Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
/generative AI: beginner's guide
41 Plays19 days ago
Can you build a working AI chatbot without any prior AI knowledge? Rachel did—and here’s how. In this episode, senior software developer, Rachel Clifton joins James to share how she tackled her very first AI project—building a Retrieval-Augmented Generation (RAG) chatbot from scratch. Rachel breaks down the beginner-friendly tools that made it possible, including Spring AI, Apache Tika, and PGVector, and opens up about the challenges of chunking, hallucinations, and missing metadata. If you’ve been curious about AI but don’t know where to start, this episode is a practical, encouraging place to begin.
Recommended
Transcript
Intro: AI Implementation Journey
00:00:02
Speaker
Yeah, I think honestly demystified is ah is a good way to describe it. It did feel like, oh gosh, I've never done anything like this before. I could never. It's way past anything I've ever done. And now I'm like, oh, no, I feel like I could, you know, implement an AI solution with the libraries and things that are out there. I feel like it's not as scary as I once thought it was.
Meet Rachel Clifton: Generative AI Experience
00:00:30
Speaker
Welcome the forward slash podcast where we lean into the future of IT by inviting fellow thought leaders, innovators, and problem solvers to slash through its complexity. Today we have with us Rachel Clifton. Rachel works with us here at Caliberty. We're going to talk a little bit about she had a recent experience where she was building a generative AI solution. It was an internal tool we have here at Caliberty.
00:00:51
Speaker
Thought it might be fun to talk about that. Rachel is a senior software developer, mom of two, and wife of a trap rock superstar. Extremely passionate about mental health awareness. So just using this as a shameless plug that May is Mental Health Awareness Month and...
Trap Rock Explained: A Musical Interlude
00:01:11
Speaker
988 the crisis support hotline supports both calls and texts if you need help please reach out rachel welcome to the podcast hi james thanks for having me okay you gotta explain this i got i had never heard of trap rock before so i gotta hear about this and Okay, so to be honest, I'm i'm going to butcher this. I'm going to show this to my husband. He's going like, what are you talking about? I can't believe you said this is what that was.
00:01:36
Speaker
He's a superstar too. Like, come on, you got to get this right, lady. Okay, so trap is just a particular style of rap and hip-hop. It's loud, though, right? It's like there's like then some screaming. So that's that's like the rock part.
00:01:50
Speaker
That's the rock part. So trap is just a very particular, not very, but it's a particular style of rap and hip-hop music. It's just a genre of music that is in that realm.
00:02:00
Speaker
And then... This is like a fusion with a harder rock as well. So they have some screams. It's really cool. that there isn't There aren't a ton of bands out there with this with this combo style. So it's pretty unique.
00:02:13
Speaker
ah love it. I got to hear it.
Caliberty's 'May I Month': AI Project Unveiled
00:02:16
Speaker
um so yeah we were we're gonna you know kind of dive into ah this this experience you had with with gen generative ai so we have a kind of this we're gonna doing a lot of stuff with with ai in the month of may we're kind of talking calling it our may i month um that was my like i came up with that idea and i was like usually my ideas get shot down so i just throw out really silly ones I don't know how that one made it through, but it worked. It stuck. So we're going with it, right?
00:02:47
Speaker
I like it.
00:02:49
Speaker
All right. So i thought if we're doing AI, you know, I want to, you know, we'd like to do dad jokes on the show. So I'll, I'll, I'll tell a dad joke here. So what, um, a robot walks into a bar.
00:03:02
Speaker
The bartender asks, what do you have? The robot says, well, it's been a long day and I need to loosen up. How about a screwdriver?
00:03:12
Speaker
That is very dad. That is yeah epitome of a dad joke. I got to use the database line yesterday. i got I got an email from my daughter's school about a dad's event or whatever, and he had the wrong name.
00:03:25
Speaker
he said, hey, Matthew. And I sent an email back and I was like, I think your database is messed up. Yeah. No, you didn't. I sure did.
00:03:37
Speaker
but I'm understanding more and more why your kids roll their eyes at your dad jokes. Yeah, yeah. Some pretty bad ones. Yeah.
Building a RAG System at Caliberty
00:03:45
Speaker
Okay. So let's let's get into talking about real stuff here, not silly robot jokes and whatnot. ah You were recently working on, we we have an internal we had an internal tool that we were trying to build here at Caliberty. Tell us a little bit about what what what even is this thing that you were building?
00:04:01
Speaker
Absolutely. So what we were working on building um is a generative AI solution. So um it's called RAG, um which I'll just give ah just like a brief overview of what that even means.
00:04:12
Speaker
and It's essentially an AI solution. that is using a generative model of information retrieval and the LLM, the large language model. So you're essentially providing this AI, um in this case, a chat bot with a source of documents that we want it to reference when creating its responses. So instead of when you just like go to chat GPT, you're asking it questions, it is using the entire world that it was trained on.
00:04:46
Speaker
And ah in this case for a RAG system, we are processing our own documents and having ai use that as its source of truth and what it's getting information from.
00:04:59
Speaker
um For this case, it was a sort of a help chat bot for Caliberty. And it was to be able to look through all of our our documents, our posted case studies that are online, ah podcast transcripts just like this, any blog posts, things like that, in order to help sort of our sales team find relevant ah material for anything that they need.
00:05:26
Speaker
um where We called it Cali. Cali. Cali chatbot. Going back to Cali. So with Cali, so you said it's RAG. So that's that's retrieval augmented generation, right? So you're retrieving things. and hand it Okay, we'll get into that further a little bit about what that even means.
00:05:48
Speaker
So you...
Challenges in AI Development: Overcoming Daunting Starts
00:05:50
Speaker
You're getting into this and it sounds like you're just this. This is old hat for you. You've been doing this for years and then you had all the expertise and everything, right? Not in the slightest. I had never done anything with AI at all before for development. So I was honestly, when the ask first came up, I was like, hmm, I have zero experience with this and I will not be able to do any of it. But at the time I was in but between projects. So I was like, sure, I'll give it a shot.
00:06:19
Speaker
But I was honestly pretty, um was little overwhelmed and felt it felt really daunting, um the thought of doing this very specific type of AI ah rag. I was like, well, what does this even mean?
00:06:33
Speaker
um yeah I was nervous to say the very least. Yeah, there's there's a there's a lot of information out there on you generative AI, almost too much right now, right? Just sifting through all of that and knowing what's what's the garbage and what's the good stuff, right? So yeah, it can be very hard.
Technical Deep Dive: Tools and Techniques
00:06:50
Speaker
So how did you how did you get started? what did what did you How did you kick things off? What did you what you do? ah Well, um I was, thankfully, i worked with Mary here at Caliberty. She gave us some good places to kind of start. There are so many libraries out there. So I'm a Java developer, and so that's what I was writing this solution in.
00:07:10
Speaker
And so thankfully, I was pointed to some various libraries and tools. um I used... for for each kind of piece of it. So there's a whole process of it needs to essentially ingest all of the documents in a way that the LLM is able to use it.
00:07:28
Speaker
um And so it was suggested maybe trying PG vector. um And there's so there's a whole library that makes that easy. And PG vector, that's that's like an extension of Postgres, right? Like the right the regular old Postgres database to to treat it like a like a vector store. OK, cool. Yeah. So um I was able to use that. And then i am actually, like i said, I'm a job developer with Spring. There's actually a Spring AI. So I was, I found that and was like, oh, perfect. Like, um hopefully that'll be easy to use. And it was, it was extremely straightforward to create a chat client using so Spring ai that library. it was amazing.
00:08:09
Speaker
honestly, shockingly easy. I was surprised how how easy it was to just build a basic chat system and then expand on it to make it this RAG model. Okay, so you load up Spring AI and you're you're getting started. So what's what's kind of the first thing what the the first hurdle? What is the first thing you kind of have to tackle when you're when you're kind of implementing these these RAG solutions?
00:08:33
Speaker
So the first thing is um making the the vector store, populating it with your data. So um I wasn't familiar at all with what a but vector store is outside of learning vectors when I was in...
00:08:49
Speaker
in math in school. And so I know that it's, you know, vectors, it's direction, magnitude, relationships to things. And I still don't know the nitty gritty of how a vector store exactly works, but I understand that when you are storing it this way, it's essentially pieces of a document and how they are related to other pieces of other documents.
00:09:10
Speaker
um So that the LanguageVal is then able to pull relevant information. um But so this PG vector, you essentially have to take what I did was I took files and I had to use a document reader.
00:09:24
Speaker
used a Tika document reader, which just kind of processes the document, splits it into chunks, um and then mutates it to be applicable to this vector store and then stores it.
00:09:39
Speaker
So ah that's the first step. Yeah, I mean, that's, uh, yeah, I mean, so you're, so you're like, like the, you're hitting on a really important part of, of these RAG solutions. One of the, one of the quick hurdles you run into is like, okay, I've got this big old document.
00:09:54
Speaker
yeah How do I break this thing up so that it's like not too little and not tell us a little bit about that. What's, what's kind of involved that that chunking mechanism. Absolutely. So, yeah, so you don't want to store the and entire document because there's so much information stored in each and every document. And so having those relationships, you are splitting the documents into chunks. You don't want to make them too small because if it's, let's just say, a sentence without the context of the sentence around it, it might not make any sense or be able to have a valid relationship to other pieces.
00:10:28
Speaker
If it's too big, you might miss um some other relationships. So, yeah. Depending on your use case, You have to play around with how big of a chunk you are. So you like it's you don't want it too big. You don't want too small.
00:10:41
Speaker
Just right. So like that Goldilocks size for your particular solution, your particular documents. And so if you are implementing a RAG solution and you notice, oh, I'm missing, it's not pulling back the data I thought it would. It's not referencing what I thought it would.
00:10:57
Speaker
You might have to go and be change your your sizes of your chunks in order for those relationships to be to be processed in the way that you are expecting and wanting.
00:11:08
Speaker
um Also for my particular solution, I also wanted the file path. So I wanted to be able to find the file later that it was coming from. So part of the problem is you are splitting up these documents into chunks.
00:11:24
Speaker
I kind of had a struggle with getting my chat bot to access the file path. I had to do so much debugging. because I saw the file path in the metadata, but the way that the language, the large language model worked with accessing this data, it only had the text content of the chunk. It did not have the metadata. So I had to do sort of a, kind of, in my opinion, a little bit of a gross implementation of manually adding a string with the words file path and then the name of the file to each and every chunk that I had so that I could later access that. So one thing that you also have to kind of battle is what your um but your ai client, your chat client,
00:12:14
Speaker
what that has access to. um If there's anything important that you end up missing when you are doing that chunking, um you kind of have to keep that in mind as well when you're creating your vector store.
00:12:27
Speaker
And the way, ah the reason you wanted that that file path, kind of the solution was the idea since it's to help me find files, that's kind of the idea that helped me find assets. Knowing where to go get that thing and so I can share it with someone that's, that's kind of important as opposed to the chat bot, just spitting out just the, the answer of what you're doing just talking to you.
00:12:47
Speaker
The, the goal is to actually, actually be able to go get that file. So that, that was why that was important.
AI Implementation: Challenges and Surprises
00:12:51
Speaker
Yeah. Okay, so and then I have heard people talk about chunking and and one of the things that people, and I don't know if you even got into this or not, but like they they talk about like kind of doing like an overlap so that that you, you know, just in case you have a, you know, ah paragraph and it might have some context at the end of the previous paragraph that kind of tees up the next one. and so So you kind of overlap your your chunks together so that you don't lose context between things. and did you got Did you get into that at all or...
00:13:22
Speaker
So I did not do that at all. But in my research, I did see an option to... It was actually kind of cool. Yeah, it was like k it was like almost like you would have the AI create a little bit of like a summary of what came before and what came after. and you could add that sort of information around the chunk. Oh, cool.
00:13:46
Speaker
Yeah. And so... I might be misrepresenting it a little bit because I didn't dive into it, but that's what it seemed. it So yeah, it's it was kind of cool. It was like kind of like adding context for before and after um as you're splitting it. But for this implementation, I ended up not really needing meeting to do that. But it is kind of cool the amount of, guess, customization that you can do with simply this vector store piece of it. um It's more than just...
00:14:18
Speaker
storing pieces of a document in a database. So it's, it's, there's so much that goes on in the background. It's really interesting. So, okay. So you're, you're able to get up and running very quickly, but I'm guessing it took you three or four weeks, something like that to get something actually that, that works.
00:14:38
Speaker
Actually, it only took me a couple days. I was shocked at how quickly i was able to get it working. um The longest thing was honestly that file path issue that I was trying to debug.
00:14:49
Speaker
I was able to get this chat client set up with um this doc. You essentially attach this document store as what's considered an advisor. And with the Spring AI, is so straightforward that it literally just took me I think a day to get something up and kind of running, but it was, even though I told it not to hallucinate, it was hallucinating file type extensions. So it was telling me that like, oh, the file name is, you know, XYZ.
00:15:18
Speaker
And I was like, that's literally nowhere, anywhere. Why did you say that? And so honestly, battling that was what took me the next couple days where I ended up just manually hammering that onto the end of each um file type. But getting something actually up and running and starting to work, it only took a day.
00:15:39
Speaker
Oh, wow. Don't you love how confident the LLMs are with their hallucinations, too? You know what i mean? It's like, yes no, this is absolutely true. I'm telling you, it's true. And then you and then you call them on it. Well, maybe not.
00:15:50
Speaker
you know Maybe I made something up there. yeah Okay, right, all right. theyre Yeah. very funny Well, you know, it's actually funny. The first time I tried to call call mine on it and be like like, that's not the name of the file. Like, where is that? They were like, what file are you talking about? So that was one thing I realized where mine didn't have memory. So that was something that I also had to look into where you I had to add on sort of a um chat memory advisor where I added in memory information.
00:16:17
Speaker
In-memory chat, a chat memory, in-memory chat memory. And um then then I was able to actually talk to my chatbot about what it had previously said. um But I actually found one thing is that that quickly makes your your tokens go up and it it quickly costs more money.
00:16:36
Speaker
um So I kind of put that memory on pause for a bit while I was refining other things. So I noticed, oh, i'm I'm running through these tokens pretty quickly. But... Yeah, so the the way that works is like the bigger the the request, basically, the more the more text you send to it, it's going to eat up your your your money, right? you're You're paying by the kind of the token, so to speak.
00:16:59
Speaker
Yeah. Or not, so to speak. And granted, you know, this was still just a couple bucks, but I had only put a couple bucks in it. So by no means breaking the bank, but. Yeah. What was the the most surprising thing that that you learned on along your journey
AI Implementation Simplicity: Reflections
00:17:13
Speaker
with this? I mean, when you dug into this, what was most surprising thing you learned?
00:17:18
Speaker
I guess the most surprising thing was how little you have to know about ai to implement the solution. At least with Spring AI, it was very much just...
00:17:31
Speaker
similar to calling any other sort of API. um You don't have to know the nitty gritty of how large language models work. You don't have to know with the vector store, you don't have to know exactly how um all of those work. You just have to kind of know what it does, you know, that the vector store is splitting these documents, you know, it's taking pieces of documents and storing how they relate to each other.
00:17:57
Speaker
um You have to know a little bit about like you know, the fine tuning aspects, but how it does what it does, you, it, it can stay a bit of a black box. If you are just wanting to do sort of a basic implementation, i was worried that I wouldn't be able to do anything without really, really understanding how AI works in terms of processing the, the chunks of the documents or how it works, um, in, you know, having that, that chat back and forth. But I didn't have to learn any of that nitty gritty to make a functioning solution.
00:18:32
Speaker
um That honestly was, I think, the the most surprising part. so you don't have to like walk around with like a big white overcoat and and talk like this. Exactly. stuff okay let's good audience Yeah. And um it is nice because then, um like I said, you know, I kind of dove into like the the chunking size and finding my Goldilocks answer to that. And, you know, you're able to slowly get dive more and more. the more refined you want your model to be, the more ah you want to make it do
00:19:05
Speaker
exactly how you're wanting it to function, but you don't have to understand it all right away. You can kind of do it piecemeal and understand a little bit more here and there and then build what you need.
00:19:18
Speaker
and And I think like with with those solutions, it's not necessarily it's not just um just going and grabbing document chunks out of a database and throwing it at the LLM and say, and tell me some good stuff, right? There's, you kind of have to give it a little bit of instructions and, and and how, how was your experience with that? How, how brittle was that? How easy was like, well, what about that part?
00:19:42
Speaker
I would say it was easy and brittle. So it was easy to figure out how to implement this with the ah Spring AI. So you have a default, you can um sort of add like a default advice that you're able to give it to tell it, hey, like I said, hey, you're an assistant for Caliberty. user will ask you questions. It's your job to return a list of related files.
00:20:05
Speaker
yeah It was easy for me to find where to give it that information um and also where to add information. further about like, Hey, this is an example of the, the kind of how I'd like you to return data to me. It was brittle in that it didn't always listen. Like I had told it to, uh, you know, not, if it couldn't find information about anything in particular to not hallucinate or make it up, um, to, you know, just provide, um a list of the articles that you found and,
00:20:36
Speaker
it Like I said, it heusi it created file names that did not exist. um It took a lot of sort of tweaking to get it to actually return a list format of, I found this file that is related and here's why it's related.
00:20:50
Speaker
um So figuring out how to prompt the chat client to respond to me in the way that I want, it took sort of a lot of tweaking and It's still not perfect. And I don't, it's, it's hard.
00:21:05
Speaker
It was hard for me to figure out how to get it to behave exactly as I wanted it to. um But it was easy to figure out where to do that. If it if it makes sense. Yeah. Yeah. And I think you hit on one of the, like, it's kind of a cool technique and they, they call it so that the general term is like, what is it?
00:21:23
Speaker
In context to learning, I think is the general term for it. And, you'll hear it referred to as like a zero shot or one shot or few shot. And what you were doing by saying, hey, I'm asking you a question. Here's an example of what I want you to respond with. That's that's the the one shot, right? When you do multiple of those, that's few shot, right?
00:21:43
Speaker
So you're, but you're letting the model learn what you need in the context that you send it. That's the the concept, but that's, so yeah, that's, that's one of those, those techniques that you have to kind of do. And, and yes, it can be a little brittle and you got to worry about those hallucinations. So that's always fun.
00:21:58
Speaker
So since you you had that experience with Callie and digging in, have you had any chance to kind of, you know, poke around with anything that that did any further learning or reading or anything like that? Have you had any opportunity to do any of that?
00:22:15
Speaker
I haven't, um but I definitely feel more comfortable if I... were to dive in more in the future, but I started a project and kind of just dove right into that. So ah a project assignment at work. yeah um So unfortunately I have sort of left my Cali and AI research a little bit behind, but I feel, i feel a lot more confident if anything were to come up at my job in, in the project, that's like, Hey, this has to do with AI. It feels a lot less daunting than if I,
00:22:52
Speaker
We're told to do that or asked to do that at a client without this little bit of this little taste. So you haven't had an opportunity to go back to Cali. I'm sorry. just I just love that that name. i
00:23:07
Speaker
I like LL Cool J. What can
Overcoming AI Fears: A Personal Journey
00:23:09
Speaker
I say? I can't help myself. ah And so if if you are, and you kind of hit it a little bit, um You know, there's there's a big trend right now of of kind of demystification trend in the market when it comes to ai and gen and in particular.
00:23:25
Speaker
ah But would you consider yourself demystified at this point that you you've you've kind of gone through the journey and it wasn't as scary as you thought it was going to be? Yeah, I think honestly demystified is ah is a good way to describe it. It did feel like, oh gosh, I've never done anything like this before. I could never. It's way past anything I've ever done. And now I'm like, oh, no, I feel like I could, you know, implement an AI solution with the libraries and things that are out there. I feel like it's not as scary as I once thought it was.
00:23:55
Speaker
Yeah. So even if it was a more complex type of a thing you needed to build, you would you would be able to take you you'd be able to take a running start at it and and be and be very confident that you can tackle it, so to speak.
00:24:08
Speaker
Yeah, I think so. the The resources that are out there, um you know, i I did this purely just through finding documentation and examples online. So if I were to dive sort of If there was something more complicated, i sure that you know even doing like a tutorial or watching videos or something more in depth, I barely had to, I was so shallow.
00:24:32
Speaker
I did not dive deep at all and was able to do this. So um yeah, I feel like with the resources that are out there in addition to the libraries, I could create a more fine-tuned, complex solution um if needed.
Guest Segment: Deciding on Topics
00:24:48
Speaker
So we should probably start our ship it or skip it round. We have a section of the podcast, a segment of the podcast, if you will, that we call ship it or skip it.
00:25:00
Speaker
Ship or skip, ship or skip. Everybody, we got to tell us if you ship or skip.
00:25:06
Speaker
Okay, so the idea here is it's kind of like ah hot or not, right? Like, is it is this, yes, let's go with it, or no, let's not. um I think we talked a little bit about this, but I would, i would um you know, maybe sometimes people are like, it depends, or maybe they're they'll they'll kind of catch their answers. But like, what about getting formal AI training? Like really, you know,
00:25:30
Speaker
going into understanding things a lot more, at at a lot more deep level. Do you you think that's a ah ship it? Like if you had the time, would you do that? Like, or or is it a skip it? Like, no, you're never going to need it. Just just go on with it.
00:25:44
Speaker
I think, um honestly,
00:25:49
Speaker
If we're going to do just like a hot take um blurb, it's going to be skip it. But I will add some sort of um ah more details there. So I think if you're just getting started and if you're wanting to implement um something that isn't extremely complex, skip it. You don't need to know it all. You don't need to really do formal training.
00:26:14
Speaker
you can just, like I had kind of mentioned, you can keep it a black box until there are pieces that you need to fine tune and you can dig in a little bit more there. um However, if AI is something that you're really interested in, i think that the getting some more formal learnings is almost necessary just because of how much is going on. like Inside that black box, I'm sure like it's it seems i can kind of tell like it's going to be some pretty complicated things happening. And if you really want to understand that and you really want to um know that, I don't know if just by Googling and just you know reading you know random
00:26:59
Speaker
documentation about it will get you to sort of an expertise level of what's really going on. i don't think that's necessary to add to develop using AI. I don't.
00:27:11
Speaker
So, you know, kind of two so two sides of the same coin, I guess. but We get a lot of those answers because we have a lot of people who are a architects and they're like, it depends, right? Architect answers.
00:27:22
Speaker
Yeah. I agree with you. I think, and um I heard a phrase the other day, somebody said, it was, you know, let the complexity earn its place ah in in your solution. I love that, the way that was phrased. like that. I think that's really cool. um And I agree, like, as you said, treating it as a black box, it it works. I can get something, a solution together that adds value to my life, makes my life easier.
00:27:47
Speaker
without having to dig into it. Okay, that's great. Um, now of course ours was an internal tool and if anything went wrong, it's not really, you know, no no planes are flying out of the sky, you know, falling out of the sky or anything. But yeah, I mean, if you, there would be some due diligence if it was a more, you know, polished product or something, of course. But yeah, I think from a business case standpoint there, you can get a lot of bang for the buck with very minimal investment upfront, right? You can get quite a bit of value out of these things. So I, and I agree with you now I'm a math nerd, right? I mean, and I know you are too, right? Like that's, yeah.
00:28:20
Speaker
Very much a math nerd. so it would be interesting to see if you, if you did do a deep dive, how, how much you would geek out on the, on the math involved. But, um, yeah, I, I don't think you have, you don't need to know linear algebra. Like you don't have to like have a bunch of GPUs flying around, you know, heating up your house or anything like to do these things. It really is fairly simple.
00:28:40
Speaker
Yeah.
AI Tools in Development: Pros and Cons
00:28:41
Speaker
All right. So the next ship it or skip it question, we're going to kind of shift gears a little bit, but like what I'm sure, you know, everybody's kind of encountered this. If you're a developer, a lot of the tools nowadays are incorporating AI. did do Do you leverage that? Did you leverage that? What was your experience of of of that? And do you say, yes, this is good stuff or like, oh, my gosh, it's a hot mess.
00:29:01
Speaker
but you What do you think? well I think, yes, it is good stuff, but... not to just blindly use and accept. the there it's It's nice when you have to do some kind of boilerplate, repetitive um code. I've noticed it can kind of pick up on things and be correct.
00:29:23
Speaker
The number of times that it has tried to autocomplete my code to utter garbage is just, it's it's a lot. I would say the amount of, it's probably 50-50 split, if whether it's helpful or if I'm just like, why did you even suggest this? Please go away.
00:29:41
Speaker
So I would say it can be helpful if you're trying to just get rid of some, you know, repetitive things that you have to do, like te when you're writing tests and things like that.
00:29:51
Speaker
um It can be helpful because it picks up on what you've tested before. But then there are times when I'm writing code and it It is just, it makes a suggestion that seems like it's out of left field.
00:30:03
Speaker
um And so if I were to just blindly accept like, oh, yes, cool. You just suggested a whole helper method for me. It's named exactly as I'm what I need it to do. Cool.
00:30:14
Speaker
And let's, let's go with it, but not actually look at what it's doing. um No, yeah you definitely need to be taking it as a starting place. and go through everything that it puts out because it I don't know where gets some of these ideas from.
00:30:30
Speaker
I, it's, it's really funny. It's, I recently went through a lot of this. I'm writing ah a little side project and I, and I was using chat GPT and then the built-in thing in my, in my IDE as well.
00:30:42
Speaker
And it's funny. Some of the things is like, wow, that was really nice. Thank you very much. Right. And it works perfectly out of the box. But the number of times where it's like, like, especially with like testing, I like to, I have to use it for testing. It helps me write tests very quickly. I can usually pick up on good patterns there, but like,
00:30:57
Speaker
it was It was trying to generate a test. I'm like, no, no, that's not the test I want. I deleted it and i wrote the test that I wanted. And then I'm like, okay, let's go to the next next test. So it suggested that same exact test again. Like, oh, no, no, I really want to do this. Right. No, that's not, I don't want to do that. Like it kept doing it over and over. How do you get stuck in this loop? Like, yeah.
00:31:16
Speaker
And then same thing with like documentation. Like it is usually pretty good about, you know, like it's Java. So if you do like a Java talk or, you know, just ah a documentation block for a piece of code and you start, you know, you begin the block and it'll, it'll kind of generate and say, well, this is the documentation. A lot of times it's good.
00:31:36
Speaker
But a lot of times when it's wrong, again, it gets, it's stubborn. It gets mad that you don't take a suggestion. So I don't know, I'm, you know, anthropomorphizing or whatever. And then like, it's, yeah I'm like, no no, that's not what this method's doing. Like, and then I type out what it's doing and oh, okay, fine. I'll help you complete your thought.
00:31:52
Speaker
Fine. And then I go to the next method. It tries the same stupid docs for that. And it has nothing to do with the code. Like, ah no, I don't want that. Don't give it to me. I, and then the other thing that I've found is When it's wrong, again, the hallucination, it's convincing. And then the the particular library i was working on was it and it did involve some math.
00:32:13
Speaker
And it's like, yep, here's the algorithm. Here you go. Perfect. Ready to go. This is going rock. And then it pats you on the back and says, you're you're doing such a great job. You know? So I love that kind of stuff.
00:32:24
Speaker
Yeah. And then I'm running tests and all that. I'm like, man, there's something wrong with this. And I paste the same exact code it generated back to it. And I say, are we sure about this? Oh, no, that's got a bug. up that's Yeah, this is wrong right here. Like, it it corrected itself. I knew it was wrong. Why'd you give it to me?
00:32:40
Speaker
Yeah, yeah. that's That's the best when it when it corrects itself. Yeah. Yeah. So it's like, oh, yeah. Yeah, I completely told you the wrong thing there. Yeah. i So I am, I would say overall, though, I would say I'm kind of a ship it, but I'm with you. Like you can't just blindly do this stuff. It's not, it is not ready to take over our world completely yet. Right. It's, it's definitely got a long way to go.
00:33:03
Speaker
And I do have qualms with my IDE because once it started implementing the AI autocomplete, it stole my keyboard shortcut for just like autocompleting the variable name or the class name. So I used to click tab just like it was like, oh, cool. It's in the dropdown. It knows I'm about to select XYZ class.
00:33:23
Speaker
I click tab and instead it autocompletes this AI line and I'm like, oh, shoot. backpa So i'm sure I know there's a setting to change it and It's irritating enough that I get frustrated, but not irritating enough that I am going to change the shortcut back.
00:33:39
Speaker
So I just complain about it. we are We are very weird creatures. It's funny, like one one little thing can make us mad and we're like, I'm going to write an entire library to solve this problem. And then there's another thing that you know bugs us every single day and we just, I'm going to live with that.
00:33:54
Speaker
yeah Yep. yeah And still not learn. Like it's been, I've had this AI on here for, I want to say a year and I still will click, click tab to complete. I just can't click enter. i don't know why. That's a trip.
00:34:06
Speaker
All right. So we, we do have a a fun section of, of our show. We call our lightning round. Now this is the very, you know, and it's fun, but it is serious. Like there are, um, there are correct answers to these questions. So you want to, you know, we can't share them with you ahead of time. Cause you, I don't want you chat GPTing this and getting the right answer. So, uh, we, we, these are ones that we will spring on you. And then it's, it is important that you take it seriously, uh, cause you will be graded.
00:34:33
Speaker
Yeah.
Lightning Round: Personal Preferences
00:34:52
Speaker
Number one, Ariel or Jasmine? Jasmine. Fair enough. My dog's name is Jasmine. I'll go with that. ah
00:35:02
Speaker
Favorite number? 93. ninety three 93. Probably during your own born. Nope.
00:35:12
Speaker
yeah Do you want to know? Well, sure. Yeah. Why is it 93? um Because, ah well, that the current in this day and age, an a is, you know, 90 to 100 or 91 to 100 or whatever.
00:35:25
Speaker
But um when I was in school, it was the seven points. So 93 was the the lowest you get for an A. So it just it just became my... I don't know. It's like you wanted to get a 93 or above. So 93 became my favorite I was in that 93 era myself. Yeah. When my kids were like, Oh, I got a 90. I got an A. I'm like, no, you didn't.
00:35:43
Speaker
No, that's right thats's not even a B plus. Like that's a B. Yeah. Yeah. They're a little easy on these kids. Anyway, back in my day. Right. Okay. ah What's your favorite carb, bread, pasta, rice, or potatoes?
00:36:00
Speaker
All of the above. All of the above. Yes. That's surprising. and That's the right answer actually. yeah done How many cups coffee do you drink a day? Cups of coffee? drink one cup of coffee, but I have two or three cups of caffeinated beverages.
00:36:17
Speaker
All right. What's the maximum number of spritzes of perfume before it's too much? I would say and depends, but no matter what, three would be the max.
00:36:32
Speaker
Like... Never go to four is what you're saying, but three is... Four is right out, as they said in Bonnie Python. Okay, I got it. I like it. For a journal, if you're journaling, do you like to do it on paper or on a digital format like a computer or something?
00:36:49
Speaker
You know, I like both. um I think they both have their place for different um different types of journaling. and But I think ah pen and paper is is what I prefer.
00:37:02
Speaker
I guess said it depends on what I'm journaling about like note taking, I guess is one thing I do a lot of. Right. But yeah. Yep. If I'm trying to recall something, if I want to be able to recall it, I find pen and paper to be.
00:37:15
Speaker
Yes. If I want it to stick in my brain. Yeah. Pen and paper are so much better for me. Yep. What's the most boring thing ever?
00:37:26
Speaker
Generative AI. I'm just kidding.
00:37:30
Speaker
So easy. It's boring. um the most boring thing ever. and don't know. Is this a question you ask a lot of people? Do people have things that they think are boring?
00:37:41
Speaker
um yeah I don't think I've ever asked anyone that question before. I didn't. It is kind of a tough one to think of like what is the most boring thing ever. say I'm a very ah impatient person. So a lot of things are boring to me.
00:37:57
Speaker
Oh, yeah. Maybe treadmill running for me. I don't don't like. Oh, treadmill one. it's Yeah, that's the worst. That's awful. Yeah. Okay. I'll steal that answer. That's fine.
00:38:08
Speaker
All right. ah This one's an interesting one. this This gets some different answers. We have asked this before. What temperature do you do you like your thermostat at? Oh, 72. 72.
00:38:18
Speaker
okay seventy two 72. Yeah. All right. No, my, my rockstar husband and I disagree. Yeah. All right. This is fantastic. So that, that concludes, uh, you did very well.
00:38:30
Speaker
i I must say I was surprised at how well, honestly, I, I think it's the best ever, honestly. Well, thank you. and I'm sure you say that to all your guests. I do. Yeah.
Closing Message: Mental Health Awareness
00:38:45
Speaker
Uh, so any, I know you mentioned something about, um, in your bio about May being Mental Health Awareness Month. Any anything closing reflections, thoughts about that in particular? I know that's ah an area of passion for you. Yeah, I would just say to anyone out there listening that you matter. You are important and you are not alone.
00:39:07
Speaker
um So many people are have gone through or are actively going through struggles with mental health. So please reach out.
00:39:18
Speaker
Even if you feel like you are the only one or nobody cares, people do care. And that 988 number, like I said, sometimes texting is easier.
00:39:29
Speaker
Just reaching out to that instead of calling, reach out to a friend, a family member, you are loved and you are important and there is hope.
00:39:41
Speaker
And there is not an AI on the other side of that line. That's a person. Correct. yeah It is a person. It is trained crisis counselors. So yeah, you will not get the runaround like you do with all these chatbots these days. So correct.
00:39:54
Speaker
Okay. Awesome. Well, thank you, Rachel, so much. This was great. I really enjoyed it. Always enjoy chatting with you. We enjoyed having you on. Welcome back anytime. Absolutely. Thanks so much for having me.
00:40:06
Speaker
If you'd like to get in touch, drop us a line at the forward slash at Caliberty.com. The forward slash podcast is created by Caliberty. Our director is Dylan Quartz, producer Ryan Wilson, with editing by John Corey and Jeremy Brown.
00:40:18
Speaker
Marketing support comes from Taylor Blessing. I'm your host, James Carman, and thank you for listening.