh, Talk. It is a daytime for us. We're doing this in the day, in lunchtime, uh, which is unusual because normally we are disappearing into night time. And then we get distracted and go for dinner and start tasting wine and stuff like that. And then we run out of time and never get it done. So we thought we'd try something new.

>> Steve: Um, that. And because of the busyness of personals lives.

>> Anouk: Life is bad. Well, it's your fault. You have this huge charity event where you're trying to make 20 grand for children with bones.

>> Steve: Which is a good thing.

>> Anouk: It's a. Certainly a higher priority.

>> Steve: Yes. It's taking a lot of my time these weeks.

>> Anouk: Yeah, part of, um. Good, good, good. All right, so here we are, we are recording and we're going to start a series. So what is our series on?

>> Steve: Our series is on, um, a little. Some. Something we talked about, um, and it's about AI prompting.

>> Anouk: A little something. A little something not at all popular with anybody else in the world at all for the last two years.

>> Steve: But it's not just AI prompting. We are not going to tell people how they need to need to write a prompt or anything like that.

>> Anouk: No, I think we're way beyond that. I mean, uh, we've both done sessions one way or another with prompts. I remember attending a prompt hackathon with Microsoft fonts, which was quite fun for business people. Um, but I think there's still work to do in that space. I think people don't really understand. They kind of expect that six words will create them as 25 page document. And that obviously is not correct. No, but that's, uh, not really what we want to look at. What we want to look at is a, uh, series of short podcasts, um, to talk about prompt at the organisation enterprise kind of level. There is an issue with prompts at that level. There is an issue with things like consistency, uh, and there's a problem with things like sharing and collaboration and all that kind of stuff.

>> Steve: We have the same issue with prompts like we had with documents at the beginning of SharePoint or from a file

>> Anouk: server to sharepoint and the real frightening thing is that people still have problems with documents.

>> Steve: Yes.

>> Anouk: I had a, um, brief conversation with a member of my team. All right. And so people that are supposed to be, you know, it, competent and literate, and the conversation was about a little thing called metadata and whether we should create multiple libraries or whether we should actually have a managed taxonomy to, you know, allowing us to search through all the libraries. And the response was, oh, let's try this metadata thing.

>> Steve: Let's try it.

>> Anouk: Yes. And I'm thinking, okay, it's my fault, I haven't educated my team. I mean, it can, it can only be me. Uh, but anyway, yeah, it's, uh.

>> Steve: But we, we see the same issues coming with the metadata, the labels, uh, documents, like we had in the past, still have. Sometimes we see it now with AI and M prompting.

>> Anouk: Yeah, I think so. I think it's interesting. So, look, we, we've got a number of subjects that we want to cover. Five particularly, and hopefully will be no longer than 20 minutes or 25 minutes. So lots of short podcasts. Um, and, um, we have five. We have prompt chaos, which we're going to talk about now. And so we'll look at that a little bit. Then we're going to kind of move on to how you can create maybe a prompt library or a prompt agenda structure thing, um, prompt metadata. But we'll talk about that as a way of providing consistency. Human prompting. Yes, I'll let you add to that

>> Steve: the one. I wasn't sure of what we were going to do. You were going to let me add a.

>> Anouk: Yes, it was intentional and I do apologise, not really, but that's okay. Um, now, I mean, we know that humans have to have prompt. This is direct. It's not like they can get away with it. Prompt is going to come out of a human person's brain, so we'll take a look at that. Um, governance and creativity. So you don't want to control things too much, allowing some creativity. And then we're going to start looking at the future of prompting across the organisations. So that's what we're going to do. Um, and, uh, we hope you enjoy the series. So they'll come out quite quickly, we hope. Um, but yeah, and NOOK does have this very large charity event running over the next two weeks, so we'll have to see how it goes. But we're on to prompt chaos. So what's the biggest issue with this?

>> Steve: Um, I think the biggest issue with the prompt, what we call prompt chaos, is everybody uses AI copilot, ChatGPT, Claude or anything else and they do it on their own. They start writing a prompt and they get good results and the prompt ends there. It's not being shared with others. It's not being hand over or not saying to your colleague, hey, I've tried this prompt. Uh, what could I improve? Or how can I get it even better to get better results. So they stayed with the user that created it.

>> Anouk: It's kind of even worse than that, though, in the. Do you have a personal list of prompts?

>> Steve: I have a few prompts listed.

>> Anouk: When was the last time you added one to the list?

>> Steve: Last week.

>> Anouk: Oh, that's okay. That's brilliant. I don't. My brain doesn't work that way. So, uh. But I know, I don't know many people that have talked about the fact that, hey, I have my prompt library here or I have my prompt list

>> Steve: in notepad, but the reason I have it is just because of the trainings that I'm giving. Ah, they are great examples to provide in a training to show people what AI can do for them.

>> Anouk: So they're all.

>> Steve: Especially when they're going to copilot.

>> Anouk: Agreed. My apologies. So they're not for your personal prompting? No, no. So if you created a prompt for some script and you then try and do it again, you don't keep that prompt around? No, no.

>> Steve: Then I search in my history and try to find it again.

>> Anouk: Yeah, no, that kind of. I don't say it makes sense, but yeah, it's, uh, it's. One part of the problem is that there is no magic tool. I now have a password store. Um, maybe, um. Actually, I'm going to cancel this now. Let's spend the rest of the time designing a new application for a prompt safe. All right, See everybody later.

>> Steve: I don't think you need to go that far, but you need to have something where you can share your prompts in your organisation.

>> Anouk: No, but that's the second stage. I still don't think there's a way of me personally storing my prompts. Maybe it just needs to be in the tool itself, in AI that it says, hey, look, this is your history of prompts.

>> Steve: You need to have an add on in, uh, your brow, in your AI tool or your browser that you can say, save this.

>> Anouk: Yes, that's what you want. And I think it's important when you're using prompts for coding because quite honestly, 95% of that might be right. And all you want to do is to add an Extra button somewhere, in which case you want to be able to go back.

>> Steve: True. But then what I do when I'm doing that kind of coding, I don't go fully back. So. So I just say in to GitHub, uh, co pilot at the button here.

>> Anouk: Okay. So that's personal prompts. And then of course we have the other issue which is the real chaos here is shared knowledge.

>> Steve: Yep. You don't share it with your colleagues. They don't know of the existence. They don't know what you're trying to do with AI yet. So everybody lives a little bit in his own world based on the eye.

>> Anouk: Yeah. And it's a bit of a shame really. It doesn't sort of work. We have, we have that cut and paste culture which is what we tend to do with, with everything. Um, and uh, yeah. Microphone troubles. There we go. Um, so I paste and cut and paste culture. I did exactly that this morning. I was comparing some contract information to try and define something. Um, and I had two choices on how I could get that information to, uh, somebody else that had asked for it. Uh, one was just cut and paste the results. But of course you're actually doing research, so something to. I mean we're actually now sharing a team collaborative AI tool set and that's a little easier. Something that I might need to think about doing at work at that level. Um, but also basically all I could do was ask Chat GPT at the time was it chat might have been copilot, but as copilot to just read do me the prompt that I should have used to get the same information. So I cut and pasted that into teams along with uploading the two files that I was referring to and said, look, this is how I got the information. This is the subset of the information. Go ahead and try it yourself and get and see you've got any more.

>> Steve: And it's all got the same results.

>> Anouk: Then I haven't checked yet.

>> Steve: Okay. Because that's something I see when I give a prompt into my trainings to what they can do with it. Um, and they try the same thing. They change one word, of course, not my data, but their company data. Change word to every. What everything is. Everybody is using. They all get some kind of different results.

>> Anouk: You know, as long as the essence is the same though.

>> Steve: Yeah.

>> Anouk: Because don't forget, it'll. It'll pick up on your personal preferences to start off with. Um, language may well be an issue, but that's fine. That's the same as any other knowledge sharing we do, to be fair. Um, but yeah, so we're in a cut and paste paste culture where we end up cut and pasting our prompts and not having anywhere to store them. Um, so again, that causes, ah, a little bit of chaos around.

>> Steve: Yeah. And also the lose of data. Not really your data, but you're losing some kind of good prompts you don't have. You write something, you say, all right, this is very good, I need to keep it. You don't directly and then, damn, I did it sometime. When did I do it? Where do I need to go and search for it?

>> Anouk: Yes. Bit like that elusive email that, you know, you received 10 days ago and you can't find it again. Mhm. So. So yes. So that is a risk from an organisation perspective. And let's be fair, we're not just talking about two people working together a few desks apart here. You know, um, if you've got a prompt that is providing me, how much annual leave do I get or have left, uh, based upon my account, blah, blah, blah. Um, that prompt would be useful for everybody. You know, in fact, it could even be an agent. But that's not what we're talking about now. Um, it, um, it's just a prompt that only applies to me and my information.

>> Steve: Yeah.

>> Anouk: Um, and that prompt then gives clear indications of what is needed. If you have a project team, that project team could be international. And so you were saying, hey, look, I want you to go through my emails and I would like you to find me relevant information and tasks for this project name managed by this person, this user. That information is consistent there, even the triggers and the, the search points. Um, and then that could be shared out by a large group of people. So all of those good information, it's not there. I wonder why people don't do that though. Is it a power struggle again, are we back to the I have the knowledge and therefore I'm looking good and so I'm not going to share it with everybody else. I mean, is it, uh, a culture change first?

>> Steve: I think it's a culture change. I think it's more like we do it, we are not sure what we are doing and we don't know if we need to share it so that people don't trust their own instincts.

>> Anouk: I have to say today when I cut and pasted the prompt, that's one of the reasons why I sent the prompt through Copilot to say, okay, can you cheque the spellings and make sure this is all correct? Uh, and what prompt? What should the prompt look like to get this information? Because I was passing it on somebody else and if I put a, ah, a search string for our intranet, then I'm going to get the same results. But with AI, of course you don't. Uh, because there's so many variables around, um, so you sometimes wonder why the hell we're doing it. But, but yes, there is a, there is that kind of issue and that's down to adoption and understanding again, that people need to understand. AI is not going to be the same. Um, but it doesn't matter. But those great prompts that we have, uh, are something that should be shared, should be useful, should be able to reuse them. Um, but why don't you not just do Agents?

>> Steve: That's a very good question.

>> Anouk: So maybe that's another culture thing to remove chaos.

>> Steve: Because with your copilot, you can use it in Word, you can use it in Outlook, and Agents is not always on a place where you would like to have it.

>> Anouk: No, it's very true. No. So maybe there's a different kind of culture for both.

>> Steve: Yep, I think so.

>> Anouk: M. Oh no, that is interesting. So the, the risk of losing good prompts or just simply redoing the same information. Um, and uh, yeah, so, so then adding to prompt chaos.

>> Steve: It does. So I forgot what I was going to say.

>> Anouk: He was probably going to say something about reworking prompts. Yes, she's hungry.

>> Steve: Uh, no, I was not going to say about reworking prompts. I was going to say, um, prompt chaos. Uh, if you just drop it on one big stack and you don't uh, describe them well, where for what they need to be used, chaos will stay.

>> Anouk: That's true.

>> Steve: Because your sales team can have a good prompt, but your IT team doesn't maybe have anything to do with that prompt. And if they try to use it, or see, oh, this seems interesting to try, there will be chaos.

>> Anouk: Yeah, that's true. So whatever repository is used for storing the prompts, it needs to have a lot of um, kind of tags and descriptions as to what it's being used for. That's true. But is a prompt not a description in its own right? I'm not dissing you. I think you're right. I think that there's certain stuff we'll get into and I think in episode two when we start looking at libraries for prompts, maybe we can talk about that. But a prompt is already a, uh, word written description of what it's supposed to do.

>> Steve: Yes, but do you think users always read everything what is in there?

>> Anouk: And just cut and paste.

>> Steve: Yes.

>> Anouk: Oh, there's a prompt. I'll try that one. I mean, there's no cost to trying it. Well, they don't think there's a cost. Depending on which engine they're using, it may well actually cost them. But that's all right. Uh, in this space for the end user kind of prompt, which is what we're talking about. Um, that's a good question. So losing good prompts. All right. And making sure that you've got the supporting information so people know what the expected result will be of using that prompt. M. That's good call, good call. Um, the next area in terms of that chaos is that everybody is just using their own prompts. And so, um, there's 25 people redoing the same prompt.

>> Steve: Yeah.

>> Anouk: Which of course is just a waste of resources. Or is it, is it not just a learning opportunity at this early stage of the game?

>> Steve: It's not a waste about the resources because they will definitely learn from it, but they would even learn more if they know that somebody else is doing the same prompt because now they all creating the same problem themselves. Slightly different worded, slightly different changes in it.

>> Anouk: Yeah, uh, I get it. I was also just thinking whether, whether you share the good ones or whether you don't.

>> Steve: I think you at this stage of AI, AI still rapidly developing, changing a lot. You should share every prom to make sure you become better in it. But other people as well, that you can learn from each other.

>> Anouk: True. And I think it also thinking about that takes a lot of the fear away from people. So people that have never done it before, if they have access to prompts that other people have used, potentially that

>> Steve: would help, especially if there are direct colleagues that are using it, because then, um, if they are not sure, they can easily go and talk to that colleague instead of going to it.

>> Anouk: True. That is also very, very true. M. So this allows people not to reinvent the wheel if there's some way of being able to group these prompts together, which is cool. Um, what about improvements and constant improvements? Does a prompt improve? Do you get version one, version two, version three, version four?

>> Steve: That's something I'm not sure of.

>> Anouk: Well, there's also many variables here, isn't there? So let's, let's assume that you run a prompt on your email and then three days later you got 400 more emails in there and you run the same prompt. You're going to get different backs. It's not a search tool that says, hey, that's the email I want. And so every time I use this prompt, I'll get that email back because I won't. But we should improve our problems. So what? Whatever we have. All right, let me ask a different question. In your experience, do you get the information out of AI first time around?

>> Steve: No. It's always a conversation.

>> Anouk: So you get to a point eventually where you've got everything together.

>> Steve: Yes.

>> Anouk: So then you're going to need to ask AI, uh, hey, when I want to do this next time, tell me what the prompt should be to get to this point and then that prompt gets saved or shared.

>> Steve: Yes.

>> Anouk: Mhm. So then we need to have some kind of rules about the prompts that we share with people.

>> Steve: That's something we will discuss in the governance part, I'm guessing.

>> Anouk: Yeah, yeah, that's four along. We might be referring to that one as we go along. The final prompt.

>> Steve: Yeah, I don't think you have with prompt something like with documents that they do underscore final. Underscore final dev. Underscore final dev. Do not remove. Something like that.

>> Anouk: Do not remove this prompt.

>> Steve: That's something you probably will not have in prompts. Uh, because every prompt will have his value and giving you some kind of information back. Maybe not directly what you want, but it will become more and more obvious and you will become better in understanding it as well.

>> Anouk: Yeah, I think you're not wrong. And I think as we move towards the next episode, which is around prompt libraries, um, the noise in the background is we're recording, not in the studio, but we're staying in the daylight here and of course less people unloading on my street outside. So there's some frustrated drivers. Frustrated drivers. Um, so yes, uh, as we get on to the next episode where we start talking about how we store these prompts and how we get to reuse them, I think understanding a successful prompt is interesting and how you describe it and tag it, uh, and what the expected result will be.

>> Steve: Yes. And I also think, um, especially if we talk about copilot part of it in what application you used your prompt.

>> Anouk: Yes. The conditions.

>> Steve: Yes. It's all different in everywhere with different results.

>> Anouk: That's true. And if you're moving from company to company, they may have different rules and regs about what you can and can't do on their prompt and how it's filtered. And of course then you get your personal learnings as well. So it's all part of the fun. All right, so there we go. Uh, a little bit of prompt, uh, chaos Unchaost.

>> Steve: Yes.

>> Anouk: Straightened out.

>> Steve: Something like that, yeah.

>> Anouk: Cool. All right, well we hope that was a little bit of uh, AI prompt prompting information. Something that uh, is prompted a thought or two. So um, how many times are you

>> Steve: going see prompt again?

>> Anouk: Oh, lots, I think. So, uh, well we've got prompt Chaos, prompt library, the human prompt governance prompted the future of prompts in the organisation. I think this is all about prompt.

>> Steve: It is, yeah.

>> Anouk: Somebody should rate a. Some music, you know, prompt and circumstance.

>> Steve: But do you know why we are still talking about prompt?

>> Anouk: I have one answer in my head. But why?

>> Steve: Go ahead, give your answer.

>> Anouk: Because it's actually just an I O. It's a way of getting information out.

>> Steve: It's a way of getting information out and still people are not always sure what a prompt is and how to write it.

>> Anouk: Ah, uh, very true.

>> Steve: Something I learned from my trainings I'm giving.

>> Anouk: Yes. Microsoft did that when they did the prompt hackathon. Promptathon.

>> Steve: I had a training yesterday and we were. I was explaining search in teams. They all have co pilot licences and I said to them if the team search maybe not giving you what you want, you can always try to ask copilot. Yeah, but how do we ask Copilot was a question. Yeah, you write a prompt. We do what?

>> Anouk: Wow.

>> Steve: People don't understand prompting or don't know what it is or how they need to start on it.

>> Anouk: That's where we started, isn't it? I mean it, this is it so new that not many people know. Don't trust. It would be interesting set of numbers to look at. But for those organisations that really are moving in the frontier into AI, then yes, prompt chaos is part of that process and we're going to talk about how we can fix that tomorrow with or next. Not tomorrow, but next podcast and we maybe look at how we can actually list these or store them or distribute them.

>> Steve: Perfect.

>> Anouk: Bye for now.

>> Steve: Goodbye.