How'd you like to listen to dot net Rocks with no ads? Easy? Become a patron for just five dollars a month. You get access to a private RSS feed where all the shows have no ads. Twenty dollars a month, we'll get you that and a special dot net Rocks patron mug. Sign up now at Patreon dot dot net rocks dot com. Hey there, this is Jeff Fritz, the Purple Blazer guy from Microsoft, letting you in on
a little secret about my friend Carl Franklin. You know, the guy who started dot net Rocks, the first podcast about dot net in two thousand and two, The guy who's been teaching Blazer on YouTube since twenty twenty. Yeah, that Carl Franklin. Well, Carl's joined up with the folks from Code in a Castle to teach a week long hands on Blazer class at Are you ready to get this? At a castle slash villa in Tuscany. It's sort
of a luxury vacation with Blazer learning built in. Carl's calling it the Blazer master Class. You'll learn Blazer from the ground up, finishing the week with the ability to build and deploy Blazer applications. Since The training happens for only four hours in the morning over six days. You can bring your significant other your partner with you and you should right This part of Italy is absolutely beautiful.
There's so much to see and do and in Larion Marco from code In to Castle are organizing daily activities both at the castle and in the area. The castle is in the Marema, a less touristed region of Tuscany, offering both classic Tuscan hill country as well as easy access to the Etruscan Riviera, with sublime local food, wine and olive oil around every corner. Breakfast is included every day. There will be two communal dinners at the castle, book
ending the experience, and most other meal and all activities are included. And did I mention you'll learn Blazer in person from Carl Franklin listen. Space is limited and for very good reason. This is quality training in a beautiful setting. Go to code Inacastle dot com slash Blazer twenty twenty three that's bla z O R two zero two three to take advantage of this amazing opportunity to join Carl in Tuscany for an unforgettable week of La Dolce Vita while advancing your programming
skills in this important new technology. After building software for a while, you know it's only a matter of time before you see an HTTP timeout or a database deadlock. In software, it's not a case of if things fail, but a case of when one mishap like this and valuable data is lost forever. And these failures occur all the time, but it doesn't have to be
this way. Introducing and service bus the utimate tool to build robust and reliable systems that can handle failures gracefully, maintain high availability, and scale to meet growing demand. For more than fifteen years, end service bus has been trusted to run mission critical systems that must not go down or lose any data ever. And now you can try it for yourself. End service Bus integrates seamlessly with your dot net applications and could be hosted on premises or in the cloud.
Say goodbye to loss data and system failures and say hello to a better, more reliable way of building distributed systems. Try end service bus today by heading over to go dot particular, dot net slash dot net rocks and start building better systems with asynchronous messaging using end service bus Welcome back to dot net Rocks. This is Carl Franklin and it's Richard Campbell and Mark Miller is here joining us. Nothing. I keep telling you nothing. It's going to go
wrong on this show. It is not going to be Mondays. However. This is clearly dot net Rocks. But we'll talk to Mark in a minute. But first of all, buddy, how you doing. I'm all right. Things are good up here. We're you know, getting organized, and we're knocking out a few shows. Air clean. Yeah, you know, it's nice being on the West coast where the wind comes off the ocean. Yeah. Yeah, there's a forest fire. You know, it's forest fire
season started very early here. Yeah, so there is a fire inland a bit and it's creeping. Depending on how the windflow goes, it creeps him a little bit closer. But it's all right there. New York City got hammered, but we we sort of were on the outskirts of it, but we still had over two hundred two hundred and fifty air quality poor air quality. And next well, we're evacuating towns that are burning to the ground and trying to save so, yeah, breathe carefully while we try and not kill
everybody. Yeah, that's that's true. Good good luck up there. Yeah, the good news is that the US is sending firefighters and water bombers. Like we all help each other during the convenient for us to have our season early, so when your later season arrives, we're available to help out. That's so awesome. Thank you for that as part of the system. All right, well, let's roll the crazy music for a little thing we call
better Noah framework awesome? All right, man, got Well, you probably heard Jeff Fritz in the introduction there talking about this crazy idea that a friend of mine had to rent a castle in Tuscany and have a week long training class there, and Glazer and toward it, I sort of coined the term traincation, yeah, because it's just as much a vacation with sightseeing and touristy stuff as it is training. In fact, the training is only for four hours a day for six days, and the rest of the time you get
guided tours of the surrounding area. So the only problem I think is that it's a little pricey. You kin kind of have to consider it both training budget and vacation budget, right, Yeah. So yeah, well, Tuscany's not cheap, and flying is certainly not to cheap these days. That's it's not Yeah, it's not cheap. And the other thing is that you can't just this castle is not a hotel. It's a villa that you have to rent the whole thing right for a week at a time, and that's completely
outside the range of most people's ability, you know. So that's it. It's at code in Acastle dot com slash Blazer twenty twenty three. Cool. And today when this comes out, is the end of the deadline for the early bird discount, so you'll you'll have to pay full prose. But we need to we need to get you know, who's interested in going. Richard, here's going our friends Hunter. That's fun. He's he's not committed yet, but he said there are a couple of things have to work out and
then he's there. That's cool. Yeah, well yeah, I'm committed to playing No s Ferrato and hiding in the shadows and scaring the people trying to learn all the basics. He really, I sent in my confirmation, Carl. I know you didn't ask, but I sent you that confirmation. Okay, great, it is mondays, isn't it all right? Well, might just be. It might just be so, Richard, you got a comment
for us? Who's talking to us? I grab a COMMENTOP Show eighteen forty eight, which is the one we did at Techorama in Antwerp with doctor Jodi Burchell from and I figures. As we were talking about large language models with her, I say, spect, we're gonna talk a little bit large language models today. It kind of made sense. Yeah, And so Nick, as a Garion said, there's an excellent episode. I really liked the idea of using AI to validate and verify an AI. Yeah, that's an arms
race for you right there. And I understand why it's never going to be able to be used long term or even now. We'll see. I think one possible regulation would be similar to other rights. AI generated visual visuals like images of movies should have a watermark that can be tracked, and players can display that the images are flagged when an image has been modified, or show
an infull length to the problem that generated it. Yeah. I wonder if AI tools could just detect that in the first place, may or may not eat a watermark. They kind of leave watermarks anyway, because of the way they make their images. This would be similar today's copy machines not being able to copy currency or only being able to display blue ray content versus an HDMI contem protended cable. Boy, we love that feature. Remember content protection?
That was the best, the best. Look how popular that is today? There you go, Yeah, bring back copy protection blue rays. Oh boy. I think the AI overall is going to have a huge benefit demand kind,
but only if they are appropriate regulations. I think a top secret committee could be created that inventors could divulge their plan adventures to regulators to be ready at launch, but IP could be protected via NDA's I think that's what patents are for, actually, Nick, like you know that way, you can divulge the stuff, but if anybody uses it, they get to pay you anyway. I mean, I appreciate that folks are thinking about how we're going
to manage this stuff. I also think we're in the midst of a really serious hype cycle that's distorting reality pretty heavily. Yeah, and then that hype cycle will pass yep. Yeah, well so, Nick, thank you so much for your comment and a copy of music cod Buy It's on its way to you and if you'd like a copy of music cod Buy, write a comment on the website at dot and rocks dot com or on the facebooks publish every show there, and if you comment there and everybody on the show,
we'll send you a copy of music Code Buy. And you can follow us on Twitter if you want to. But the real fund happens over on masadon. I'm at Carl Franklin at tech hub dot social, and I'm Rich Campbell at Massodon dot social, and yeah, send us toute we like to hear over there. Hey, before I introduced Mark, you know, you were talking about the watermarks there, and you know how we can identify images.
It's very easy to identify. Just look for people shaking hands and the fingers are all fused together, or extra digits creepy stuff, or an extra ear, an extra eye. Yeah, but that software shirt does dive straight into the uncanny valley, isn't it? Like it's not even the uncanny valley, it's oh my god valley. All right, So Mark Miller is here. He's a seven year c sharp Microsoft MVP, probably a ten year by now
right, you gotta update your bio. I was kicked out of the program for a few years when I was on the run from the authorities, but you know, I'm back now. He's also leading expert and user interface design chief architect to the IDE Tools division, to developer Express Any Streams, Live c sharpcoding, and design on Twitch, dot tv, slash code Rushed.
Mark has been creating tools for software developers for four decades. And of course he's the star of that old show we used to do together called Mondays, which you should never let your children or grandparents listen to. Every that's right or anybody with a slightly even civilized nature, you know, really, yes, yes, In fact, the three of us would like to formally apologize for Mondays right now again to everybody who has ever been offended by it.
Again. Yeah, my attorney says, I can't apologize because we got pending lidication about a previous apology that I messed up. Yeah all right, So some Mark, You've had a lot of thoughts about open AI and what they're doing in AI in general and all these large language models. I mean, I'm just gonna open it to the floor. What the hell is going on
here? Well, you know, I'm like I'm sitting around on my butt for about six months this year, you know, listening about chat GPT maybe the only five months I suppose, you know, listening to it, acknowledging what's there and just kind of shaking my head watching what's happening as well with regards to get hub code Pilot. And you know, ultimately we had made a decision actually early this year that dev Express was not going to jump in
this space because we thought it was going to require too many people. It was, you know, kind of a lot of unknowns, high risk kind of place to go. Did you even have any ideas as to what dev express could do to embrace it or jump in the space. I did? I did? I had. There's a feature we have in code Ross called duplicate line and duplicate selection, and basically it duplicates the line and it uses
heuristics to anticipate what you're likely to change. So if you've got a line that says with is equal to size dot with and you duplicate the line, you can get height as equal to size dot he for example. But that
particular feature I thought was ripe for AI exploitation. Yeah, and it was a unique entry point that nobody else was going to step on for a while because it's based on a unique feature inside code Rush, but it doesn't seem like a life changing like do or die I kind of have that feature, you know, it doesn't seem like that, right, I think, you know, to be fair, I think I think there's maybe only one or two features in code Rush that are life changing. Yeah, I gotta have
it, And they're kind of they kind of creep up on you. They're kind of under the hood. They're kind of basic features. They're been are
for like twenty years. But there's a combination of things that are working together that the sum of which I think kind of gets you to that point where you are like, Okay, I can buy into this idea of working with less effort and working with less though it's always been my experience with code Rush, right, It's like it's not any one thing, but when you use all the things, you do so much more in the same amount of time. That's right. Yeah, Yeah, I think that's fair to say.
I know that, you know, I'm I'm using the product live on the Twitch stream, and you know it enables me to write code and talk about it at the same time, Right, So I don't need so much cognitive load to pay attention to the code I'm writing because I'm using shortcuts or features things like that. So you're using a GitHub Copilot and code Rush at the same time. No, I'm not. I actually don't have GitHub Copilot installed. I you know, I you know me. If somebody says don't do
something, we all agree we're not gonna do it. What do I do? Whatever you do, don't do this. In fact, in fact, management at a company we won't mention. But one that I work for has no idea. I'm here telling you about everything that I've done. Okay, they'll find out later and I'll be in trouble, I'm sure. So you So what you're saying is A da Have Express decided not to jump into AI space and b Mark Miller decided to jump into the space rings product at da
Have Express. That's exactly right. No, and they they kind of don't know, they kind of know, but now they don't know. They don't I offered, I say you want to see a demo, and they were like, well busy, and I'm like, okay, no demo for you, all right. I don't see any downside though, right, I mean, this is just making code Rush a better product. My whole point of view on this is, I guess not so much. Oh, I'm gonna
make code rush a better product. What I'm really trying to do is I'm really trying to get really good features to developers and code rushes, like the it's like the conduit the medium through which I can do that. Yeah, it's my vehicle. Yes, you're right, I should have that mentality.
I should like have a product ownership mentality. But I think primarily I'm really thinking about is there something we can do to make developers more productive that's a significant step that's not going to be stepped on by anybody else, you know, that sort of thing. Okay, So so that was the approaches. So let's talk about some of those features. Sure. I included a link in the show's description, and that link is to a video that shows a
little bit of what I'm working on. So if you want to visual and kind of a motion context to see things in movement in motion, take a look at that video and you have a stronger sense of what's going on. But specifically, answer your question, Carl, It first started when I started interacting with open ai on the playground, and I started, you know, asking questions, you know anything wrong with this method? And open Eye comes back, no, this method is fine and it's going to work correctly.
And the method was a method that took a first name and a last name concatenated them together with a space between the two, So if I passed in Mark and Miller, it would come back with Mark Miller with the space between the two the first name and last name. Are you specifically talking about chat GPT here or are you using another interface chat GPT. I'm on the open Aie playground, which is essentially using the chat GPT engine. Okay, guy,
is what's happening? So? And it says nothing's wrong with the method. And I say, well, what happens if I pass in an empty last name? And it says, well, you're going to get back you know, the first name and it actually put in quotes Mark with a space at the end of it in quotes. And I said, the space is a problem. Uh, And it said, oh, I see. And I said, can you generate a test case for me that tests against this particular problem? And it and it popped it out like that, and I
and the test name was well, named. It did in fact test what I what I what I was asking for. And after that experience, I started thinking, Okay, what can we do here that can help people out? And I'll say this my conclusion now I've been working with it for about two months, and my conclusion is is that AI specifically chat GPT is like a wandering child. Yes, okay, it's I couldn't agree more. It's smart enough to give you some really high quality answers, but sometimes it'll kind
of go astray a bit. Yeah, it could either mix stuff up or just could be confident in you know this works and it doesn't. Yeah, oh right. Yeah. We had a massive argument with chat GBT on live on the stream where chat GBC was claiming things were We're real and existed, and we were like, no, they're not. It was like, well, I'm sorry you think that he was very patronizing. I'm sorry you think that, Mark, but I've looked up some of the shows you've done in
the past and I don't think you're qualified at all. Right, I think that's it didn't say that, did it. No, it didn't say that. Last pure comedy right there, that was God, pure comedy. Kids so hard to tell though. There's a lot of anthropomorphization going on with some software here. I know Richard doesn't like the anthropomorphization of chat gbt um, but I think it's funny though. I mean, it's hard to tell what's what's satire here and what's real because chat gbt is really, like he said,
it's a small child. It's like a wandering child. And so part of the challenge in getting from point A to point B, in other words, point a idea for features and point B execution on features that actually really consistently work is learning how to take that small child and corral it, get it to run down the path you wanted to run. And then also,
you know, take a look at what is producing and contour it. There's a number of pieces that are It's kind of like it's kind of like the idea of like looking down on cattle that are moving from one place to another but not seeing any fences. And that's kind of what we're doing, is we're taking the fences and hiding those fences from the users. I considered it like a junior developer that knows some things and sometimes gets things right, but
you have to double check the work. But also speaks very confidently the whole time that all their work is perfect. Oh yeah, absolutely, yeah. And if you think of it as like a rubber duck, now that that kit can be useful for like, let me bounce this problem off you and you might it might come up with one or two suggestions that you hadn't thought of, but you know, it may probably it probably won't come up with the actual code that you need. Um, that's not That has not been
my experience in working with us and in using this. It's been hit or miss. Yeah. But but realize, though, Carl, we're doing we are doing a couple of things that you're probably not doing if you're just in chat GPT alone. Uh. And one of the things that we're doing is we are providing a really strong context to the questions that's hidden. For that context is hidden from the user or it's kind of collapsed. They can expand
it out to see it. For example, I can be inside of visual Studio and I can say, Hey, what do you think about this method? And that's all I have to say, And then now I can get analysis on the method. I can get a description of what it does. Right. I find that when I'm looking at other people's source code or something really big that I don't quite understand yet. Starting from an AI analysis is incredibly productive. It makes me much faster because now I have something to compare
against as I go through and look at the code. The AI analysis might be not one hundred percent correct, as you you know, as you pointed out, but but it now at least gets me started, whereas before I'm just looking at things and I don't quite see them yet. Right. It's like having somebody to bounce it off of. Yeah, and the AI can instantly tell you what's going on in that method. It can instantly come back or or much faster than I can come back and say, look at this,
and look at this, watch for this. This is what I think is happening. This is what the method appears to do, and that I find, yeah, really really useful. So and it's never been like catastrophically incorrect. It was only when I was using the playground it was maybe not catastrophically but absolutely incorrect. It said that there was a c sharp way of
specifying ignore case in the red X instruction strength, but it's not. It's a parameter to creating the new redg X and but it was claiming that it was there, it was conflating it with another language, and uh, you know, we were going back and forth on that to the point that I was like, Okay, this guy is not back and down. But that was in the playground and when I've since put it in the kind of the corral where I've got these railings kind of you know, set up, I
have not seen that. In fact, you know, for me, with the one of the features that I've been actually using on a daily basis that I demonstrated on that video is automatic generation of XML dot comments. To get that, there's a prompt that starts it out that's, you know, setting up all the rails right right, that says, here's what I want, here's what I don't want, that sort of thing. Well, and I really like that it's not asking You're not asking you to write code. You're
asking you to write a good description of this code, right. And also realize that in the case of an XML dot comment, if it's wrong, the part you have to fix is effortless, and the part that it gets right, which are the actual ce refs with the links, which are impossible for a human being to right themselves because they take forever to write, and the exception analysis as well, those things it gets and so you get that, you get the return types. And the other thing is in addition to
that corralling, we give it context. So we check to see, for example, what are the types that are used inside there, what are the method calls inside there? And we give chat GPT more information on those types right right, and those method calls that are going in, so that when it comes back with the description, it has more information then you could give
it then it could have from just the text of the method alone. Yeah, and as you've already described, like this is laborious to do by hand, right, but the bot can knock these parts out because they're and they're very very viable. You can go and chase them down if you need to. Yes, that's that's the thing is that it is. It's an area that where and it looks like nobody else is trying this yet. You know, Copilot doesn't appear to be doing this yet. I think the reason they're
not is because when you first try it, the results are inconsistent. It's the wandering child. Yeah, and so a lot of the effort is about getting the child pointed in the right direction, and then when it comes back, checking to make sure everything is as you ask for and cleaning up anything that comes in, because sometimes chat gpt likes to be a little flowery and maybe add things that do not make sense right, add maybe, for example, a bunch of c refs, you know, at the end, outside
of the parameters for parameter description, outside of the remark descriptions, right outside of anything that's a common thing that chat gpt will do. And so you clean up these kinds pieces that are this extra noise, and what you end up getting is something that's pretty high quality. The other component that was really initiated my reaching out to you guys is that to make this work right, we created a UI that's essentially a representation of the AI working in your code.
So when you say I want to create an XML dot comment for this method, you get a little animated you know, composing right there where the XML dot comment will go, and you're free to continue to edit inside the code. In fact, you're free. You'll see this on the video if
you watch it. You're free to execute this feature many times over right, and so you can actually see in the video multiple instances of the code agent working alongside me inside the same file, And it's a peak into what I think the future of AI code agents can be, where I might have a code agent to some serious analysis of you know, kind of a deep dive analysis of my architecture while I'm having another AI agent generating um uh, generating
test cases, generating XML dot comments, that sort of thing, while I'm
going through and reviewing and learning and looking looking at the code. All of that could be having in a single file if you wanted to wait, remind which you're describing reminds you of like Carl and I working in a Google doc at the same time, right, Like yeah, very sexy too, Like it's thrilling when so that a fact of the agent kicks off and works on that problem while you're continuing on another problem, and another agent kicks off here
and works on this next piece, Like that's pretty cool. It's almost impeded by the whole AI piece. Like this is just an interesting idea that agents that evaluate your code along the way. The fact that it happen to use large language models is a sort of ancillary to the point. So here's here's another thing. To think about is that these things could walk towards just becoming
too intrusive to your code experience. Sometimes I like to just look at some code and think and just stare at the screen without things happening, you know, Whereas if I'm if I haven't completed my thought and I'm writing some code, you know, I don't want some AI in there going oh you should do that. See yeah, I was going to do that, but I'm also thinking about this, like I don't want to have those conversations. I
let you shut up and go away and then let me think. Well, at the moment it moves my cursor in any way, I'm going to kill it right like, or you know, or it starts putting in that comment like, it better be shifting it in the right direction. Those are U acts of facts. Yeah that, but yeah, just a warning. It's the it's And I would argue code Rush is brilliant at this. Yeah, don't interrupt me, Like I've never seen a product better at not interrupting me
than code YEA. Like, generally speaking, when it takes an action, you're delighted by it. You can look around and look for those contrast clues and those h you know, icons and colors and things to and interact with them if you want. But are you right it doesn't, So thanks for
that, Mark. I just want to you know, respond to the implied concern and the you know, the answer is is that already with the current you know feature, the XML dot coment generator, there's no shifting of the view when the insertion occurs, without moving the carrot, without changing selection. You are completely free and fully powerful to do editing and invoke other features including intelisense and telecode whatever you want during that time. So we've kind of already
got that. Also with regards to the don't distract me, you know, we hear you loud and clear on that, Carl, and that is not it's only These features are only invoked if you ask for them, and to ask for them you have to specifically do it. For example, if I want to enable voice commands, I have to opt in by specifying Azure Cognitive Services Speech API key, and if I put that in there, I can then choose options like if I hold the control key down, it'll listen.
So I can hold the control key down kind of like a microphone on button, and I can talk leaning the microphone talk and then release the key. So is that working right now? Is that something? Yes? Yeah, yeah, it's working right now. In fact, I've got it on my machine control. Left control key speaks to the agent, right control key executes voice commands. Okay, is that a code rush thing or is that just in visual studio code rush thing and it's an AI thing we're actually using.
You have the ability to specify your voice command, like I have a voice command I added today called clear chat, and I have a checkbox under that that says allow semantic equivalents, and it's checked by default, so I can say, hey, deck's clear the chat while I'm holding down the right code control key. I no longer have to remember the exact phrasing and say it for voice commands to work, right and it as it's code rush, there's
context that are bound to this. So some commands will only work if you're, for example, debugging, and it makes sense for the debugging commands to work right. Yeah, that makes sense. Now, hang on, that's my favorite way way way way. Wait, well, you you're just sort of tossing out the voice control part because for the most part, we haven't seen a lot of voice and chat GBT. I know there's a few folks
tinkering with it. But and you, I don't think you've used voice in code rush before either, So we just casually saying this, but like you're
introducing and you use your interface. Yeah. Yeah. The reason we've never done voice commands before is because the interface sucks horribly, right, there's there's just no way to solve the problem if you don't have if you don't have AI and or you don't HIT and also you don't have like a back end like as your cognitive services helping out right, Okay, a cognitive services is fast and nicely accurate, really loving it. Um uh and for speech recognition,
for speech recognition exactly. Writing the feature, the voice command feature took like two days. Maybe it was not hard to build, right. We take in the voice in information, we send it out to open AI with a specific corralled messaging including uh, well not this is this is to be done what I'm about to say, but including symbols from that are in scope because I also want to add voice dictation of comments. So this is this whisper you're talking about on an Open AI. No, No, I'm not
using whisper. I mean, I think whisper. I'm only barely familiar with whisper. I think it's I think it's speech recognition, Isn't it? Is that true? Or no? Or just speech to text speech to text? Yeah, no, I'm using as your cognitive services speech to desks. Okay. The last time I used cognitive services, the technology they were using was LUIS the language UI service or something like that, and you had to do quite a bit of work in order to get it to figure out what the
operative words are and recognize them within some sort of context. I'm assuming that has changed, because there they sort of shut that down, yep, and now that now they have some AI behind it, I guess. So I'm not familiar with what's changed, but I will say that we are able to add our own phrases to recognize and that's where the symbols come in for the purposes of dictation, right, dictating comments, things like that, which I'm
still interested in implementing. But but yeah, are My experience with it is that it's like well over like nine accurate without any tweaking to the models or whatever, right, without tweaking, And if I tweak it, I can then basically it gets name is wrong. Yeah, like essentially is what it
gets. More most like, the our AI agent is called Decks. And if I say hey, Dex, and I don't give it any context, it might say headaches, or it might say it might say, hey, comma d E c K S instead of d e X. So what I do is I say, well, there's a character named Dex. I give it freezeology, I give it things like c sharp, I give it things that we're going to talk about it coded. So you give it some proper names in say that these are some of the names that you can recognize.
That's that's an old trick. You know that we've been doing that in speech recognition forever. But the key is the key that makes this whole thing work, Carl, is that having ninety five percent accuracy right is all I need when I'm talking to chat GPT. If I'm talking to the agent and I get it ninety five percent correct right, and I might have like a flub in there or something like that, it doesn't matter because chat GPT understands what
my intent is and gives me the answer one the question. I love it. Hey, let's pause here for just a brief message or two and we'll be right back. Hey, we're back. It's dotting It and Rocks. I'm Carl Franklin. That's Richie Campbell over there. Hey, and he's not responsible for the wildfires, we get that straight. And that's Mark Miller. Of course. He's talking about all these cool AI features that he's put in code Rush, and he's had a lot of thoughts about AA in general,
and we were just talking about this, the speech recognition technology. Again, this is a feature that's in code Rush. This isn't a visual studio kind of thing. The visual studio doesn't have anything like this, does it. No, I'm not not built in. There may be plugins that you can get, but you know, one of the one of there's there's a couple of commands that I've added, and I haven't added a whole bunch because I'm still very fast with fingers. Yeah. But one of the commands that I
added is new folder, which I really like. Nice. I like that, and it's semantically It allows semantic very so I can say, you know, let's create a new folder. Now. Nice. You know, I want a new folder. Give me a new folder. Give me folder, new folder. You know folder that is new and then it highlights it so you can just type the name, right, yeah, it highlights it, just do type in the name. Also, you know show watch window number
one. Yeah, you know, something along those lines, right, those kinds of pieces I've built commands for because I don't like reaching for the mouse going into the you know, debug Windows watch and then watch one that's like four menus. Yeah, of course, you know. It's the worst is like the package manager window. If it's not already at the bottom of the of the screen, it's hidden. You got to go to Windows other windows and then find the package man doesn't It doesn't even have a keyboard shortcut.
So really obscure things like those things might be interesting for us to ship with commands voice commands out of the box. Um, but voice commands, I think it's you know, even though I'm leaning in this direction and I'm starting to use them, I still think that for the vast amount of coding that I'm doing, it's gonna be keyboard driven. Yeah, I'm gonna be you know, coding a navigation still off the keyboard, and we're back to the
same old thing of taking your hand off the keyboard. You get to the mouse is dumb, yeah, right, like it slows you down. The idea of voice as that additional interface, like this is the problem with Visual Studio as a whole. Right, the whole cockpit of the seven forty seven is stuffed in those menu bars like it's all in there. You just can't freaking find it at the time. Windows in general, the idea that you could just say I need to do acts or do this for me and it
does the navigation give me more of that? Yeah, yeah, can you write a plug in for Adobe Premiere now once you're done with that would be so great? Yeah, I know, you know, speaking of which, are there any folks from Adobe? Okay, guys, we've got to fix your UI. And I am an avid user of a number of your products, and you get me on the phone and I'm gonna get you. We'll do a call and I'm gonna get you. We're gonna fix a number of problems spaces in your UI. They're easy fix. I know, Carl Richard
are both like, Mark, what are you doing? What are you talking about? This feels like Mondays again. Right, We're gonna we're gonna need the lawyers again and we're gonna there's gonna be a series of poorly made apologies that will require more. Boy, I've seen this playbook, like this is not the first time down this rodeo. Yeah. Actually to that point, Mark, like all of this voice interface stuff, whether it's spoken or written, seems to be an alternative to the gooey. Like, you know what's
cheaper than trying to fix that flipping gooey. Provide a new interface so you don't need it. Yeah, that's interesting. Well, you also get in this other problem, which is you have an entrenched customer base that knows where their stuff is that gets really angry with you every time you move the cheese, right, like, anytime you change anything on the interface, and yet
it's utterly unapproachable by new users, right right. So you create a new interface that a new user, inexperience user, or somebody trying to use a new feature can use, and don't move the cheese for the incumbent. Yeah, yeah, don't take the menus away just to provide that what you're doing, which is a nice yeah voice user interface. You know, there are like something there's something like three thousand commands. I believe that ship with visual
studio. Wow, that's my understanding of the total count, and you would know better than almost anybody in this planet. Yeah, I don't. I haven't seen that. I haven't counted them up. But I've been told this as I recall, and this is what I'm retaining in my head. I'm pretty sure this is correct. So, um, code rush is free, but AI isn't free. So how are you gonna how are you gonna do
this? So you're gonna turn those features on with a credit card or something or no, we're gonna turn the features on with a kind of You'll have to create your own open AI key, okay, and answer that key inside the options folder or inside of your environment settings. But we won't be able to speak it because it's not yet, won't be hocked up the right we pay. I got a text the poor problem. We spent weeks actually developing that feature. That was our first thing, and then we realized, don't
don't. No fireplace doesn't go there, it goes over there. Yeah, Deuxpress managers are like and this is why you don't develop features. Mark, I'm sorry, Okay, I'm gonna think ahead next time. So his speech seems like just the beginning for you? Are you? Is your goal to go after Copilot X? I mean that's nuts. Not so not to go after them, but to basically live in a compatible, harmonious world as long as it takes for them to start copying our features. Okay, you know
what that's that's great. Well, how do you describe Copilot X different from
Copilot first? Well, so I'll say this, I'm not I can't answer that question because I have not installed either of the products, in part because I want to develop kind of in a dirty white room, I guess or whatever, right, not quite you know, not quite white, fully white, because I'm aware of the product and I've actually gone through and see, you know, looked at its feature set, looked at what they were advertising, because I wanted to make sure I wasn't going to, you know,
step on it or collide with it. But I'll tell you, Richard, I think that I think that they're that with the information that we have inside of code Rush and also our skills that we've already built in terms of generating code, you know, for what's been over twenty years of development of this product, I think that we have the ability, with fewer people to still do innovative things in a way that's not you know, that's not redundant.
The other piece is is that you know, you may be in a it may appeal to you to use code Rush instead of Copilot because you have control of the direct pricing. You pay only for what you use, as opposed to I think Copilot is on a monthly subscription. Yeah, fixed rate per month, right. I mean A big thing for me with X is they've gotten much more into the workflow stuff generating you're you know, parsing the code
changes you've made and making a pull request from that. Very cool. Um, the automated testing parts like those are a lot where Copilot was much more asking for code you know that kind of thing. This is getting a little more integrated into your workflow. Yeah. I think that's the direction you know, where where everybody where code agents are heading. I think that the more that the code agents can understand, the more helpful it can be, right,
the more helpful information that it can provide. Yeah. And the other part, of course is GPT four. You know, I know we're in a Gartner hype cycle and we're like right at the top of the peak of of unreasonable expectations. But it does feel like we're coming down the other side of it because folks are starting to talk about just how expensive GPD four is. Well, okay, that's a reality. Yeah. I don't like to
play in that space. I mean, if you want to talk about cost of AI because of all the machines and stuff running like that, Yeah, that's gonna I'm just gonna be sad if we don't. If we go there, what I'm but but I'll say this about GPT four. The four oh is kind of like taking your you know, nine year old child and turning him into like, you know, a twenty something year old child, you
know, now moving around a little bit. The the the rational thought, the understanding, the explanations are at a whole other level from my perspective in terms of what I've seen making clear it has no rational thought. It is a prompt response engine, still just a very large one. I'm an illusion, Richard, and you're you're you know, diss in my vibe. Man,
I'm just deeply concerned with people personifying a piece of software. Yeah, old chat GPT is going to be upset when he hears who said that about him? Exactly what I mean right it, But yeah, they the school. You know, we're already starting to see the issues around how many customers are going to need, like just what is it going to take to make this thing pay for itself. That's a huge chunk of Azure processing all the time for every request for you know, this titanically large model. Is the
value sufficient? Yeah? No, I think that. I think that's a good realistic question as you're moving forward, right, Um, you know something else on that, But I think I kind of lost it. I'm just like, I guess I'll live right, We'll leave it there. Well, because we got some pretty good results from three and three is a lot more practical from a science perspective. Yeah, but it was also testing the waters
and that's why it was free. You know, it's limited, it'll it'll forget things after a while and it will start, yeah, going insane. And that was you know the reality is And this is what we got from Jody Burchell too. Yeah. I mean is they got to a point where they'd written as many problems as they could think of, so they basically turned into public to write more, right, you know, like we're just the test rabbits. It's like, hey, throw some stuff at it we haven't
thought of. Turns out we wrote existential questions for two months, which is bizarre. It's like, you know, get a dog, for God's sake. I have a subscription to Chat GPT four and I'm not I'm not two worked up about it. I think it's worth it for me. Yeah, But I'm also really aware of just how I mean. They built one of the largest supercurities in the world to create Chat, to create GPT three, which means almost certainly they did make the largest supercomputer in the world to make
for and and now they're trying to operate it like that. It's a good thing NFTs tanked when they did, because we needed all that compute. Yes, dearly beloved, we got it here. Yeah, that's that's say goodbye to the NFT. Yeah yeah. I think what happens is you get to a point where you're like, oh, it costs money for a little more intelligence, yeah, right, and do we really need it? Do I
want to pay it? What level? And you know, one of the things that's kind of cool about open AI is that you can choose which of the engines do you want to use? Yes? Right, And so you can go for the cheaper, faster models if that's a particular concern for you, right, and especially if they're sufficient, if you're not relying on those pieces. But here's the other side of it. You know, we've all
built enough software over the years. You know, when you're still exploring a particular domain space, you just keep going bigger until you get your head around it, and then you tune, you sort of settle it. And this still feels like they don't know what the destination is, so they're just building bigger. But it's to me seems very likely that at some point you go, you know, that's bigger than it needs to be. Let's dial that back. Yeah. I think what happens is we move forward into the future
and we kind of maybe push up against that. But what's also happening is they're making the AI calculations cost less as well, through hardware innovations and other kinds of factors that are going to I think drive that cost down. So I think what happens is we may kind of slush up against that, and there may be folks who need that higher end LLM, right, that really really intelligent analysis, and there may be a lot of folks who don't need
that at all. Right, Yeah, And I think you're exactly right. It's like that might be the edge of exploration, but the engineering part, it's practical and profitable, there's a couple of steps behind it, right, Like most engineers look at the eighty percent case because it's reliable. There's the
other problem that edge case. There's another problem that people just don't know what they need in terms of the eighty percent of the twenty percent, right, and so they'll just throw money at the big one, you know, without knowing that they need it or don't. But I kind a story that kind of illustrates that. I have a friend who has a product and they sell
mostly on their own website, right, and their business is tanking. And you know, I did a little just a conversation with her, the owner, and her husband who sort of does it stuff but he's you know, we're his head, and they decided, we all decided they needed to be on Amazon. They needed to have their products on Amazon. I said, okay, so what's keeping you from being on Amazon? And her husband says, oh my god, I just you know, every time I go to
figure that out, it's just it's a nightmare. He's completely over his head. He doesn't know what he doesn't know and everything. He's tried it a couple of times and things have been screwed up. They've hired some people, it's been expensive, and they've screwed up. I said, all right, here's what we're gonna do. We got on a zoom call, got out chat GPT, and I said, type your question, your specific question about Amazon into chat GPT and lo and behold, there's an answer. Oh and
yeah, but it doesn't in these complaints. Says, yeah, but it doesn't say a blah blah blah. And I said, ask the question, right, and there's another answer, and and so then he's off, you know, trying to apply this to the stuff he's already gotten. He said, Okay, maybe I need to do this, and maybe it did,
but I don't as it. I'm like, ask the question. So it just came back down to them not realizing that they had the tool that would help them get the information that they need just by asking simple questions, and they're reticent to ask. This guy was reticent to ask the questions because he thought that it didn't have an answer right. And it turns out that they got it all worked out, like within a couple of hours just by asking
questions. Could they have done that with a search engine if they had asked the questions of the search engine. Well, obviously he had been on the search engines and he had been he's over his head and it still reticent, right, absolutely actual an issue with his reticent absolutely right. But there's there's also noise, right that there's noise in the search results sometimes, and there's less noise if you ask a direct question to GPT of how can I do
this? Or what's your recommendation? Right? Yeah, how should I proceed? Because they haven't they haven't steeped open AI in ads yet because they're still having worried about a revenue model, where on search engines they are steeped. I'm just telling what this experience was like for this guy. It was life altering because it took away the anxiety of not knowing exactly you know, what the answer to a particular question was before you could move on and knowing how
to do things the right way. Yeah, my kids had a similar experience at school. At school, they wanted to do their own project based learning elements and it was outside the curriculum, and so one of the administrators at school said, well, give me a proposal, and it kind of felt like a blocking demand right right out, a proposal of what you want to do. So he said, allright, kids, we're gonna learn about chat
GPT today and we're gonna have chat GPT right the proposal. And whenever you have somebody blocking you that's asking for essentially words in some form, right, you can use chat gp to unblock. And so let's start that. And so they started by with the prompt that says, I want to study animatronics and puppet building, and I want to combine the two with actually open an eye. This is Campbell who wants to do this. I saw the Facebook
post. Yeah, it's really cute. Yeah, and he wants to combine open AI so we can create a puppet that talks to people using synthetic voice and voice recognition, speech to text, and open AI and a puppet that he makes. So he wants to do all of those things. And he said, you know, build that proposal for me, and it writes out the proposal pretty much instantly, right, and then we look at it.
We say, okay, how do we want to edit it. Well, let's go back to the prompt make a few changes, or when it's good enough, we just go in and make those changes by hand. And with about fifteen twenty minutes worth of work, we have, you know, a page and a half of text a proposal. We send it back to the administrator and now the ball is in their court instead of it, you know,
being in you know, at a thirteen year olds court. And then you know, and they maybe let it drop because they just don't know how to proceed where to go. And I think this is one of the things that tool is phenomenal at, is getting past blank screen syndrome. Yeah right, whether us block for code, for that business proposal, for an essay, like any of those things. Just to again, it's very much a rubber Duck effect, Robert Ducking describe the problem, which helps you think more
clearly about it. And then the fact that you get some fairly creative words back that that give you a place because it's easier to edit than it is to write. Yeah, yeah, and I'll but then you're still engaged with the material, Like if you're dumb enough to just send out what it made, you're you know, going to reap the consequences. And there's a few lawyers dealing with that right now, where you know, used chat GBT to generate a summary on a case that cited a bunch of cases that did not
exist and they didn't check them at all that but the judge did. Yeah, it's really happened. Yeah, wow. Yeah, they're gonna lose. They're gonna lose their law licenses because they misused a tool mark. When are we going to hit the singularity? Um? I think it's about two years and two months about from right now. Hang on a second, how would you define the singularity? So for me to say, okay, fine, well, I wasn't gonna say that part because now it's gonna mess up my
prediction. Maybe I was going to define it, you know, and retrospect as well with the day I brought brownies out of the oven. But now that you've put me on the line, I'm gonna I'm gonna have to say
no. For me. The singularity is that we have human level intelligence close enough human level intelligence, and it's relatively cost effective, so it's not it can keep itself running and then it can create improved versions of itself, right, it's the moment it has the ability to create improve versions of itself. We're essentially there that's ushering in our eventual robot overlords. Is not. Yeah,
that's that's essentially that's that's where we are only in science fiction. But okay, no, I'm no all right, Richard the Voice of Reason please no, no, no, wait, the Voice of Unreason still trying. I want to get Richard after you though. Yeah, no, I'm like, I'm at a point where, uh, you know, I'm We're already seeing stories about people who are being fired from their jobs because they're being replaced
with chat GPT. They're hired to uh to now train chat gpt to replace them, right, so that all of this is ridiculous, like they're good news stories that sell cling. CHATCHYPT doesn't replace your job. Somebody using chat GPT replaces you. Well, that's that's step one. Step one is that turns out that's the only step. It's not. It's not the only step,
because what happens is is we approach the singularity. The ability to manage other bots can be distributed to a specialized bot managers, right, and that distribution point now leaves fewer people hired, fewer people making money. You're talking way beyond chat GPT though, I think that's where Richard is talking about. Yeah, well this is beyond this is beyond m but this is science fiction. It's not made me in two years, Richard. Yeah, okay,
all right, Richard, the voice of the reason go. I mean, the original singularity is defined by John von Newman, the guy who basically came up with the concept of the modern computer that we've been building on ever since. So I kind of count on him. And you know, he was describing it with Stanis Laws back in the nineteen fifties. But the real point was when does technology transform civilization to the point where you can't predict with civilizations?
Look like he didn't say anything more than that, you're pretty close to that. I would say, we're well past it. Yeah, you could not describe to someone in nineteen hundred two thousand and have it make any sense to them. Right. So all of the hyperbole that's been added to that is purely for marketing purposes, right. But civilization has already been transformed by
its technology otterly. And that's and yet here we are. It's only when you want to sell more widgets or collect more clicks that you turned into an apocalyptic concept. I saw this great meme on Facebook or one of it now. It was on the Mastodon and there's this. The caption said, I would like to go back to nineteen ninety five and explain this picture. Does
somebody appear? Right? And the picture was a woman in a mask, right, a COVID mask holding a phone an iPhone profile so you don't really see what it is, and the other hand an ice cream cone, right, So just think about that from it before cell phones? Why are you wearing a mask? What is that thing you're holding? And what are you doing to that ice cream cone? Well? What are you going to do with the ex room cone because you're wearing a mask? Yeah? Right,
It's just very confusing. But well, anyway, I don't know. Is that a good place to leave it? Mark? Uh? Yeah, yeah, yeah, I think I think it's good except for Richard's wrong. It's all gonna be over soon, it cut. Oh man. I miss you so much. Mark. Thank you so much for sharing your thoughts with us, and I'm sure everybody else does too well and more importantly, building cool
stuff as usual. Yeah, like what I like about you, Mark is you only show up when you've made something cool that we need to learn about. And I really appreciate that, you know, and I didn't ask you. Where can I get this new dev Express voicemail voice recognition thing. Well, it's so as of today, not out yet. I'm I'm pushing to get it out in about thirty days, okay from now, so that's July
ninth round roughly there. Uh. And you just go to dev express dot com, forward, slash code rush h and also, you know, come see me live at twitch dot tv slash Code Rushed with an eed at the end. Do you have a private repo for your friends? No, don dude, Now, I have so much trouble with the Express when they find out what I've done. I just want to mess around voice in visual studio now. Yeah, I'm looking forward to this. All right. With that,
we'll talk to you next week. Thanks for listening, and we'll see you next week. I'm dot net Rocks. Dot net Rocks is brought to you by Franklin's Net and produced by Pop Studios, a full service audio, video and post production facility located physically in New London, Connecticut, and of course in the cloud Online at PWOP dot com. Visit our website at dt N et R o c ks dot com for RSS feeds, downloads, mobile apps, comments, and access to the full archives going back to show number
one, recorded in September two thousand and two. And make sure you check out our sponsors. They keep us in business. Now go write some code. See you next time. You got a Jadmtlevan