S1. Ep6 - What's New In The AI World

Katie: Welcome to the AI Automations for Business Podcasts. We're your hosts, Katie and Null, and we'll be discussing how you can use AI for your business along with the latest news, updates and automations to help you stay ahead of the curve, allowing you to grow and scale your business more efficiently. Please be sure to subscribe and I hope you

Noel: enjoy this episode.

Katie: Hello, welcome back for another episode with me, Katie,

Noel: and... Hello, it's me again. It's not.

Katie: So we're doing a slightly different episode to the ones that we've recorded so far for you. And we want to give you some updates. And I'm also going to be asking Noel some questions about things that I have seen and heard online. So I hope you really enjoy this episode. And of course, if you've got any questions that you want us to answer, then just let us know you can either send us a message or you can come into our free LinkedIn group, which is AI Automations for business. It's open to anyone. And you can come and just drop your question in there. And we would absolutely love to hear from you and answer your question. Okay. Should we get into it now?

Noel: Let's do it.

Katie: Okay. So let's start with any updates or anything new that's come out recently. So as of of recording this. We are in March of 2025.

Noel: So yeah, there's been quite a few updates recently. The AI model arms race is continuing. It's going strong. So within the last few weeks, we've had Anthropic have released Claude Sonnet 3.7. That model is incredible. It's really, really good, really fast. It has incredible, you know, coding abilities and things like that. So it's it's a lot better than their previous 3.5 model. And, yeah, I would say like it's the speech or like mimicking human speech is also a little bit better than 3.5. And for me, 3.5 Sonic was probably the best out there. So, yeah, it's only getting better.

Katie: Can you give us a bit more detail on what makes it so good, like what this update is? Can you give us any more details?

Noel: So I think what makes it so good is it kind of, it understands the prompts a bit more. easily as well so when I've been testing and using this it's like you know with 3.5 sonnet you could ask it a question and it might get it right might get it wrong or especially in automations and things that it might you know hallucinate and do different things that you don't want it to do but 3.7 is like next level it really understands what you're after it really thinks about it also it's they've kind of added that there's the whole reasoning is like a big thing at the moment where models you know go through like a change of thought to think about a response before it gives it. So, yeah, that's also been added in as well. So, you know, it takes a little bit longer to get a response back, I think.

Katie: But the response is are. How much longer are we talking?

Noel: A couple of seconds.

Katie: Right. Okay. Yeah. Yeah.

Noel: Some of the other reasoning models, they might take a minute or so to think, whereas, yeah, with 3.7, it's a little bit quicker. But as it's thinking about it, you know, you can see all of its chain of thought what it's gone through. So if you don't get the response you like or was expecting, you can always go back and review what it did and then maybe adapt your prompt to make it work better.

Katie: Yeah, that sounds really good.

Noel: Yeah. No, it's awesome. Yeah. It's a similar price, 3.5 sonnet as well. So if you're using it for automations and things that you're then using and the API. So, yeah, the costs are very similar, and you're getting a performance boost as well, which is always nice.

Katie: Yeah, sounds good. Okay, so any other new updates?

Noel: So the other one is another model update. And who's this for? So Anthropic release 3.7, and then, you know, Sam Altman at OpenAI, went, hold on a minute. We can't have this, you know, go unchecked. So they decided to release DPT 4.5 research. That's really, I say again, it's another really clever model. They're only getting smarter. But what this one does is you can give it a question and say, well, do me some market research on this niche for this product. And it will go away and it will think for ages. It will troll the internet for 10,

Katie: 12 minutes to get all of the information.

Noel: So even like a simple question, it will really go into it. But before you, or before it goes off and does the research, it will take a look at the question or the prompt that you've given it and then be like, well, hold on a minute, that's probably not everything. So it'll give you like a list of questions and say, well, you know, if you're looking for research on this topic, you know, have you thought about this? Are you interested in these different facts or whatever? Or, you know, are these features of a product? Are they interesting to you? And then, you know, you can get the chance to like add it in a bit more context. So, yeah, you're not sat there waiting for 10 minutes for a response to get something that you don't want.

Katie: Yeah, that's not really the best thing, is it?

Noel: But I say with 4.5, that's probably their best model for replicating human speech. So the response is far less robotic,

Katie: which is, yeah,

Noel: it's really nice to use.

Katie: Yeah,

Noel: that was, that's that one, a little update. But, yeah, opening, I did have one more, though.

Katie: Okay.

Noel: Well, they haven't released it. They've talked about it. I'm not sure for me personally I would ever use. Is it, I mean, I'm pretty sure you wouldn't want me to subscribe to it.

Katie: But they're going to release,

Noel: they're released like a super agent. And they've been quoted as pricing it at $20,000 a month. I mean, they haven't really said.

Katie: So do you know what's actually included in this $20,000 per month?

Noel: They haven't really gone into all that much detail on it. They've just said they're going to release this agent.

Katie: Yeah.

Noel: It's going to, the rumoured prices. 20K a month but yeah yeah really knows

Katie: which is quite interesting seeing is meta recently announced that they're going to have a free AI agent across yeah facebook instagram and WhatsApp yeah i'm really like intrigued to see what that's going to look like and how it's going to perform yeah

Noel: that is a as exciting one but I think that one came out after the announcement that Open A.I. did. So they've probably seen this ridiculous price tag and then gone, you know what, let's just do hours for free and everyone will use the hours.

Katie: Yeah. Well, you know, Zuckerberg likes to, I think, take inspiration from other platforms and, you know, why not? I mean, you know, and then he kind of puts his own spin on it, doesn't he?

Noel: Yeah, yeah. He's making enough money somewhere else to let us have that for free. Yeah, yeah.

Katie: Thank you. Thank you, Mr. Zuckerberg. We appreciate that. So while we are on the subject of AI agents, there's been a lot of noise on the internet, in particular, YouTube, about people saying that AI agents are great, but actually they're not very consistent. They're not reliable because they only work a small percentage of the time. Yeah,

Noel: so I've been using N-A-N and their AI agent, and they provide you like a really cool visualization of when a request goes into the agent. It will show you the mapping of where it's going to and from, so you can see it live and see it all thinking, which is great. But then during my testing, I've got like a massive system prompt, really detailed. And even with that sort of level of detail, the agent would just be like, well, I could use the social media agent, you know, link within the tools, but I can't be bothered with that. I'll just make it up myself. It completely ignores the tools. And we're like, well, no, that's not the point. No.

Katie: Okay.

Noel: Yeah. And I think for like businesses and anything that's like client facing, and if you're using an agent that's doing like booking systems and things like,

Katie: you could find

Noel: that the agent would say, yeah, I've booked you in for two o'clock. but actually within your business calendar, that doesn't exist because it just went, yeah, that'll do.

Katie: Okay, you turn in.

Noel: It's a bit, but it's only about five-ish, 10% of the time. But the kind of, the thing you've really got to do is like really jump in and like, you know, sort out your system prompt. Make sure it's really well structured and the things that, otherwise, yeah, it'll go off and do its own thing.

Katie: So you almost need to tell it to not ignore what you've had. previously told it. Is that correct? Yeah.

Noel: Yeah, yeah, you do need to do it. You also need to describe all of the tools that it's got access to as well. In the NAN, you've really got to say, well, you know, there's a tool there to do this particular function and that function will produce whatever it does, you know, like social media posts or images and you've got to be really, really strict on that. And then you've got to name that tool in a way that the AI would then go, well, yeah, so I need to go down that route.

Katie: Because if you don't,

Noel: if you're unclear in any of those two areas, it will just start ignoring it and then doing things on itself. Even for me, it was like, I have like a hierarchy of agents. So I've got like a top level agent that dishes out tasks. And then there's sub-agents that then do the tasks. And that was kind of what I was testing out. I was getting either like a social media agent and it would go off and create posts and images. But then what I was finding was that agent was doing the job right. So that was good. but whatever it was sending back to the top level agent, the top level agent was like, I don't care about that. It just made it up and then just sent me something back that was nothing to do with what the other one did. And I was like, oh, come on. It's tricky. It's not perfect.

Katie: Yeah, okay. But then I guess we are still very early days with AI agents.

Noel: Definitely, yeah. I think for businesses, we need to be looking at more linear workflows. so you have one input that goes into an AI and then that then does one particular output.

Katie: Yeah, nothing too complicated yet. Yeah,

Noel: yeah, that's a safe bet. Okay.

Katie: So the other thing that I've seen popping up and people kind of like going to town on is that businesses who are using AI and almost like trying to make us feel bad for you, using AI or AI automations because we're losing the human touch. And, you know, a lot of us started our own businesses because we wanted that, you know, that service piece. We wanted to be able to help people. And by then using AI, we're taking some of that away. And I would, I guess, like, I would really love your thoughts on that, no, because obviously I, I understand. I understand why people are saying that, but I feel like also at the same time, people who are saying this probably don't know AI and AI automations particularly well or to the level of what you can do with AI and automations.

Noel: Yeah, I do get that with the personal touch and it's, yeah, it's kind of subjective as well. So like some customers would be like, I want to do this particular thing and then you give them the output and they're like wow that's not really you know the kind of thing we're looking for and it's you know it all really comes down to the prompt engineering and that that's that's where the secret key is so so like i've always said is like we shouldn't be automating absolutely everything within

Katie: a business

Noel: i mean we've got to do something at some point yeah you know we can't just sit back and let i just loose on absolutely everything but there are certain tasks that you know that with AI automations we can speed up you know so where where it'd be like you've got like a product that you're selling online and you want to be able to send them like a personalized email that that's something that's really easy to do and it can be done within seconds of the order being placed or they could wait six hours you know because they've ordered it in the middle of the night and it might take us like 10 50 minutes to write an email that's tailored and personalised, you know, to give them that extra experience. Whereas, you know, with the automation side, we can just do that instantly, almost.

Katie: Yeah. It's picking and choosing, isn't it? What's going to save you time? What's going to save you overwhelm? And instead, where can you put that energy in that time into, like, wowing your clients and providing an incredible service elsewhere?

Noel: Absolutely, yeah. So another thing I built the other week was kind of something that you could easily automate was like the trust pilot reviews for companies like most people ignore them for most companies ignored and they just never bother to respond.

Katie: Yeah, they just kind of like collect to them and then go we've got so many 100 or 1,000 five star reviews on trust pilot.

Noel: Yeah, but they never look at the bad ones and they, oh, they might even just like respond to the good ones because I don't really want to get into any confrontation with a one star review. But the system why I built out as like a quick demo was it would go off and it would look at their reviews and it would create like a series of reports. So that's the sort of thing that's like really good for AI automation is because, you know, you haven't got all the time in the world to go through and answer reviews, answer comments. But this AI would then just put it into a report, you know, a couple of pages that you could read through and then go, oh, right, well, you know, these are the things that I need to improve within my business to

Katie: ensure.

Noel: that everybody's getting the best, you know, customer service or, you know, the best experience from using the services and things that. So, you know, there's little things that that can, you know, make a massive difference. It doesn't take all that much effort to.

Katie: Yeah. And I think as well, like with AI, if you're responding to, for example, the Trust Pilot reviews, if you're, if you're, you know, obviously it's very, very important, I think, even to reply to negative reviews, because that's where you can improve and obviously you've then got that chance to make it better for the customer and have a completely different opinion and view of your business. But if you're like, I don't even know how to start this, this is where AI can really help you.

Noel: Yes, definitely. I wouldn't know where to start. Yeah.

Katie: But that's the thing. Like for those who haven't done any customer service roles, like obviously I have done that. You know, I've done it for my own businesses. I've done it in, you know, retail. So, you know, I understand, like, how to make it better for, you know, the person who is complaining. But, yeah, for someone who doesn't know where to start or, you know, it's, you know, they're taking it personally, which you should never do because it's not about you. It is about your business, even if they are criticising you. you can then go, okay, how can I make this better? Get AI to help you with that. And instead of having like, you know, a really angry response that, you know, other people are going to see and then go, do you know what? Because of how they've responded, I'm not going to touch their business. Yeah. And it doesn't matter what kind of business you have, you know, the reviews and how you respond to them are so, so important.

Noel: Yeah. And another thing I built into that automation, was also, so the demo I did was on energy companies in the UK. So I was posing as like I was the company Octopus Energy and I was looking like British gas, EDF, Ovo and all the other ones at. So not only was I looking internally at the reviews that, you know, I was getting in per se, but that was also then going off and then looking at other companies reviews, my competitors are then going, right, well, what have they done good? What are they doing bad? You know, how do we, you know, how come we make things even better for our customers and maybe steal some away from somebody else. And that could be doing in any sector, you know, where there's, you know, you've got lots of reviews and stuff going on. So, yeah, yeah, I think that report was quite powerful. But, yeah, I've only got a couple of people. I want to buy that off me already.

Katie: Yeah, I bet. I bet. If anyone wants help with their trust pilot reviews.

Noel: I won't write them myself, don't worry. Go and see no. Okay,

Katie: so is there any other updates or anything else that you feel that you want to answer on this episode, Noel?

Noel: So there's a couple of things. One of the biggest tool that I use, really, is make.com for the all the automations. And they've got some incredible tools coming out soon. And I've been very lucky to get early access to all of their new tools that are coming out. And they're incredible. So I guess I'll go through them one at a time. but the first one is really important. It's called human in the loop. So what that does is you can input that into your automation. And what it does is it would then create like a dummy web page. And then you can then put all of the information that you would like the human to review. And, you know, it then sends like an email to like, I don't know, either to yourself or to like a manager within a company. And then they can then go through and look at what the AI is done and, you know, review it properly. and then they could either amend it or approve it. And, you know, if it approves, it then triggers the next workflow, which then, you know, posts any of a social media post somewhere or whatever you've got in there. So, yeah, that human in the loop site is really, really clever. And I think it's going to be a massive tool within Make.

Katie: Okay, well, you'll have to keep us updated.

Noel: Definitely, yeah.

Katie: Yeah, and let us know what you've managed to build with it and things like that.

Noel: Yeah, but Make also have their own AI tools that I've got access. to us also. So when I went to Make Waves, their annual conference last year, they showed some of their AI tools there and it's like, and they can see like most people that are using make.com are using OpenAI's

Katie: module.

Noel: So they're like, well, obviously, you know, this is kind of a big trend. So they've built like their own like sentiment analysis AI kind of thing. So I use that within the reviews so I could work out what was the positive, neutral or negative. It would automatically do that. There were things that you can categorise text that goes in. So again, for that review one, I was then going, well, what team would that most suit would it be, you know, the sales team, the support team?

Katie: And it would figure

Noel: all that out and then just output which one it was. It wouldn't give you anything else. It would just figure out that one little piece. So yeah, they've got loads of really handy little tools that are coming out that just, you know, just drop into a scenario really nice and easy. It also means you don't need to do any prompt engineering. It's all you done for you. Amazing.

Katie: I guess the other thing to a lot of people's ears.

Noel: Absolutely, yeah. So, there's a lot of time typing.

Katie: Yeah.

Noel: And then the last one is something that you've called the grid.

Katie: The grid? The Instagram grid.

Noel: No, no, it's not as cool as that. It might be. It is to me. But basically, we've made.com. Like, I built a web app which has 80 scenarios, which are all kind of interlinked together. They all interlinked into the same sort of data. base and things like so it was quite complicated but the one thing i was lacking was like an overall view and what the grid does is it allows you to have like a 3d view of how everything within your scenarios are all into linking so if you've got any sort of issues you're like well why isn't this going from this scenario to the next one you could look in the grid and then it will go you know it shows you in real time where things are going to and from and you know whether the data's going in and out or just one way. Yeah, it's incredibly clever. But yeah, they showed that at the conference, and I had to go on it while I was there. But yeah, now I've got access myself. It's, yeah, it's really impressive.

Katie: Okay. Again, you'll have to keep us updated.

Noel: Yeah, definitely. A lot of people in this automation space are really excited by the grid just because it looks cool.

Katie: Yeah, okay.

Noel: They can't really work out really what the use case is, but they like it because it looks nice.

Katie: Yeah, well, definitely keep us updated now.

Noel: yeah 100% will do

Katie: thank you so much for listening to this week's podcast episode if you've liked it please subscribe and feel free to leave us a review and we will catch you next week for another episode