AI Plus WordPress Plus WooCommerce: Trends, Tools & Transformations

Mark: Sure. Thanks for having me Bud. Mark West Guard from WSS form. I’m the founder of WS Form and WS Form is a WordPress form plugin and I’ve been in the WordPress space. I’m quite a newbie compared to some people. I’ve been in this space about, I don’t know, 10 years and actually doing business in the WordPress space for about five years.

Yeah, happy to be on the show. Thanks for having me.

Bud: And now Alain.

Alain: Yes. Thanks for having us Bud it’s really an honor to talk on a podcast here. My name is Alain Schlesser. I’m the director of technology and innovation with XWP. I’m also the maintainer of WPCLI. I have been active in the WordPress space since 2014, I think which is neither a newbie nor one of the old God, I have been doing IT work for a lot longer than that, but I decided at some point that I was in the WordPress space for the community and I have been enjoying and regretting the choice ever since for several reasons. It has always been a technical challenge to deal with WordPress, but it has always been just just a pleasure to deal with the WordPress community and the people that make it up.

Bud: We have a good fortune in WordCamp New York 2018 to have you come, and I don’t know if you remember what you spoke about, but it was using Gutenberg and how to transform your existing code so that it will work with Gutenberg. That shows how long ago that was, right? But I’m glad you came. Tell us, how did you get to XWP.

I know at WordCamp Europe, you told my wife that you have been tinkering with code since you were what, three years old in the crib or something like that for a long time. Yeah. So talk a little bit about that journey. I know it’s a long one, how did you get to where you are today?

Alain: So it was not at three years, but rather at seven years that I started doubling with. Computers for the first time in the beginning, that was a purely theoretical exercise because I was not actually allowed to touch the computer. It was my dad’s computer that he brought, bought. And I was very curious, but he assumed that was not a kid’s toy.

So I was not allowed to touch it. I was just allowed to spectate as he was trying to figure this thing out. And what happened was that he was attending evening courses and I was always going through the material that he brought home from these evening courses, trying to understand it. And at the time, so the computer he bought was a Commodore C64.

And at the time, when you started into a computer it just started, it booted up into a programming language. There was not much else to do than learning to program to get started. So if you wanted to use an application or something, you first needed to learn how to program. That programming language to load the application and execute it within the program.

So that’s how I how I started in a very theoretical way, basically doing a pen and paper version of learning to develop a BASIC program. So BASIC was a programming language at the time. And after… A lot of more heated interactions with my dad when I told him how to do this and how to do that.

At some point, he just gave up and let me also have a go at it. And in the beginning, that was all under supervision. And at nine years old, I got my first own computer to to work on, to play with. So that was how I got started with this. And I thought A lot about also doing that as a career.

Opportunity while I was growing up, but I grew up in Luxembourg and in Luxembourg, there was never really an industry for that. I was a bit in the wrong country to do that at such an early age. There was just nothing happening in that regard. So I ended up doing something completely different in terms of education and work, but I Always did development as a hobby.

At that time it was mostly game development. I was always fascinated by the ability to tell interactive stories and to make pixels move on the computer and that, that was actually a good way to learn about not only programming on its own, but really learn about the entire ecosystem around hardware drivers, operating systems, because at the time for game development, there was no hardware accelerated graphics processes and things like that.

It was just, you had a machine that was meant to do text processing and to try to abuse it in such a way that it also was able to do games. So if you wanted to play music, you needed to hack into the driver of the built in beeper to change the voltage so that you could modulate the frequency and things like that.

So it was really a more adventurous way of programming, but it also allowed me to really learn about way more than just the structures of programming really understand how a computer works in terms of the hardware and the bits and pieces that make it up. And so I went from there always as a hobbyist working on all sorts of different things building a known Linux distribution and all that sort of stuff.

And in terms of working for the government in the administration of a high security prison. And in the beginning, that, that work didn’t involve any IT work at all, but they quickly figured out that I had way more experience with most of these things than the rest of the IT department. I got pushed more and more into that role of also taking care of that.

And that’s where I did a lot of more government and enterprise type software work. Dealing with Oracle databases and SharePoint intranets and things like that. So I have all sorts of certifications now from that government work. I’m a certified Oracle database developer and certified on all sorts of Microsoft technologies.

I don’t use any of that anymore, but at the time for that type of work, certifications were actually more important than the results you end up producing most of the time, unfortunately. So that’s how I got pushed more and more into that type of work. But at some point, I decided that working for the government, not the type of work that I was doing, but the…

environment in which I was working was not for me. I I started with government work because I didn’t know what else to do. And it seemed like that’s the type of work where you don’t burn out, where you can focus your energy on your Hobbies and pastime instead, but I didn’t know myself at that time.

And I ended up finding out a hard way that I was actually the one who needed to get his work done to set this satisfaction anyway, and in the government space, that was really hard to do because it depended on a lot of people that had no intention of doing their work properly and there were no incentives to do there was no negative impact for them of not doing their work. You ended up doing the work for many. other people as well, if you wanted to have success in your own projects. So that was yeah around 2010 or so that it, that I decided that I want, didn’t want to continue with that for a longer time.

And the initial plan was to slowly phase the work out. But for personal reasons together with my wife we decided at some point that we wanted to accelerate that And so we decided that we would just both quit our jobs. She also worked for the government she worked for a social office at the time leading a social office.

And so we both decided that he wanted to quit our jobs and do something else with our lives that was more fulfilling. And we ended up selling everything in Luxembourg to get rid of our debt, moving to Germany to reduce our cost of living because we went from to rather high salaries to no income at all, which requires a bit of a transitional phase, of course.

And so for me it was clear that I wanted to do something professionally with software development. And I decided that web development was the one big portion that I hadn’t yet done. So I had game, had done game development. I had done application development and I hadn’t yet tried web development.

So I decided to do that instead. And at the time, I just looked at market shares and WordPress seemed to be. Pretty popular and seemed to be a safe bet. So that’s what I started with. And that was yeah, around 2014 that I just jumped into the WordPress space and tried to make a freelancer career out of that.

So being a freelancer and working remotely was a very. Intentional choice because when we started over, the first thing we did was to establish principles for ourselves to not fall into this same trap again and ending up doing something we don’t like just for the money. So we put up a few rules for ourselves, and one of them was that we didn’t want to be geographically constrained because we didn’t know whether we wanted to stay in Germany or not.

We want to remain flexible. So that was why since then I only have ever worked remotely. And being a freelancer was mostly to be able to fire my client if I don’t appreciate the work anymore so that we can reverse that type of role. And that I also don’t feel. Unnecessary obligations that I need to stick to.

So that I can really also psychologically have control on my side to decide how I want to spend my working hours.

Bud: Yeah. And now I know you’re with X WP and how long have you been working there? And what more precisely do you do? And before you answer that question, though, you’re in old time Germany ? Did I say that correctly?

Alain: Yes. Yes. Very correctly.

Bud: And that’s where you settled when you left Luxembourg. Am I right about that? Yes. Yeah. So you’ve been there for a number of years now, right?

Alain: So we moved to Germany in an area where there’s actually almost no local jobs. So it is

very cheap to live here. There’s a rather high quality of living, but there’s not many local jobs around.

Bud: It’s very picturesque. I saw some beautiful pictures of it today. Yeah. Very nice. Yeah. A lot of farmland.

Alain: Yes, we basically live in the middle of nowhere in Germany. There’s not much around. We live directly next to a forest and it’s nice because we travel a lot. And usually we travel to conferences where we end up in large cities, which is very chaotic, very noisy. And although we both appreciate that from time to time, it’s also, there’s a nice contrast where when we get back home, it’s the absolute opposite. It’s just completely quiet. And so we have. Right now we’re, we have about six months of the year where we travel, where we have adventure and meeting people, making new friendships, and half of the year where we’re just at home, recharging batteries and enjoying quiet time and focusing on the internal side of our lives.

Bud: And what about your current work today?

Alain: Yeah, so I have been with XWP for four years now. Initially, I was in, in contact with XWP for a while, but it was never interested in doing the traditional agency work. So even though we had been in discussions, there was never really a project that makes sense for a while until they approached me where.

Google requested me for a specific project that was technologically very challenging. And that, that piqued my interest that was to work on the the M4 WordPress plugin, where the point is to take whatever WordPress produces and fully automatically transform it into something that can provide guarantees in terms of what the worst performance would be.

So I joined XWP because Google, I also had been in, in discussions with Google for a while, but at the time that was pre COVID, they were never open to remote work or freelancers joining a subcontractor. So that was not an option, even though I had a lot of friends already, where we always discussing the latest projects, it was never an option to work with Google, but here through XWP, they basically.

They paid XWP to hire me so that they, for them, it was a service that they were paying for. So that, that got around this remote work and freelance work limitation at the time. Now it’s changed since COVID.

Bud: And you’re being much too modest. You are a Google developer expert. How did that come about? That’s an official designation. How, you got that obviously through Google or through working with them. I’m just curious about that.

Alain: So the Google developer expert is a program where they have people that already have an audience and that share knowledge about their products to a developer audience so that they can use that as leverage to better educate developers to do to have more use of Google technologies.

And I’m a Google developer expert for web technologies which is basically the one category where it is not directly tied to a Google product like Google cloud technologies or Google maps or so, but it’s actually, Tied to Google Chrome as their product, but which means it’s web development in general, and because I had done a lot of education in the web development space and also with web performance, for example, which is a big thing with Google.

That, that was a good fit. And basically what you get as a developer, if you joined a program is, so you sign an NDA, then you get early access to information. You get access to things that are not yet public to run tests on. You join meetings where you can ask questions and things like that.

So it’s very, if you’re really into. Into the Google ecosystem of products, or if you’re working on the web, it’s very, it’s a very valuable program because you get better information earlier and faster, and you have the right context to ask questions if something’s not clear, because oftentimes, as you may know, documentation is not always highest priority when something is created. Having the right access to ask questions is always super valuable.

Bud: I’m going to take you back to WordCamp Europe in Athens a couple of months ago. And I know. At during the after party, you and I were on the street talking and I thought, God, this would be a great idea if we could just continue this conversation on Do the Woo.

So that’s one of the reasons why we’re here. And I’m going to take you back a little bit to that. And we talk a lot about AI. And so I’m going to ask a silly question to get started. I have a whole list of questions, but. I’m just going to do a silly one. Maybe it’s not so silly. We’ll find out. Will AI, and Mark, you can answer this too, obviously, because you know an awful lot about AI with your plugin. Know that. Will AI maybe usher in an era where we don’t have to have usernames and passwords anymore? Or am I nuts?

Alain: So that’s an interesting question. I don’t think that has much to do with AI. More so with how identity is defined and treated and there’s already right now there’s technologies that you can use.

That don’t require you to use, to have a username or a password to authenticate yourself. So there are already technologies for that. The thing is that right now, these technologies, they need to be properly integrated everywhere to make use of that. And usernames and passwords are just the standard that everyone falls back to if they.

They don’t know about any other approaches yet. It is something that will take time for the ecosystems and old third party providers to adapt to. But just for example, there’s past keys that is one approach to doing that. There’s more work being done on using federated logins where you have an identity provider of.

Unknown form where you identify yourself, and then the knowledge about that is just being used everywhere else. Where that the fact that you are authenticated is shared instead of letting you authenticate everywhere. So there’s different approaches to that. I don’t think that has necessarily to do with AI, especially In the type of AI that we’re talking about now, so there is some AI involved, but that is more of the machine learning variation where everything you do is still somewhat deterministic where machine learning is, for example, used for fraud detection to detect patterns that are usually, that, that look sketchy when logins happen to files and things like that.

So there is, in fact, AI involved. But the I, the process of identity is one that it should never involve guess of any guessworking of any kind, because it is super dangerous not only in the sense that it can be abused, somebody could identify as yourself, but also because that is one of the, one of the areas where you don’t want to have any bias involved.

For example, if working, if authenticating yourself is already working against you because of some inherent bias of an algorithm. So that is really bad place to be. Being able to identify yourself is a very critical piece of the architectural landscape, and we need to make sure that there’s always a way of doing that where you don’t rely on a machine to take any decisions on your.

Yeah,

Mark: I agree with you, Alain. I think that AI is, as you were saying, being used to analyze user behavior. We’re using that to make sure that you’re coming from the same place or you’re you’re using the same computer and things like that. But I personally don’t really want AI determining whether or not I’m the right person or not.

There’s too much fuzzy involved. And these techniques for analyzing user behavior to determine whether or not you are who you are can be used against you as well. Hackers will use that to, Develop whatever we develop, hackers will go in the other way and turn it, turn the technology the other way around and try and use it against you.

I do, Elaine was talking about the whole passkey system trying to go passwordless. I know Microsoft is pushing that quite significantly these days. That’s, that stuff is all great. We have all these two factor authentication. We have, we, we can put a USB stick in our computer and touch a thumbnail pad to get in.

But at the end of the day, you can still click on, use other method. You use other method, and you end up back to your password. We’re still a of passwords and it’s the vast majority of how. Hacking of people’s data still occurs, a Verizon study that said 81 percent of hacking comes as a result of a poor or stolen password.

We, we still got to make sure those passwords are long and difficult and, a pain to work with, unfortunately. But yeah, the things like the passkey is quite interesting where you. You unlock your phone and that effectively unlocks everything else that, that you own. I think that’s convenient for me.

I don’t particularly like the idea of that. I hope that I like having each of my individual accounts secured in, in different ways and with different passwords. Make sure you don’t use the same password on everything. I know we’re guilty of that with some things. I know my Netflix password’s the same as my HBO password, but yeah, I don’t.

I can’t see it taking over user authentication or or changing passwords, but we’ll see where the technology goes, huh?

Bud: Let me ask a different question altogether. I said, that was probably going to be a silly question. And I got the right, I got it. Hopefully this will be a little more poignant, which is, in the general press, there’s a lot of talk about.

I think it’s called apocalyptic AI. The worst case scenarios, what kinds of questions should we even be asking about AI today, where it’s going, what we should expect of it, guard rails, security, all that kind of stuff. I’m just curious what your thoughts are on that. In general, what are the things that we should be looking at right now?

So right

Alain: now, I don’t

think we should worry as much on what AI could do, but rather more on what humans can do with AI with it, because it’s really humans now get a type of access to information. That we didn’t have before and also so on one hand, the I can help us really figure out things that would have been very difficult to figure out before then, even if those are malicious things.

On the other hand AIs. Can also just like social media can have this accelerating effect with echo chambers and so on. So there’s a similar danger with AI where the humans will use the AI to figure things out everything they figure out will go back into the AI as training data for the next generation.

And so it has this feedback loop where if people use the AI, AI is now to do very malicious. things, the AI will be trained to learn that’s what people usually want to do. So it will be the first suggestion for next time around. So there’s this feedback loop that is pretty dangerous with the training data, especially as trained AIs are now being used to train other AIs that even accelerates this feedback loop.

But on the other hand, it’s also there’s there’s a fine line between making information hard to access and censorship. We always had to, we’re always trying to figure out what the best way is to do that. So censorship generally is not a great thing, but having impressionable minds have access to the bad type of information is also a good thing. So with AIs. It is even harder to even have a say in that and how that works.

Because for example there’s a lot of work now trying to censor these trained models, which is harder than expected because it’s not as simple as removing. The training data that was talking about certain things, because as the models become smarter, they can actually deduce the stuff that you remove from the training data.

Because just as humans came up with that in the first place, the AI models can come up with that as well. So on one hand, it’s really hard to, sensor the input that you’re providing to an AI like that. And on the other hand, if you try to put rail guards in place, usually these rail guards are what’s called a sandbox for the eye, for example, or a jail.

But it’s pretty easy to jailbreak those. Because the ai, on one hand, it has seemingly absolute knowledge, but on the other hand, it behaves like a five year old and you just tul it in your rules and it’ll happily do the AI itself is also a very impressionable mind, and you can very easily abuse it to just.

Do whatever you want it to do. So I think right now the AI itself, it is not capable of doing that much harm directly, but indirectly through what humans make of it as a tool it can already have pretty pretty severe consequences. So that’s one thing we need to be immediately aware of. It is a tool.

But a very powerful one. And so right now we need to make sure that the tool is not abused. Then in a second phase, at some point, we might enter a stage of these AIs where the AIs themselves have enough access to the real world, the physical world or the digital world that they could have their own direct effects on our lives and our society.

And at that point Then the AI itself is really dangerous because as much as we seem to consider something like chat GPT or all this bots to be human, like they are not at all human, like in terms of how they operate it is, it’s not even. Close. It is it has nothing to do in terms of how the decision making happens.

There’s no moral process to it. There’s no reasoning process to it. It’s just a statistical evaluation based on a training data. And if you give an AI the wrong access, it can just very. I would say ruthlessly, but that is already the wrong word, but because that assumes that it would know what it is doing, it will just through cold calculation to things that might be great for humans or do things that might be the demise of humans, and it doesn’t care either way.

Because it’s just a statistical evaluation. So that assumes, of course, that the I would have access to anything that can have that drastic of an effect. Hopefully we’re smart enough to not do that while we still have no clue how these things work, while there’s still these black boxes. But that is what people usually refer to as this apocalyptic outcome.

That is not because the I turns. Evil or something like that. It is just if you give the goal, it will pursue that goal without any other validating process in mind. There’s this nice there’s a, like an online game that is like a thought experiment where you have an AI system. that you train to produce office clips.

And it has the mission to become the best possible production system for office clips. And in the end, and it has consumed the entire universe. It has gone rid of humans because they were in a way of producing paper clips. Effectively, it turned the entire planet into resources. Et cetera, et cetera, because the goal was to produce more and more paperclips without any limit to it.

That’s if you’re working on your database, if you’re doing a query without adding a where clause the results turn out to be rather catastrophic most of the time. And this is a similar thing. You don’t have a where clause for your AI system, so it will just do it at infinitum. Without any reason to stop at some point.

Bud: Okay. I need you guys to help me out with something that’s very practical that I’m working on right now. So I’m working on a blog post for GoDaddy. And the nature of it is how to use AI to create a custom post type. Okay. And you have to understand this is coming from this perspective. I’m not a programmer.

I’m not a developer, but I know just enough vocabulary to, to be a little conversant with it. And that’s one of the things I found out is they don’t call it chat GPT for nothing. And I went through an iterative process to get. To an end that I wanted my question to you guys is how did it do that?

All right, it spit out or generated custom post type code. Does it look at other WordPress plugins to understand how to do this or what’s going on here

Mark: With open AI it essentially starts its life as a spider, right? So it’s going through the internet. It’s looking at webpages looking at content putting that all into a big database.

I don’t confess to know how open AI works inside and out But it’s essentially a database of information It is then is it’s a trained tool There are thousands and thousands of people that have taken prompts, it has given back a response, and there are humans that have said, yeah, that response makes sense, or no, that response doesn’t make sense.

And just like Alain said, it’s like a five year old. It’s being taught slowly. And the results that we get back sometimes can be a little bit. Wrong. We can be given snippets of code that are not correct. My, my fear with that is that if you haven’t learned how to program correctly, you can take that code on face value and end up in real problems.

And I’ll give you a simple example. If you were to ask it to maybe create a post based upon a user input, I’m sure it can give you some code for that, but it’s not going to take security into account. It’s not going to make sure that the input values have been sanitized, it’s not going to prevent your database from having a SQL injection hack, because it doesn’t really care about that.

A lot of the examples that you see on the internet are fairly rudimentary. And they get the job done, but in a commercial environment or in a, if you’re a plugin developer, you want to make sure that everything is sanitized your inputs and your outputs have all been programmed correctly, and you’re not necessarily going to get that with an open AI output.

Interesting story. So my wife’s family, they have a laundry machinery manufacturing company. They came to me with some software that simulated one of their washing machines that was written in Fortran for the old IBM machines. And I could understand the Fortran, but it was quite a lot of code.

And I thought I’ll put a chunk of this into chat GPT and ask it to convert it to JavaScript. And lo and behold, it did. It even took into account, indexing on a race starting from one instead of zero and all this kind of stuff. But if I gave you anything more than a few lines, it started to break down.

So there’s definitely limitations with it. I would certainly say in terms of. Using OpenAI to develop a blog post about creating a post. Don’t take what it gives you on face value. We use. open AI with our AI generated forms product. So you can basically type in a prompt and say, build me a form that does this and it will build your form.

And the way that we do that is we ask it to give us back a packet of data that we can interpret into a form and 95 percent of the time it works fine. Some of the times it is a naughty five year old and doesn’t give us what we want and things break so So I think we have a responsibility, you know going back to your previous question about you know Is it gonna take over the world and everything?

I think there’s a responsibility of the people developing these products to make sure that they don’t do anything Negative, but also us as users, we’ve got to make sure that we use this as a tool and don’t take it on face value and qualify everything that we get back.

Bud: And that’s one of the things I love about it is that I look at it right now as an assistant, as a teacher, I could ask you questions.

What do you mean by so and it’s great. I, it’s all about using and working with AI. And as a smart person has

said it’s not about. Whether AI is going to take my job away. It’s whether or not the person who knows AI better than me will take my job.

Mark: You can get wrong answers from Google just as much as you can from OpenAI, when you go onto a site and say, Hey, give me some code about creating a post, someone may have written a bad post about it. But yeah, it’s a completely different way of absorbing information and using information.

Bud: And Mark, I know I’ve mentioned this little blog post to you and you said, is it going to code to WordPress standards? Could you ask it to do that? Could you ask it to give it sanitized code? Could you? I haven’t tried any of that.

Mark: I haven’t tried that myself. No,

Bud: I’m going to try it.

Mark: Yeah. And I would be surprised if it could do that efficiently. I wouldn’t trust it.

Alain: I would assume it probably could. The thing on one hand, what is important to know is that the way this is trained is that it can only be trained on data that is publicly available, which is great to work on WordPress, for example, because of the open source concept.

So it has access to everything on GitHub, to everything on other portals. So it has read a lot of WordPress code. But also code quality in the WordPress space is very. Questionable in general. So it has actually learned to perfectly reproduce bad code that is all over the net for WordPress, because most tutorials are just that most code is wrong.

And the chance that you get something that reproduces broken code that was already broken in the tutorials is very high, but that being said, if it also will know about. The security functions to use and the code style principles to use. So if you just give it a random request, create a custom post type for me, it will base its answer on the types of tutorials you find about how to create a custom post type.

So it will be very rudimentary code and it will lag a lot of nuance. However, if you know what to watch out for, you can actually tell it, okay, I want to make sure that you adhere to these and these standards. I want to make sure that you’re using these and these security concepts and it will happily adapt the code to do so generally that works well. However, There’s an additional problem with that, which is the context window. The initial systems, they were not ChetGPT. It was basically you did a request and you got the result. It is a statistical model where it takes what you try, what you’re typing, and it basically auto completes it.

So that was how it initially worked. And that also is the reason why you get these random errors, sometimes these hallucinations, because it is a statistical model of just auto completing what you wrote. That means it, it returns. The result that it found that had the highest probability of being correct.

The highest probability might mean 0. 3 percent though. It might not actually be correct at all. It’s just, if that was the highest probability it found, just like your auto complete, it will return the result. It doesn’t by default have the concept of being confident in its answer or not. It just produces the one that is the most correct, even if it’s not at all correct to begin with.

That is, that gives you an idea of why these hallucinations happen. To get rid of these, you would tell it to only reproduce an answer if it has a high confidence of being correct. Otherwise, say that it doesn’t know or something like that. You can actually teach it to not give these hallucinations most of the time.

And then with ChatGPT, there was this. This iterative approach that was added. So it’s not called ChatGPT because it has a chat interface. It was actually called ChatGPT initially, because it allows you to have a progressing dialogue with with the engine to refine the answer that it is producing.

It was not about the interface that you’re using on the web for it. But this iterative approach that allows you to get a first result and then let refine it until you’re happy. So now add code style, now add security functions and so on, but then you’re hitting the limit of the context window. So the context window is how much context it can consider at one time.

And that is like a sliding. Window that moves with your, in your dialogue as you’re interacting with it. And depending on which engine you use, the context window might be larger or smaller, but it is very finite because the larger the context window, the more expensive and the more slow the computations become.

The context window is the most limiting factor in producing really complex solutions, because at one point you will ask it the next refinement. And it has already forgotten what you told it in the beginning.

Bud: I want to make sure I understand when you say context window, that is the window in which the code appears. Is that correct?

Alain: No. The context window is. is the entire input data and deduced knowledge it already has from your conversation with it.

Bud: So the entire conversation, including the generated code.

Alain: Yes. And the context window for the initial versions, because it was just one request, one answer, it was not that big of a problem.

You already had trouble getting more complex solutions anyway, because of that format. But now as it is a dialogue, you can quickly. Hit that limit because you’re just blabbering about blabbering about with it and you don’t realize that it has already forgotten what it started with. And you might your solution might actually shift in focus. And at some point it’s completely unrelated to what you actually started out with.

Bud: And so I can ask, how sure are you that this is correct? I don’t know if that’s the right way to phrase it, but if I ask, generate a custom post type with one taxonomy and three terms. Okay. And then it generates, and then I can say, how sure are you that this is correct? Is that a legitimate question or don’t bother? Cause it’s going to say…

Alain: It is a legitimate question, but the answer actually doesn’t have much to do with AI. You have the same methods of figuring out how correct it is than when you’re dealing with subcontractors or when you’re doing it on your own.

You write tests, you test your assumptions, you figure out what the problem domain is to figure out what all the different bits and pieces are that you need to consider. And most of the time, It will not be fully correct. It will be correct up to the point where you consider the problem space.

Mark: Probably what will happen is Bud will send me a message and say, is this right?

Bud: Hey, don’t put that past me. I already have you on the list to check over the code. Okay. So it’s going to be Mark AI. Okay. So we’re going to, I’m going to run it by you to make sure it’s right. But that’s so interesting.

I’m going to switch the conversation a little bit to ecommerce, because it’s on my list. And of course, this is Do the Woo. And where do you see, where is AI being used now in woop? And that’s a very big subject, obviously, to ask you, but are we going to get to the point where AI will, we’re there already, predict user behavior and then suggest, I don’t know, product choices and things like that. Where are we with all that? If you can make sense out of this question, which I’m sure you can.

Alain: Yes, I can. So in ecommerce for a while now machine learning has been used to recognize user patterns or patterns in user behavior and produce suggestions based on these identified patterns.

So if you buy. dog food, it will assume that you try to buy other articles that other people who bought dog food will also try to buy. So it doesn’t know anything about you, it just has identified that usually when people buy dog food they also buy a pet grooming accessory or whatever. So this has already been happening for quite a while.

And that is always being fine tuned to be better at recognizing these patterns, but it is happening after the fact, let’s say. This just identifies patterns that have already happened and that only works. On a large data set where you already have these patents to begin with.

So if you have a completely revolutionary product that you’re putting on a marketplace, while there are no patents to go by, this is something new. So it’s hard to get started with that. It’s only over time that you’re figuring out what patterns have emerged with that type of product that you can then base your suggestions.

What will probably happen. With these new families of AI, which are more generative in nature and which don’t happen after the fact, but can actually precede the fact is that probably instead of finding a pattern that anyone that buys dog food will also buy doesn’t this accessory, it will move to anyone that buys dog food is dog owner and dog owners will have need of this and isn’t this so that.

The first time around you’re doing that, it does not only recognize these patterns, but it attributes a property to you. That it has identified and that is only a first step, but imagine that for example, on Amazon, you have these personal accounts where people buy the most random set of things, of course, but then you have these Amazon business accounts, which are usually tied to a business that has the core business model.

There’s a certain topic to a business account usually. And in the future, you can imagine that while it figures out from what you’re doing or the data that you provided, that you are a carpenter and then it can actually figure out what you will need in the future, it might be, might figure out.

What could make sense to you that is very innovative and could help you improve your

business that no one else is yet using. So the way you can already use these AI systems now to create Completely new knowledge where it actually can reason in some capacity to figure out solutions to problems that didn’t have a solution before it could do the same in e commerce where I don’t know as a carpenter you could ask a question like I have, doesn’t this.

Problem how can I solve it? And it would actually give you a guide on how to solve your problem and put the necessary ingredients in your shopping cart. It could it could prompt you upfront okay, you are a carpenter. Right now, stocks seem to indicate that from this, in this country where you’re normally getting your wood there will be a shortage.

So you should prepare for that. You can you can think about more and more forward thinking approaches to this, where it just tries to be proactive. Instead of reacting to user behavior, it is proactive in helping to shape user behavior instead. So that is one area in, in, in which that might go. And the other area is.

Right now our user agent is the browser and one of the things that I think might be possible is that in some future, everyone will have their favorite AI system that that is trained on their own preferences and so on as their actual user agent. So instead of having a browser and going on the Amazon site.

You will communicate with your AI user agent to let it know what you’re planning to do. And it will make sure that you have all the ingredients you need, accessories you need, and we’ll just take care of it. And you will not even go to the website anymore. You will have a user agent that is not a browser, but that is your agent that acts in your behalf.

Bud: Very personalized experience that will fetch what you need for a game.

Alain: And you might have, I don’t know, you might have to choice between a frugal user agent and a luxurious user agent, or you might be able to just, I don’t know upload your brain dump and have one that just matches what you normally do. The possibilities are endless in that regard. But I think the web and also ecommerce as a result will go through some fundamental changes as clicking about on a website might not forever remain the interface of choice for interacting with the digital world.

Bud: I have a feeling you’re right. So before we wrap up, I just want to ask Mark, and he alluded to this a little earlier about WS Form and his, I guess this is a new feature. Of generated forms. This is so mind boggling. I saw it this morning and can you tell a little bit about this? Wow. We don’t need templates anymore for forms.

Mark: Yeah. It’s the next step in form generation. Yeah, a lot of people use AI right now for generating text content or for image content. And also, spitting code out and using that to create posts seems to be a very common use of open AI. I, I started playing around with it and just asking it to.

Give me, I, I might get a little bit too technical here, but I was basically asking it to give me a JSON packet back, right? So a JSON is essentially a description of an object in programming language. And I said to it, could you please give me a J ss o packet back that describes a form that is a contact us form?

And it gave me some data back. It had some preamble on there about, Hey, Here’s an example of how you can do that and everything else and using prompt engineering and what prompt engineering is that you’re basically engineering what you’re asking it a bit like Elaine was talking about earlier on with programming.

So you could say to it, make sure you use these coding standards or make sure you code it in this particular way. We trained it and trained it until we got to a point where we could get a pretty consistent packet of data back that we could then use to generate a form and we got it to a point where you can literally now say, create me a contact us form or say, let’s do a better example.

Maybe a a dog adoption form, right? And you want to be able to ask what breed you want and how old the dog is. And, what if, what, any other questions that you want to ask on that form, and it will then build a form for you. So we’re taking that output from OpenAI and then it then generates.

The form for you using the engine that we have in ws4 because we use that for other things like creating Forms to create posts or integrating with third parties like mailchimp So we extended that functionality with open ai and it works remarkably. We’ve also extended it to be able to you can ask it to add help text To fields to aid with accessibility and it will add that information So we give it parameters and we tell it, you know where we want to see that in the data and on the whole Yeah, probably five percent of the time.

There’s a blip and it will just come back with something completely crazy and we can’t work with it and but it’s interesting because every time you run it you get a slightly different style of form back And it’s interesting what it comes up with. You can literally, it would also populate select dropdowns, checkboxes, radios.

So you could say, yeah, I want a country selector on that form. Please populate that with countries. And it does it. It puts all the countries in for you. So really we implemented it because we wanted it to be a time saver for developers. We didn’t want them to have to drag and drop all the fields onto the form to get to where they wanted to be.

This gives them a good step forward. They can just start off with that form and then they can then adjust the form layout and adjust the conditional logic or anything else that they want around that form. So our next challenge with that and it’s really quite fun to work with, is we want to see if it can generate conditional logic style output.

So we could say. Okay, create me a form that has a billing and shipping address and Add a check box that will copy the billing address to the shipping address. And it’s interesting Just playing with it and seeing what it comes back with.

Bud: And once the form is generated, there’s the only way to iterate on it is to actually manipulate it with a keyboard and mouse, right?

Mark: Currently. Yeah. Yeah. But as Alain was saying earlier on you can use that iterative process where you can add more content to it to modify that form. That would be another great use case of it. So once that form’s created, we could literally put a chat prompt at the bottom to say, okay, add a country field to this form and it would add it to it.

So Yeah, it’s interesting. I enjoyed doing it because it was a different use case for OpenAI that I hadn’t seen being done before. Actually using the coded output that it gives us and using that in an actual application. So I’m sure we’ll see a lot more of that.

Bud: And one last thing, when I’ve experimented with this, I’ve noticed that unless you have the vocabulary to describe exactly what you want, it’s just, it’s not brilliant. It is brilliant, but not brilliant.

Mark: Yeah. Yeah. You need to give it some prompting for sure.

Bud: You have to be able to define in a certain way, what it is that you want.

Mark: If you were to just type in, create me a dog adoption form, it’ll do it, but there’s probably things in there that you don’t want personally.

You can ask it to add a file upload field to the form and it’ll do that. You can tell it what you want the labels to be. You can even say, create me a I don’t know, hotel booking form in German, and it will do that. So it’s actually, cause OpenAI is not only English, it’s absorbed every language on the planet, basically.

So you can ask it to generate things in different languages too, which is pretty cool.

Bud: Really phenomenal. Hey guys, thank you so much. This is a treat to have both of you here. Wow, I don’t know two better guys to talk about this stuff with, really. And Elaine and Mark, thank you so much.

Now, before we go, I want to find out where… You can be contacted. So let our listeners know. Mark, where is a good way to contact you?

Mark: You can get hold of me through our website, wss form.com, or I’m on Twitter or X or whatever it’s called at the moment that you listen to this podcast on X. My personal is @westguard, w e s t g u a r d or WS underscore Form.

Bud: Fantastic. And Alain our guest thank you so much for doing this. And please tell us where we can contact you.

Alain: Yeah, I have a personal website alainschlesser.com which has all the other information where you can reach me, but generally the easiest way to reach me is on X Twitter under S C H L E double S E R A. That’s my X Twitter username.

Bud: And you can reach me, Bud Kraus at joyofwp.com. That’s Bud at joyofwp.com or my website joyofwp.com or I don’t know, Slack, X, Facebook, whatever WordCamps, whatever I am. And And that’s it, really. And a big thanks, of course, to our friend BobWP, Bob Dunn, who let us carry on for the last hour or so. Thanks, Bob.