WEBVTT

00:00:00.342 --> 00:00:00.882
<v Matt>Hey, Nina.

00:00:01.497 --> 00:00:02.135
<v Nina>Hey, Matt.

00:00:02.752 --> 00:00:08.091
<v Matt>How are you doing? I'm good. This is our first one. So congratulations to us for starting.

00:00:08.457 --> 00:00:12.650
<v Nina>Yeah. Yeah, this is exciting. Should we do a little intro or?

00:00:12.993 --> 00:00:16.297
<v Matt>Yeah, I guess so. People probably want to know who the hell we are at some point.

00:00:16.726 --> 00:00:21.181
<v Nina>Right. Yeah. So, yeah, my name is Nina. Software engineer.

00:00:21.902 --> 00:00:25.228
<v Nina>Did illustration and music before getting into software eight years ago.

00:00:26.080 --> 00:00:30.715
<v Nina>I worked for a privacy company, React Shop throughout my career,

00:00:31.446 --> 00:00:34.817
<v Nina>a lot of Node.js development, a little bit of AWS.

00:00:35.557 --> 00:00:37.647
<v Nina>And in my personal time, I do

00:00:37.647 --> 00:00:42.317
<v Nina>code a lot with Claude for building desktop tools to improve my workflow.

00:00:42.637 --> 00:00:47.837
<v Nina>So mostly for my creative workflow when it comes to music editing and yeah,

00:00:47.837 --> 00:00:49.877
<v Nina>just a lot of script automation as well.

00:00:50.487 --> 00:00:54.225
<v Nina>That's kind of my focus right now with AI-driven development.

00:00:55.000 --> 00:00:56.357
<v Matt>And met

00:00:59.157 --> 00:01:01.895
<v Matt>we used to work together Nina and I, so that's why we started this podcast

00:01:03.100 --> 00:01:09.035
<v Matt>I've been dev for too long, maybe 20 years now, and I've been doing things from

00:01:09.375 --> 00:01:12.046
<v Matt>large enterprises to some all-startups.

00:01:12.955 --> 00:01:16.765
<v Matt>I've been running now independent for about four or five years now,

00:01:16.765 --> 00:01:23.515
<v Matt>doing client work. And so I've been working in Java, Python, Node, a lot of stuff.

00:01:23.815 --> 00:01:28.837
<v Matt>And it's all changing now with AI, so it doesn't matter which language I'm using anymore.

00:01:30.283 --> 00:01:34.585
<v Nina>Yeah, right? It doesn't really matter because, I mean, you can learn and pick

00:01:34.585 --> 00:01:38.582
<v Nina>up anything in an instant at this point.

00:01:38.866 --> 00:01:43.927
<v Nina>Like, I had flirted with using Electron for desktop apps.

00:01:44.050 --> 00:01:49.535
<v Nina>And, you know, Claude has made it so much easier to, like, develop a plan right

00:01:49.535 --> 00:01:54.637
<v Nina>off the bat and do research that it kind of feels like we could code in anything.

00:01:55.263 --> 00:02:01.466
<v Matt>It doesn't matter. Yeah. I've tried Java. I've tried Python. It's the same.

00:02:02.046 --> 00:02:06.295
<v Nina>Yeah. Like, nothing is intimidating anymore, which is, you know,

00:02:06.635 --> 00:02:11.526
<v Nina>or you don't have to invest, like, a hundred hours in something.

00:02:13.351 --> 00:02:18.350
<v Nina>But especially when it comes to like the research aspect of getting something started, I don't know.

00:02:19.301 --> 00:02:27.490
<v Nina>Claude has been really great for that. So like recently I made an Electron application

00:02:27.790 --> 00:02:31.803
<v Nina>with Vue3 and TypeScript that I've wanted to make for a while.

00:02:32.261 --> 00:02:37.290
<v Nina>And it was kind of a daunting task because all the features I think would take

00:02:37.290 --> 00:02:40.968
<v Nina>normally a huge dev team to complete.

00:02:41.496 --> 00:02:46.830
<v Nina>It's like a tag management you tag your audio files it's like a daw system for

00:02:46.830 --> 00:02:51.836
<v Nina>like music engineers streamers you could make sound boards and it's just for,

00:02:52.471 --> 00:03:00.690
<v Nina>organizing like huge music libraries on your local machine and it has like outside

00:03:00.690 --> 00:03:05.460
<v Nina>of like the tagging it has like stem separation with uh like what.

00:03:05.460 --> 00:03:06.058
<v Matt>What's that mean

00:03:06.673 --> 00:03:12.120
<v Nina>It means so stem separation is where you have a an audio track that's already

00:03:12.120 --> 00:03:13.824
<v Nina>finalized, mastered, and mixed.

00:03:14.312 --> 00:03:19.304
<v Nina>But let's say you want to hear the guitars or you want to hear the vocals isolated.

00:03:19.914 --> 00:03:24.840
<v Nina>I installed a machine learning model that I researched all of them with Claude,

00:03:24.840 --> 00:03:26.635
<v Nina>like which ones are free, open source.

00:03:27.117 --> 00:03:35.435
<v Nina>And Demucs separates the audio files. So I was able to separate like vocals, guitars, drums.

00:03:35.874 --> 00:03:40.993
<v Nina>And that's really great if you're somebody who likes to mix and remix tracks,

00:03:41.650 --> 00:03:46.430
<v Nina>so i guess it could be good for djs too but yeah it was just a little bit of

00:03:46.430 --> 00:03:50.050
<v Nina>python on the back end for separating out those scripts,

00:03:51.040 --> 00:03:52.990
<v Nina>and i know this yeah go ahead.

00:03:53.140 --> 00:03:58.110
<v Matt>So this is just a tool that existed before are you using like claude to verify like the outputs yeah

00:03:58.822 --> 00:04:01.201
<v Matt>Are you verifying that?

00:04:01.661 --> 00:04:10.447
<v Nina>I verify the outputs. So it's something, it's open source, and I installed it on the node backend,

00:04:11.056 --> 00:04:17.341
<v Nina>and I already had streaming and buffering set up for all the audio files.

00:04:17.341 --> 00:04:21.191
<v Nina>So it makes a request on the front side, and then on the backend,

00:04:21.191 --> 00:04:25.025
<v Nina>it does the separation using the model. And the model has its own.

00:04:28.442 --> 00:04:32.991
<v Nina>Internal functions. It's like a model slash library kind of that you can use

00:04:32.991 --> 00:04:36.881
<v Nina>and it splits them apart for you and then just stream and buffer them to the

00:04:36.881 --> 00:04:41.495
<v Nina>front end and it separates audio files and they're right on your operating system.

00:04:42.041 --> 00:04:48.631
<v Nina>So the application is also kind of like it's kind of like a file explorer for

00:04:48.631 --> 00:04:52.023
<v Nina>your operating system but just for audio files. So,

00:04:53.277 --> 00:05:00.614
<v Nina>One of the things I learned with it was, okay, how do I search audio files on

00:05:00.614 --> 00:05:03.077
<v Nina>my computer extremely fast?

00:05:03.600 --> 00:05:07.884
<v Nina>And I researched with Claude, let's come up with the most efficient algorithm

00:05:07.884 --> 00:05:13.430
<v Nina>in terms of latency and how fast it could search, because that's what matters to the users.

00:05:14.018 --> 00:05:17.837
<v Nina>And so it helped me implement a parallel recursive walk,

00:05:18.540 --> 00:05:24.944
<v Nina>and it recursively goes through all the sub directories it's and you can actually

00:05:24.944 --> 00:05:28.136
<v Nina>set it up to search in batches with multiple workers,

00:05:28.975 --> 00:05:33.044
<v Nina>and in like the advanced settings i set it up where you could customize the

00:05:33.044 --> 00:05:38.756
<v Nina>workers so it can search through my entire downloads folder,

00:05:39.535 --> 00:05:45.404
<v Nina>which is extremely messy i think there's like 4 000 files in there i can't remember

00:05:45.404 --> 00:05:50.845
<v Nina>the like the gigabytes in there but And it can search through that in like a couple seconds.

00:05:51.395 --> 00:05:54.134
<v Matt>What do you like even search for? Like cool guitar riff?

00:05:55.291 --> 00:06:00.004
<v Nina>Well, first you have to point to the directory and it'll go through all of the

00:06:00.004 --> 00:06:00.954
<v Nina>files in that directory.

00:06:00.954 --> 00:06:04.714
<v Nina>So you could technically point to your root directory and it could search your

00:06:04.714 --> 00:06:08.488
<v Nina>whole computer and it'll find every audio file,

00:06:09.052 --> 00:06:14.674
<v Nina>on your computer, excluding like private audio files used by applications,

00:06:14.674 --> 00:06:17.374
<v Nina>like, you know, like sound effects for, I don't know.

00:06:17.650 --> 00:06:23.250
<v Nina>Like your pages app or when you get like a message notification, it'll omit those.

00:06:23.571 --> 00:06:28.904
<v Matt>So is it just searching? Is it an audio file at this point? Or is it like looking

00:06:28.904 --> 00:06:32.864
<v Matt>into the audio file to be like, Oh, this is a jazz song. And like,

00:06:32.864 --> 00:06:35.047
<v Matt>this is a pop song or something.

00:06:35.661 --> 00:06:36.514
<v Nina>Oh no. So,

00:06:38.563 --> 00:06:43.070
<v Nina>When it comes to classifying the files, like with tags, in terms of what those

00:06:43.070 --> 00:06:45.036
<v Nina>files are, that's up to you.

00:06:45.651 --> 00:06:50.880
<v Nina>There is a search and filter, though, where you can search by file type a string

00:06:50.880 --> 00:06:55.892
<v Nina>in there, and then once it's a string in the file name, and then once you add a tag,

00:06:56.760 --> 00:07:00.880
<v Nina>to the file, like you could tag something that's like, download 200.

00:07:01.360 --> 00:07:07.152
<v Nina>You could tag it, guitar riff. And you could organize those tags and then filter by those tags.

00:07:07.986 --> 00:07:13.097
<v Nina>So it's for people who have a very, very messy, messy computer,

00:07:13.097 --> 00:07:18.505
<v Nina>and they didn't digitally archive all their sound bits.

00:07:19.268 --> 00:07:22.520
<v Matt>So at some point, are you going to try to get Claude to do the tagging for you?

00:07:22.700 --> 00:07:26.791
<v Nina>Yes, yes. So that's an in-progress feature.

00:07:28.033 --> 00:07:34.433
<v Nina>I did add Claude. I have added Claude to other apps where I stream the output,

00:07:34.955 --> 00:07:38.350
<v Nina>in the app for like a document updater.

00:07:39.091 --> 00:07:44.977
<v Nina>And it was relatively easy. And I was able to hook up multiple models from Hugging

00:07:44.977 --> 00:07:49.520
<v Nina>Face, whichever ones don't affect my MacBook Air.

00:07:50.047 --> 00:07:52.108
<v Nina>You have to be very careful with what I install.

00:07:52.775 --> 00:08:00.335
<v Nina>But yeah, it was a huge learning process. And looking at the podcast that,

00:08:00.550 --> 00:08:02.425
<v Nina>you know, you were sending to me,

00:08:03.070 --> 00:08:08.637
<v Nina>what Claude has made the easiest is we're kind of almost like our own product

00:08:08.637 --> 00:08:09.855
<v Nina>engineers at this point.

00:08:10.070 --> 00:08:16.177
<v Nina>Like, we just have to focus on end results and then verify the inner workings

00:08:16.597 --> 00:08:18.108
<v Nina>as they're being implemented.

00:08:18.451 --> 00:08:23.547
<v Nina>Like, as long as we research things and make sure things are sustainable and

00:08:23.547 --> 00:08:26.031
<v Nina>scalable, like, anything is possible.

00:08:26.620 --> 00:08:29.957
<v Matt>You're talking about that panel podcast that I mentioned? Yeah.

00:08:30.826 --> 00:08:34.111
<v Matt>Jordan Gall is just talking about not...

00:08:34.490 --> 00:08:39.220
<v Matt>Do you even need developers? But you do, because they can't do it.

00:08:39.655 --> 00:08:43.510
<v Matt>But also, you don't want a full-time developer doing this work.

00:08:44.497 --> 00:08:46.817
<v Matt>It's a weird in-between

00:08:49.817 --> 00:08:53.700
<v Matt>of where you hire a developer, where you hire a SaaS.

00:08:54.945 --> 00:09:00.695
<v Matt>Like this could be a SaaS I feel like the Forge Audio thing like there's enough people wanting that

00:09:01.471 --> 00:09:02.812
<v Nina>Yeah, I think so.

00:09:03.485 --> 00:09:08.346
<v Matt>Like who would be type of customer, do you think? Would it be like a DJ or?

00:09:08.944 --> 00:09:18.446
<v Nina>Yeah, I think DJs, Twitch streamers, people who use definitely like game streamers

00:09:18.446 --> 00:09:21.246
<v Nina>because there is a soundboard aspect.

00:09:21.246 --> 00:09:28.146
<v Nina>And a lot of Twitch streamers, you have to pay to make soundboards and stuff

00:09:28.146 --> 00:09:32.221
<v Nina>on Discord and a lot of tools that they use.

00:09:32.400 --> 00:09:37.133
<v Nina>So, I mean, this might be a nice open source kind of option for them.

00:09:37.633 --> 00:09:42.144
<v Nina>And audio engineers. I know it would definitely help audio engineers because

00:09:42.498 --> 00:09:47.116
<v Nina>when they're in the studio and they're mixing bands, they have all these tracks.

00:09:47.116 --> 00:09:50.488
<v Nina>And I can see them working and I know they're not naming the files.

00:09:52.326 --> 00:09:55.886
<v Matt>They just like kind of make it and then throw it away and like not even try

00:09:55.886 --> 00:09:59.065
<v Matt>to organize the stuff and just make it new next time and things like that

00:09:59.332 --> 00:10:03.096
<v Nina>Yeah they'll have like a they'll have a huge directory for all the different

00:10:03.096 --> 00:10:08.047
<v Nina>takes right so let's say like like somebody's recording guitar one,

00:10:08.628 --> 00:10:13.166
<v Nina>they'll have like probably a hundred takes of guitar one and they'll have to

00:10:13.546 --> 00:10:17.097
<v Nina>write down and remember which was the good take before they mix it in,

00:10:17.787 --> 00:10:21.986
<v Nina>for anyone on the podcast i i used to do music and go into the studio and see

00:10:21.986 --> 00:10:26.706
<v Nina>audio engineers work and a lot of the times they'd be like I can't find the

00:10:26.706 --> 00:10:29.081
<v Nina>take I can't sorry I can't find,

00:10:29.758 --> 00:10:36.456
<v Nina>the one from this day so if they had Forge Audio they could be editing write an app and.

00:10:36.456 --> 00:10:38.395
<v Matt>It's now ready selling oh

00:10:40.475 --> 00:10:40.991
<v Matt>sweet

00:10:41.454 --> 00:10:42.725
<v Nina>Yeah I don't know how to market though.

00:10:44.203 --> 00:10:46.780
<v Matt>That's why we're developers and not startup owners yet

00:10:47.168 --> 00:10:47.477
<v Nina>Yeah,

00:10:48.707 --> 00:10:55.555
<v Nina>But how has AI helped you with your consulting and migrating legacy code?

00:10:56.628 --> 00:11:00.228
<v Matt>Lately, I've been doing the Chief of Staff thing. Have you been kind of following

00:11:00.228 --> 00:11:01.461
<v Matt>this trend that's been popping up?

00:11:02.040 --> 00:11:02.266
<v Nina>No.

00:11:02.938 --> 00:11:07.380
<v Matt>So basically, a lot of people are using like Rockbot. Have you heard of that one?

00:11:08.092 --> 00:11:15.208
<v Matt>Yeah, so it's the XAI thing. And it's like a cloud-based agent with sub-agents easily.

00:11:15.208 --> 00:11:19.338
<v Matt>So basically, instead of doing all this stuff on your own machine, there is just a bunch of

00:11:20.124 --> 00:11:23.735
<v Matt>random number of amount of computers you get to use in the cloud

00:11:24.438 --> 00:11:30.568
<v Matt>and what everyone's doing is like having a main agent that's like the chief

00:11:30.568 --> 00:11:37.210
<v Matt>of staff and that guy will spin off things to like go to your GitHub, check out your

00:11:37.827 --> 00:11:41.928
<v Matt>repos see if there's any issues and go to your email, check what emails you

00:11:41.928 --> 00:11:48.270
<v Matt>got check your calendar and like spins up all the agents then comes back that builds your daily to-do

00:11:48.910 --> 00:11:51.585
<v Matt>list of what to do and what agents are up to.

00:11:52.082 --> 00:11:59.907
<v Matt>What I ended up doing was not using GrapBot at all. I went and I used the Hetzner.

00:12:01.088 --> 00:12:04.220
<v Matt>It's like a cheaper AWS, but you still get a full box.

00:12:04.799 --> 00:12:08.928
<v Matt>And so I put my account onto there, and then it can go.

00:12:35.346 --> 00:12:40.668
<v Matt>And also the best part is it'll go through the PRs that have requests and it

00:12:40.668 --> 00:12:41.780
<v Matt>kind of scores them for me.

00:12:42.812 --> 00:12:46.591
<v Matt>And so it's just like, oh, this is like a 7 out of 10 good.

00:12:47.150 --> 00:12:50.680
<v Nina>On what metric, though? Do you get to dictate the metric in which it's?

00:12:50.934 --> 00:12:54.451
<v Matt>The metric is, would you approve this PR?

00:12:54.451 --> 00:12:56.831
<v Nina>But can you customize that?

00:12:56.841 --> 00:13:01.340
<v Matt>Oh, yeah, you can customize it as much as you want. I mean, it's just Claude talking to Claude.

00:13:01.822 --> 00:13:06.221
<v Matt>So it's already like agents I had in each box that like have their own special

00:13:06.221 --> 00:13:09.915
<v Matt>sauce to be like, oh, in this repo, we do it this way and blah, blah, blah.

00:13:10.525 --> 00:13:13.517
<v Matt>And so would it then go ahead

00:13:14.141 --> 00:13:15.424
<v Nina>I was going to say that please,

00:13:16.650 --> 00:13:17.636
<v Nina>Do I got beef? Yeah.

00:13:19.784 --> 00:13:21.281
<v Matt>That's a pretty deep sigh.

00:13:21.885 --> 00:13:26.205
<v Nina>I mean, it's like you're being an overseer.

00:13:26.652 --> 00:13:32.282
<v Nina>I like the analogy of, let's say you're building a building on site,

00:13:32.282 --> 00:13:36.024
<v Nina>and you have somebody who's pointing, work or go there, work or go there.

00:13:36.407 --> 00:13:39.808
<v Nina>It's like, how do you keep track of everything at once?

00:13:40.191 --> 00:13:43.702
<v Nina>What does the interface look like when you see all that?

00:13:43.702 --> 00:13:49.022
<v Nina>It's like, how do you manage all these individual agents? I mean,

00:13:49.022 --> 00:13:54.772
<v Nina>it sounds like N8N or like LangGraph or one of those other tools where you're

00:13:54.772 --> 00:13:58.110
<v Nina>kind of orchestrating agents to do a million things.

00:13:58.602 --> 00:14:03.458
<v Matt>Yeah, that'd be similar. But this one's a little more wishy-washy.

00:14:03.725 --> 00:14:09.187
<v Matt>So basically what's on the Hetzner server is just its own Java server.

00:14:09.507 --> 00:14:12.525
<v Matt>And then inside the Java server, I got like Spring AI going.

00:14:12.990 --> 00:14:17.312
<v Matt>And it has like pretty strict things that it can do it's like oh i can go to

00:14:17.312 --> 00:14:23.432
<v Matt>the box and like run this script and the script only can do so much and that

00:14:23.432 --> 00:14:25.292
<v Matt>way it's like just can't go nuts like

00:14:26.012 --> 00:14:30.052
<v Matt>there is an escape hatch but it has to be talking to me that it can just like

00:14:30.052 --> 00:14:33.261
<v Matt>talk directly to its Claude or the Claude on a box

00:14:33.737 --> 00:14:38.252
<v Matt>but it's good it comes back with good scores or at least like it's like oh this

00:14:38.252 --> 00:14:41.031
<v Matt>is like six out of ten you should probably look at this one

00:14:41.792 --> 00:14:46.432
<v Matt>and like there's ones that are just like oh this is fine it's just like a one-liner

00:14:46.772 --> 00:14:50.992
<v Matt>change don't worry about it but i also got it so it sends me a slack message

00:14:50.992 --> 00:14:51.823
<v Matt>with all this stuff right

00:14:52.467 --> 00:14:57.070
<v Matt>but slack has this thing where you can kind of put like in little apps and messages

00:14:57.952 --> 00:15:01.945
<v Matt>and so the message comes with a button to approve and i can just hit the button

00:15:02.260 --> 00:15:04.273
<v Matt>from a phone and the phone then goes

00:15:04.894 --> 00:15:09.692
<v Matt>from the slack to the hetzner from the hetzner to the client box and then hits

00:15:10.092 --> 00:15:13.311
<v Matt>what does the GitHub approved command for that PR number.

00:15:14.132 --> 00:15:17.975
<v Nina>Oh, man. I know a lot of devs who are doing that, and...

00:15:20.141 --> 00:15:24.396
<v Nina>I want to get into doing that. It's like too many. Did you take like a,

00:15:24.500 --> 00:15:27.461
<v Nina>was this something you just learned on your own? Did you watch,

00:15:27.914 --> 00:15:33.201
<v Nina>was there anything you watched like a tutorial about how to use GrokBot to like

00:15:33.201 --> 00:15:34.541
<v Nina>orchestrate all of this?

00:15:35.861 --> 00:15:40.712
<v Matt>I'm lucky. I'm just like, been doing this long enough. I just know this kind of like pattern.

00:15:41.165 --> 00:15:44.451
<v Matt>Like all it is is just data pipelines and old fashioned ETL.

00:15:44.451 --> 00:15:48.951
<v Matt>But like you don't have to be as thorough and mapping out the pipelines you

00:15:48.951 --> 00:15:52.671
<v Matt>could be wishy-washy with cloud ai between the lines it's like kind of what

00:15:52.671 --> 00:15:57.551
<v Matt>everyone was selling on ETL like 10 years ago yeah never really paid off

00:15:58.208 --> 00:16:06.151
<v Nina>Yeah yeah so it's deep agent frameworks seem to be what is going to be i mean

00:16:06.151 --> 00:16:11.221
<v Nina>if anybody makes an independent an independent app that is kind of like software

00:16:11.221 --> 00:16:13.125
<v Nina>as service that, like, for devs,

00:16:14.021 --> 00:16:16.441
<v Nina>it's just going to be deep agent frameworks.

00:16:17.368 --> 00:16:20.631
<v Nina>And a lot of these, you know, that have MCPs that work with,

00:16:20.631 --> 00:16:27.088
<v Nina>like, Figma, that work with Jira, Atlassian, it's like, having a framework that kind of maps out,

00:16:28.185 --> 00:16:31.883
<v Nina>exactly the workflow for for etl or pipeline it's,

00:16:34.127 --> 00:16:38.494
<v Nina>i always worry about just the verification process and i know it's pretty good

00:16:38.494 --> 00:16:41.763
<v Nina>at verifying itself but like.

00:16:42.604 --> 00:16:44.387
<v Matt>What kind of verification are you talking about like

00:16:45.467 --> 00:16:51.335
<v Nina>Like basic ad hoc testing with unit testing like like actually testing the app not just sending,

00:16:51.903 --> 00:16:57.454
<v Nina>Claude into like let's say the chrome tab browser to verify things like Like actually

00:16:57.454 --> 00:17:00.367
<v Nina>like testing it at the end yourself and then,

00:17:00.924 --> 00:17:04.413
<v Nina>figuring out what, because every time I implement a feature,

00:17:04.912 --> 00:17:07.704
<v Nina>I only run like two or three agents at once.

00:17:07.704 --> 00:17:08.250
<v Matt>Yeah.

00:17:08.494 --> 00:17:11.582
<v Nina>Because I say, okay, once you get to this point, I need you to stop.

00:17:12.160 --> 00:17:14.638
<v Nina>I'm going to go to the Chrome tab and I'm going to test everything.

00:17:14.975 --> 00:17:16.351
<v Nina>And then I give it feedback.

00:17:16.467 --> 00:17:23.026
<v Nina>And that's, I feel like if I don't do that, I feel like everything's going to explode.

00:17:23.787 --> 00:17:26.504
<v Matt>I've been going back and forth, right? So on the client side,

00:17:26.504 --> 00:17:31.884
<v Matt>I have to test everything, basically try to get 80% coverage type stuff and

00:17:32.204 --> 00:17:38.666
<v Matt>end-to-end tests, all that stuff. Because there's a Jenkins or GitHub built around that anyways.

00:17:40.244 --> 00:17:46.674
<v Matt>But then it's just like on the Hetzner side, I think I have like five unit tests maybe.

00:17:47.070 --> 00:17:49.558
<v Matt>And that's it. And I didn't really care.

00:17:50.143 --> 00:17:55.112
<v Matt>And it's just like so bespoke that I don't think I'll ever care.

00:17:55.680 --> 00:17:59.930
<v Matt>That it's that 100 percent like it'll always just be me using it

00:18:00.710 --> 00:18:05.944
<v Nina>Right right but let's say it's like a it's something that you do want to you

00:18:05.944 --> 00:18:10.064
<v Nina>know provide as like a subscription service or some application that you want to.

00:18:10.064 --> 00:18:14.984
<v Matt>Sell well yeah so then like you got people paying i think once you start getting

00:18:14.984 --> 00:18:18.255
<v Matt>there's money on top of the code you should probably have some tests

00:18:19.445 --> 00:18:26.337
<v Nina>Yeah, I think Forge Audio has 950 tests. Whoa.

00:18:27.350 --> 00:18:31.450
<v Matt>How long does it take to run them all? Are we just unit or are we talking EDU as well?

00:18:32.043 --> 00:18:36.470
<v Nina>No, these are just unit tests. It doesn't take that long.

00:18:36.470 --> 00:18:38.220
<v Matt>It may not be too long, just unit.

00:18:38.864 --> 00:18:43.600
<v Nina>Yeah, they're just unit tests. I don't remember, but I always make sure to run

00:18:43.600 --> 00:18:46.050
<v Nina>them before I commit, and I commit in batches.

00:18:46.450 --> 00:18:48.830
<v Nina>I don't say every single thing commit.

00:18:49.848 --> 00:18:57.007
<v Nina>I say commit a group of these. And I should be better about making work trees and stuff like that.

00:18:57.232 --> 00:19:01.664
<v Matt>Yeah, nah. I don't do work trees. Work trees are useless.

00:19:02.286 --> 00:19:07.493
<v Matt>Are you trying to do parallel feature implementation at the same time?

00:19:08.507 --> 00:19:10.470
<v Nina>I have done it, but,

00:19:12.615 --> 00:19:17.591
<v Nina>I don't like doing it because I like seeing everything that gets written.

00:19:19.502 --> 00:19:24.404
<v Nina>And I guess so when it comes to this development, then there's like there's

00:19:24.404 --> 00:19:29.564
<v Nina>another follow up question I want to ask for for for people who didn't listen

00:19:29.564 --> 00:19:32.704
<v Nina>to the podcast. I guess you could link them to this podcast.

00:19:33.740 --> 00:19:37.690
<v Nina>Yeah. Who do you think you're you're more of Jordan or Justin?

00:19:38.538 --> 00:19:45.324
<v Matt>Oh, neither. So basically on the podcast, there's Justin, Jordan and Brian.

00:19:45.724 --> 00:19:50.274
<v Matt>So brian's the programmer justin's the marketer and jordan's more in operations

00:19:50.274 --> 00:19:54.914
<v Matt>it's kind of felt like is the way the trio is working at these days so i'm definitely

00:19:54.914 --> 00:19:57.940
<v Matt>more towards brian how about

00:19:58.960 --> 00:20:04.964
<v Nina>You um i'm a little little bit of both i i feel like i'm i mean yeah,

00:20:06.544 --> 00:20:16.104
<v Nina>i feel like i'm as cautious as as justin and jordan but i mean brian is trigger happy you know it's.

00:20:16.784 --> 00:20:21.224
<v Matt>I mean his entire business is teaching people how to use ai as coding agents

00:20:21.224 --> 00:20:25.335
<v Matt>so i feel he's i mean an unintended agenda is definitely there

00:20:27.093 --> 00:20:31.332
<v Nina>Yeah and i feel like when justin and jordan they were arguing that there were

00:20:31.715 --> 00:20:35.744
<v Nina>no at-scale success stories for this yet.

00:20:36.104 --> 00:20:42.801
<v Nina>I think that was one of their angles. I think once we get to a point where these,

00:20:43.374 --> 00:20:49.397
<v Nina>agentic frameworks are less new and more sophisticated, which will probably be,

00:20:49.990 --> 00:20:51.782
<v Nina>I don't know, like six months or something.

00:20:52.334 --> 00:20:54.122
<v Nina>It's moving so fast, I imagine.

00:20:54.482 --> 00:20:57.644
<v Matt>It feels like it peaks and plateaus at the same time.

00:20:57.944 --> 00:21:02.774
<v Matt>And just like, I can't tell if it's just a steady up and to the right,

00:21:02.774 --> 00:21:08.721
<v Matt>or if we really are starting to see the curve flatten and capabilities.

00:21:11.246 --> 00:21:16.354
<v Nina>I think it will plateau, but I think once there are more of these frameworks...

00:21:17.460 --> 00:21:20.676
<v Nina>We'll probably see at-scale success stories with it.

00:21:21.297 --> 00:21:26.963
<v Matt>So would you say like non-programmers, just like building sapses and spinning them out with AI?

00:21:27.748 --> 00:21:29.200
<v Nina>It depends on the product, right?

00:21:31.440 --> 00:21:36.550
<v Nina>I think with those people, there'll be more fires to put out. But I don't know.

00:21:37.636 --> 00:21:40.688
<v Nina>It's hard to tell at this point because everything moves so fast.

00:21:41.560 --> 00:21:46.870
<v Nina>But i think that yeah we might we might get there where we don't really need

00:21:46.870 --> 00:21:52.515
<v Nina>dev teams we just have one or two people spinning agents and you know it,

00:21:53.370 --> 00:21:55.630
<v Nina>kind of makes me a little sad because it,

00:21:56.606 --> 00:22:04.000
<v Nina>it does because if one person well a because you're not you're not working with

00:22:04.000 --> 00:22:06.531
<v Nina>a team anymore which has its,

00:22:07.658 --> 00:22:13.580
<v Nina>perks and its drawbacks you just have one person with one vision and if that

00:22:13.580 --> 00:22:19.041
<v Nina>person is tunnel visioned and they make a mistake with their agent,

00:22:19.835 --> 00:22:24.300
<v Nina>I mean there goes their company if it's a big mistake you know?

00:22:24.874 --> 00:22:28.597
<v Matt>What really is that kind of mistake? Like they delete the database?

00:22:29.027 --> 00:22:35.621
<v Nina>Yeah they don't have backups or anything or you know It's something like that Something extreme.

00:22:35.993 --> 00:22:40.650
<v Matt>So like I feel like those have happened But like also They're like people are

00:22:40.650 --> 00:22:45.937
<v Matt>learning from it And people are kind of backwards Becoming programmers from this

00:22:46.712 --> 00:22:52.065
<v Matt>yeah so like i've seen people like i've been helping a friend who's been doing

00:22:52.065 --> 00:22:56.675
<v Matt>this to like just kind of go through acquisition documentation like she does

00:22:56.675 --> 00:22:59.125
<v Matt>m&a stuff and she started with just like

00:22:59.605 --> 00:23:04.451
<v Matt>Claude in the on the browser then went to Claude Cowork and then moved to Claude Code

00:23:05.325 --> 00:23:09.205
<v Matt>and she's just like become a programmer over the past six months like she didn't

00:23:09.205 --> 00:23:13.586
<v Matt>know what a database was like three months ago and now she's like asking me

00:23:13.898 --> 00:23:19.762
<v Matt>how to do like as views and postgres and i'm just like what

00:23:20.460 --> 00:23:23.955
<v Nina>That's that's a new thing i haven't heard of.

00:23:23.955 --> 00:23:27.505
<v Matt>Like i mean it's basically like they just like hit a problem and then they dive

00:23:27.505 --> 00:23:31.090
<v Matt>super deep into just a tiny bit that's relevant to them and they pop back out

00:23:31.816 --> 00:23:35.625
<v Nina>Right because they're looking at what they want in the end and so they've backwards

00:23:35.625 --> 00:23:40.585
<v Nina>learned what a database is just looking at the cloud stream is what you're saying yeah.

00:23:42.493 --> 00:23:45.245
<v Matt>Because like they'll hit a problem they'll be like what's going on here and

00:23:45.245 --> 00:23:48.405
<v Matt>then Claude be like oh it's something with this this and then they just like

00:23:48.405 --> 00:23:52.985
<v Matt>try a few things and then it doesn't work and they've learned from those things

00:23:52.985 --> 00:23:55.970
<v Matt>i mean they're just learning they're just like learning the hard way

00:23:56.573 --> 00:24:02.227
<v Matt>and but they're able to iterate so fast they don't need like books or courses they can just do things

00:24:02.627 --> 00:24:07.595
<v Nina>Yeah that's incredible that's really cool but that's also that also gets me

00:24:07.595 --> 00:24:12.484
<v Nina>nervous because people go to school for so many years to be able to do this and now.

00:24:13.385 --> 00:24:20.305
<v Matt>I mean nothing's gonna match so one-to-one like with programming and ai so programming

00:24:20.805 --> 00:24:23.070
<v Matt>is definitely gonna become different a different way to learn

00:24:23.597 --> 00:24:27.895
<v Matt>i would be like way like in the beginning when like you just be like a script

00:24:27.895 --> 00:24:29.774
<v Matt>kid and like mess around on the internet,

00:24:30.325 --> 00:24:31.950
<v Matt>and I learned that way.

00:24:32.524 --> 00:24:37.830
<v Matt>And then there would be the more scientific stuff you'd learn in school that

00:24:38.196 --> 00:24:41.307
<v Matt>was more if you just wanted to be a logic-proof guy.

00:24:43.465 --> 00:24:44.868
<v Nina>Were you a script kid growing up?

00:24:45.584 --> 00:24:49.001
<v Matt>Yeah, I was beginning just like messing around the browser. And then,

00:24:49.690 --> 00:24:53.291
<v Matt>then I ended up going to college for it, learning all that stuff.

00:24:53.587 --> 00:24:57.400
<v Matt>But my college is weird. It kind of was like half, like real computer science

00:24:57.400 --> 00:25:02.892
<v Matt>and half like reading the rest PhD, that guy's dissertation.

00:25:03.693 --> 00:25:09.310
<v Matt>I had some weird professors who were very practical about the coursework.

00:25:10.836 --> 00:25:14.145
<v Matt>So I had a weird upbringing in computer science.

00:25:14.627 --> 00:25:15.411
<v Nina>Where'd you go to?

00:25:16.044 --> 00:25:17.194
<v Matt>I was in Loyola, Chicago.

00:25:17.678 --> 00:25:23.713
<v Nina>Okay. Okay. I did not get a computer science degree. I went to a boot camp, which,

00:25:25.963 --> 00:25:30.133
<v Nina>that was kind of backwards learning as well. Even though they try to teach you

00:25:30.573 --> 00:25:31.762
<v Nina>the concepts right up front,

00:25:32.420 --> 00:25:37.813
<v Nina>I felt like I didn't get those concepts until I dove in and they just gave me

00:25:37.813 --> 00:25:42.233
<v Nina>the list of stacks and technologies and I had to be programming like 12 hours

00:25:42.233 --> 00:25:44.590
<v Nina>a day no joke it was yeah that was hours.

00:25:44.980 --> 00:25:50.163
<v Matt>That was like kind of the big idea like back then to you could just do like

00:25:50.163 --> 00:25:53.297
<v Matt>six weeks of boot camp and then get a job

00:25:53.950 --> 00:25:56.293
<v Nina>They're not around anymore really it's it's,

00:26:00.842 --> 00:26:05.881
<v Nina>I mean They have agents, like, it's AI programming boot camps,

00:26:06.230 --> 00:26:09.434
<v Nina>which that feels like it's just kind of taking your money.

00:26:09.912 --> 00:26:14.800
<v Matt>Yeah, so, like, not even LLM training, just, like, using Claude to make...

00:26:15.561 --> 00:26:20.036
<v Nina>Yes, a lot of them are just, like, learn how to use AI to program.

00:26:20.686 --> 00:26:26.200
<v Nina>They do have machine learning and LLM courses, too, but it's still more...

00:26:27.433 --> 00:26:31.315
<v Nina>They're just... You're paying for them to just point you to use Claude.

00:26:32.133 --> 00:26:38.943
<v Nina>Yeah, exactly. And I don't know how to tell a friend who wants to get into programming,

00:26:39.565 --> 00:26:43.723
<v Nina>that this is probably, it might be the best time, but it also might be the worst

00:26:43.723 --> 00:26:47.233
<v Nina>time for you to get a job because nobody's...

00:26:48.560 --> 00:26:54.856
<v Nina>It seems like nobody's hiring junior devs because AI can just do their job for them.

00:26:55.303 --> 00:26:58.650
<v Nina>And that's what bums me out about this the most.

00:26:58.650 --> 00:27:03.570
<v Nina>It's like us who are mid to senior, it's like, yeah, we can do anything.

00:27:03.570 --> 00:27:09.720
<v Nina>We could, we could like found a consulting company. We can go work for a company

00:27:09.720 --> 00:27:12.572
<v Nina>that needs AI implemented in it.

00:27:14.130 --> 00:27:20.450
<v Nina>But the juniors are the ones, I think, that are suffering from this.

00:27:20.450 --> 00:27:24.251
<v Nina>And that's what bugs me, I think, a little bit.

00:27:24.802 --> 00:27:33.090
<v Nina>Like, how do we open up the door for people to get into tech now and actually

00:27:33.090 --> 00:27:36.741
<v Nina>get a job as a junior dev somewhere, you know?

00:27:37.536 --> 00:27:41.281
<v Matt>Yeah, I mean, but are we just oversaturated? like do we just have too many

00:27:41.940 --> 00:27:47.020
<v Matt>juniors i mean that was like the big cell to like push more and more people

00:27:47.020 --> 00:27:52.060
<v Matt>in computers push more and more people into the boot camps but now they're out

00:27:52.060 --> 00:27:53.595
<v Matt>of the boot camps now it's too many

00:27:55.180 --> 00:28:01.542
<v Matt>like it's just a real a real not even distribution of people who've come into the field

00:28:02.504 --> 00:28:08.506
<v Nina>I i feel grateful that i got my boot camp certification almost nine years ago,

00:28:09.063 --> 00:28:10.607
<v Nina>because it wouldn't have been possible now.

00:28:13.030 --> 00:28:18.230
<v Nina>But I don't know. I've known people, a lot of people haven't stuck with it.

00:28:18.610 --> 00:28:22.840
<v Nina>You know, people that got pushed into it or felt like they had to get into it

00:28:22.840 --> 00:28:24.884
<v Nina>because they just wanted to get into it for the money.

00:28:25.441 --> 00:28:29.640
<v Matt>Yeah, it was for a while. It was just like the CPA, like what CPA used to be.

00:28:29.640 --> 00:28:32.727
<v Matt>People just got it because the parents made them do it.

00:28:33.550 --> 00:28:34.074
<v Nina>That too.

00:28:34.486 --> 00:28:38.637
<v Matt>And then after a few years, they're just like, I don't want to do this anymore. why this is miserable.

00:28:39.671 --> 00:28:44.065
<v Nina>I'm going to miss those days when we go, oh, this is miserable.

00:28:46.224 --> 00:28:47.098
<v Nina>Just a little bit.

00:28:47.098 --> 00:28:47.780
<v Matt>Just a little bit.

00:28:48.256 --> 00:28:53.817
<v Nina>You know what I mean? Don't you miss banging your head over a keyboard for hours

00:28:54.032 --> 00:28:57.161
<v Nina>just a little bit because it feels like you earned whatever you made?

00:28:57.946 --> 00:29:01.488
<v Matt>I mean, I still get that point with Claude. More now, I just don't bang my own

00:29:01.488 --> 00:29:05.017
<v Matt>head. I want to punch Claude in the face. It's like, what are you doing?

00:29:05.720 --> 00:29:12.218
<v Nina>Yeah. Yeah. Yeah. But yeah. So we went a little bit off topic.

00:29:12.218 --> 00:29:15.625
<v Matt>But yeah, it's our podcast. We do what we want.

00:29:16.152 --> 00:29:21.220
<v Nina>Yeah. Yeah. The future, future of AI systems engineering is,

00:29:21.867 --> 00:29:25.748
<v Nina>I feel like we, the people who, who can do it,

00:29:27.008 --> 00:29:30.710
<v Nina>I mean, we're just going to need less steps and you know, that's,

00:29:31.012 --> 00:29:34.065
<v Nina>that's just the, the harsh reality of, of things.

00:29:34.176 --> 00:29:38.158
<v Nina>So people getting into tech, they, they have to learn how to,

00:29:39.095 --> 00:29:42.310
<v Nina>I guess, The bar is raised.

00:29:43.560 --> 00:29:47.567
<v Nina>Something that makes me really optimistic about the future is not only is the

00:29:47.567 --> 00:29:54.197
<v Nina>bar raised for developers' skill sets because of AI, but the bar of what makes

00:29:54.197 --> 00:29:57.924
<v Nina>a good application is going to be pushed to new levels.

00:29:59.437 --> 00:30:03.184
<v Matt>Does it, though? I have not seen an increase in quality.

00:30:05.138 --> 00:30:09.345
<v Nina>Well, I feel like we're going to have innovation and breakthroughs. like.

00:30:11.258 --> 00:30:15.737
<v Nina>I'm trying to think of an example of like something that could be elevated or

00:30:15.737 --> 00:30:19.747
<v Nina>pushed like things that would take forever to develop and add into an app that

00:30:19.747 --> 00:30:21.057
<v Nina>now we could do very quickly,

00:30:22.052 --> 00:30:25.362
<v Nina>like creativity when it comes to i don't know,

00:30:26.161 --> 00:30:32.087
<v Nina>for whatever reason let's say you have a very feature heavy app and it's a desktop

00:30:32.087 --> 00:30:35.148
<v Nina>app but for whatever reason you need to add ar technology,

00:30:35.740 --> 00:30:39.777
<v Nina>like why would you do that maybe there's a specific case where you have to add

00:30:39.777 --> 00:30:41.520
<v Nina>ar which is as you know where you,

00:30:42.115 --> 00:30:46.047
<v Nina>you do some sort of mapping of your face for whatever reason like i'll give

00:30:46.047 --> 00:30:48.054
<v Nina>an example game development,

00:30:48.780 --> 00:30:53.957
<v Nina>let's say game engines every there's there's hundreds of them and there's the

00:30:53.957 --> 00:30:57.405
<v Nina>big ones like unity or unreal engine,

00:30:58.131 --> 00:31:02.752
<v Nina>let's say they have additional like technology inside where,

00:31:03.467 --> 00:31:08.797
<v Nina>we're using an LLM under the hood in one of these to do an AR mapping of your

00:31:08.797 --> 00:31:14.125
<v Nina>face to generate a 3D model that is then optimized by machine learning,

00:31:14.795 --> 00:31:19.707
<v Nina>to have the vertices be smooth and run great in the app.

00:31:19.707 --> 00:31:24.476
<v Nina>So that's an example of a feature that could be elevating.

00:31:25.517 --> 00:31:30.717
<v Nina>One part of the industry where you can you can just use ar technology to create

00:31:30.717 --> 00:31:38.597
<v Nina>3d models from video so that would be like a raised standard so like creative things like that,

00:31:39.466 --> 00:31:43.145
<v Nina>But I can't think of another example. Exactly.

00:31:43.685 --> 00:31:46.043
<v Matt>It feels like there's this fear of this

00:31:46.940 --> 00:31:50.685
<v Matt>silver bullet that's going to kill programmers, and it's just not there.

00:31:51.353 --> 00:31:56.246
<v Matt>There's like, the problem wasn't writing the for statement.

00:31:56.881 --> 00:32:04.362
<v Matt>The problem was translating a normal person's idea of the world into a computer's idea of the world.

00:32:05.962 --> 00:32:09.097
<v Matt>That still needs a special kind of person.

00:32:09.526 --> 00:32:16.156
<v Matt>Do that right it's like where our creativity exists that's like what we do

00:32:17.648 --> 00:32:21.154
<v Matt>and it's always been what we do but we just like had to waste a lot of time

00:32:21.902 --> 00:32:27.082
<v Matt>writing that into a computer language where now we can translate it more into

00:32:27.694 --> 00:32:32.563
<v Matt>a computer style of talking and then talk to the Claude and it can do the computer

00:32:32.563 --> 00:32:35.946
<v Matt>language it's just like a new layer if like

00:32:36.730 --> 00:32:40.870
<v Matt>a virtual engine like we're no longer doing assembly or just writing java

00:32:42.488 --> 00:32:47.563
<v Nina>Yeah i think one of well you still work with them i don't want to should i mention

00:32:47.563 --> 00:32:49.690
<v Nina>his name i don't know no no.

00:32:49.900 --> 00:32:51.391
<v Matt>Okay no names okay

00:32:51.576 --> 00:32:55.363
<v Nina>Okay no names okay yeah one of the one of the people that we that we worked

00:32:55.363 --> 00:32:59.054
<v Nina>with when i would talk about ai with him very often,

00:33:00.168 --> 00:33:05.706
<v Nina>He said, and he's been a developer for like 20 plus years.

00:33:05.856 --> 00:33:13.700
<v Nina>He said this is like akin to when JavaScript frameworks came out or when Google came out.

00:33:13.890 --> 00:33:17.558
<v Nina>Like, we're always going to need developers and engineers.

00:33:17.878 --> 00:33:22.540
<v Nina>It's just that we have new tooling to make us do our jobs faster.

00:33:23.174 --> 00:33:26.675
<v Nina>And now with the agentic frameworks, even more so.

00:33:27.540 --> 00:33:32.477
<v Nina>And if I hope we don't reach a plateau I hope we keep pushing and.

00:33:32.477 --> 00:33:35.735
<v Matt>Finding and engines I don't know it feels like it's slowing down

00:33:36.757 --> 00:33:39.084
<v Nina>Are we talking about game engines in particular or.

00:33:39.907 --> 00:33:49.647
<v Matt>Just AI LLM like we kind of hit the big surprise and then a few more now it's

00:33:49.647 --> 00:33:53.347
<v Matt>like like Fable 5.1 just came out

00:33:54.162 --> 00:33:56.385
<v Matt>I don't notice a difference

00:33:57.170 --> 00:33:58.297
<v Nina>I do notice a difference.

00:33:58.297 --> 00:34:02.895
<v Matt>With fable they do yeah like from opus to fable yeah for sure

00:34:03.540 --> 00:34:05.984
<v Nina>Oh okay that's that's where i jumped but.

00:34:06.344 --> 00:34:14.737
<v Matt>No before fable 5 just released a 5.1 oh okay uh so that just came out like

00:34:14.737 --> 00:34:16.622
<v Matt>this week and then but like

00:34:17.491 --> 00:34:26.507
<v Matt>opus of fable was like okay i noticed a little bit like sonnet to opus like was a big jump i and then

00:34:27.687 --> 00:34:32.507
<v Matt>and then i forget whatever the hell was before sonnet basically went from like

00:34:32.507 --> 00:34:34.808
<v Matt>unusable to usable mm-hmm

00:34:35.686 --> 00:34:40.013
<v Matt>I think it was just a few versions of Sonnet, but whatever. So

00:34:40.813 --> 00:34:44.772
<v Matt>maybe it's still just the same amount of increase, but just not usable-to-usable

00:34:44.772 --> 00:34:50.742
<v Matt>is such a huge feeling increase that the capabilities, even though they're still

00:34:50.742 --> 00:34:54.291
<v Matt>just incrementing the same amount, aren't as noticeable.

00:34:54.922 --> 00:34:57.603
<v Nina>Yeah, I think I've got to get better hardware.

00:34:58.363 --> 00:35:00.133
<v Matt>Oh, I have to run it locally, yeah.

00:35:00.638 --> 00:35:04.452
<v Nina>Yeah, so I can start tuning models, because,

00:35:05.212 --> 00:35:12.522
<v Nina>that's something that's having specific models that are tuned for specific use

00:35:12.522 --> 00:35:16.492
<v Nina>cases, like spatial reasoning for, I don't know,

00:35:18.073 --> 00:35:20.605
<v Nina>things outside of what we do, web engineering.

00:35:20.860 --> 00:35:27.222
<v Nina>Like that that's really interesting to me robotics and ai i just want to i feel

00:35:27.222 --> 00:35:31.572
<v Nina>like a kid in a candy shop and i just want to be experimenting constantly even

00:35:31.572 --> 00:35:34.384
<v Nina>outside of web engineering but.

00:35:35.358 --> 00:35:37.412
<v Matt>Yeah i always wonder what specialization

00:35:37.412 --> 00:35:41.031
<v Matt>gets you like if it really does pay off in the long run yeah

00:35:41.635 --> 00:35:47.502
<v Nina>I i think it does like i'm sure i'm sure mechanical engineers they're gonna

00:35:47.502 --> 00:35:53.198
<v Nina>have to be you know are using models for, you know,

00:35:53.924 --> 00:35:57.816
<v Nina>like, agentic, like, what's the word?

00:35:58.048 --> 00:36:01.108
<v Nina>Just robotics. I'll just say robotics. Smart robotics.

00:36:01.820 --> 00:36:07.492
<v Nina>But, yeah, are there any hugging face models that you in particular have experimented

00:36:07.492 --> 00:36:12.977
<v Nina>with for, like, very special use cases for apps you wanted to make?

00:36:13.360 --> 00:36:18.362
<v Matt>Yeah, I've used the Whisper one to transcribe podcasts. And that one's,

00:36:18.362 --> 00:36:23.102
<v Matt>I mean, it's good. And it's also, I mean, I use Super Whisper to like talk to

00:36:23.102 --> 00:36:24.156
<v Matt>Claude from time to time.

00:36:24.696 --> 00:36:27.906
<v Matt>And so I guess, yeah, for like audio, it works very well.

00:36:28.956 --> 00:36:32.048
<v Matt>It works better than whatever the hell YouTube's doing for transcriptions.

00:36:32.048 --> 00:36:36.888
<v Matt>That's why I use it because if I want to have Claude search a podcast,

00:36:36.888 --> 00:36:41.013
<v Matt>I have it transcribed with Super Whisper locally and then look at it.

00:36:42.678 --> 00:36:49.401
<v Matt>It does a way better job than YouTube's transcription. And YouTube's transcriptions doesn't have

00:36:50.016 --> 00:36:55.384
<v Matt>who's talking where if Super Whisper will tag who's talking about what,

00:36:55.616 --> 00:37:00.898
<v Matt>even if it doesn't know, they'll just put unknown, bark It's just like someone's dog in the background.

00:37:03.243 --> 00:37:07.916
<v Nina>Yeah. And that app that you made is really, really useful.

00:37:08.796 --> 00:37:15.335
<v Nina>Maybe, oh, we got 15 minutes. Well, maybe next time we can talk more in depth about it. Yeah.

00:37:16.881 --> 00:37:19.508
<v Nina>Yeah, Whisper. Oh, go ahead.

00:37:20.685 --> 00:37:24.726
<v Matt>Yeah, Whisper's good. I got no beef with it. But I'm never going to make a model.

00:37:24.726 --> 00:37:27.088
<v Matt>I don't think I ever would. So you let me know how that goes.

00:37:28.266 --> 00:37:31.646
<v Nina>I got a MacBook Air, Matt. I don't think it's going to happen anytime soon.

00:37:32.504 --> 00:37:33.660
<v Nina>It's not going to happen.

00:37:35.106 --> 00:37:39.196
<v Matt>I did one, make a model, like, what was going to be, like, 10 years ago,

00:37:39.196 --> 00:37:43.105
<v Matt>back then when, like, you could, like, have a telepicture was,

00:37:43.500 --> 00:37:46.890
<v Matt>like, a hot dog or not a hot dog was the trend at the time from Silicon Valley?

00:37:47.566 --> 00:37:48.623
<v Nina>That was the biggest thing?

00:37:49.128 --> 00:37:53.667
<v Matt>Yeah, it still took like 10 minutes of training on my laptop.

00:37:54.168 --> 00:38:03.977
<v Nina>Oh my God. You see, that's where you need such insane hardware that if you're just one person,

00:38:05.286 --> 00:38:10.346
<v Nina>it's nearly impossible to train something that could keep up with any of these big name models.

00:38:10.780 --> 00:38:15.650
<v Nina>So it was 10 minutes for just determining is this a hot dog?

00:38:16.126 --> 00:38:21.025
<v Matt>Yeah. It went through like 100 tagged images. It was like part of a workshop.

00:38:21.548 --> 00:38:27.835
<v Matt>It had 100 images of hot dogs and dogs. And I had to choose between if it was a hot dog or a dog.

00:38:28.846 --> 00:38:32.422
<v Nina>What if you squished up the hot dog and then you presented it to you?

00:38:33.812 --> 00:38:39.026
<v Matt>It had a hot dog that was being eaten by a dog to trick it. And it had a dog

00:38:39.026 --> 00:38:43.097
<v Matt>in a hot dog costume to trick it. Just to mess with it over.

00:38:45.206 --> 00:38:46.071
<v Nina>How did it do?

00:38:46.768 --> 00:38:50.376
<v Matt>Basically like there was enough of us like 100 people each making our own and

00:38:50.376 --> 00:38:54.440
<v Matt>like when we ran it like about 10 people would get mixed up on each one

00:38:56.291 --> 00:38:58.817
<v Nina>Was this in what this was in school like a school project.

00:38:59.513 --> 00:39:06.773
<v Matt>No this is i got lucky and like old company at the time sent me to like an training program oh

00:39:07.508 --> 00:39:13.324
<v Matt>I was doing a project for, so like basically their idea was they'd be a middleman

00:39:13.324 --> 00:39:16.481
<v Matt>between companies and like Amazon and Walmart and stuff.

00:39:17.003 --> 00:39:21.644
<v Matt>So like say Kraft would come in and be like, all right, we want our macaroni

00:39:21.644 --> 00:39:24.964
<v Matt>and cheese box on all the e-commerce sites, but we don't want to individually

00:39:24.964 --> 00:39:27.661
<v Matt>put it. So they were just a middleman and they divvied it out,

00:39:28.010 --> 00:39:28.967
<v Matt>basically retail stuff.

00:39:29.420 --> 00:39:34.604
<v Matt>And so what we wanted to do with llms was you know amazon they like give you

00:39:34.604 --> 00:39:37.992
<v Matt>the image like in your hand so you know how big or small it is

00:39:39.384 --> 00:39:43.920
<v Matt>well we knew how big it was right and we can measure the hand we took someone's

00:39:44.087 --> 00:39:48.094
<v Matt>photo of a hand we measured it and so we like had both of them so we could resize

00:39:48.094 --> 00:39:49.856
<v Matt>the image to fit in the hand

00:39:50.325 --> 00:39:53.674
<v Matt>without like having to do a whole bunch of like graphic design on top of it

00:39:53.907 --> 00:39:59.195
<v Matt>to like or like actually take the photo in somebody's hand and and then gray out the hand and stuff.

00:39:59.723 --> 00:40:04.654
<v Matt>So that was the idea. We could get LLMs to verify it because the data wasn't

00:40:04.654 --> 00:40:06.535
<v Matt>100% good in our database.

00:40:06.814 --> 00:40:09.931
<v Matt>So sometimes you'd put something in someone's hand and it would just be like

00:40:10.175 --> 00:40:12.148
<v Matt>a can of tuna the size of your head.

00:40:14.047 --> 00:40:17.576
<v Matt>Someone used meters instead of inches or something like that.

00:40:18.220 --> 00:40:20.444
<v Nina>So this was like, how big was the database again?

00:40:21.298 --> 00:40:26.572
<v Matt>For those images, we had to have had hundreds of thousands, if not more.

00:40:28.264 --> 00:40:32.634
<v Matt>We never got to actually building it the idea was to build one what we ended

00:40:32.634 --> 00:40:35.854
<v Matt>up doing is just familiar with Mechanical Turk

00:40:37.310 --> 00:40:39.571
<v Nina>It's, if we refresh my memory.

00:40:39.571 --> 00:40:45.138
<v Matt>What time is it? So the idea is it's an Amazon product called Mechanical Turk.

00:40:45.527 --> 00:40:53.000
<v Matt>And basically you can, it just like pays an engine, pays like people 10 cents to do one tiny task.

00:40:53.405 --> 00:40:57.347
<v Matt>And our tiny task was to like, be like, is this image correct? Yes or no.

00:40:57.614 --> 00:41:00.821
<v Matt>And we like show them the image and then like, they'd be like,

00:41:00.821 --> 00:41:02.926
<v Matt>no, can size of your head.

00:41:03.407 --> 00:41:06.978
<v Matt>And so we just did that, but we did it internally too. We didn't even pay Mechanical Turk.

00:41:07.325 --> 00:41:10.341
<v Matt>We just had people inside our own company whenever they had free time.

00:41:10.341 --> 00:41:13.566
<v Matt>We built an app to just be like, yes, no, yes, no, yes, no.

00:41:14.035 --> 00:41:20.251
<v Nina>Yeah. You know, it seems that that's like a huge thing now. I don't know if

00:41:20.251 --> 00:41:25.381
<v Nina>you've seen on LinkedIn where they're hiring people to just train AI,

00:41:25.912 --> 00:41:28.692
<v Nina>doing stuff like that like a hundred times.

00:41:28.872 --> 00:41:29.633
<v Matt>Yeah, yeah.

00:41:30.295 --> 00:41:35.371
<v Nina>But they're paying like 150 an hour for doing something like that, which is insane.

00:41:36.091 --> 00:41:37.517
<v Matt>Get paid to have your job destroyed

00:41:38.141 --> 00:41:45.391
<v Nina>Yeah essentially yeah get get paid to i mean who knows what you have to go through

00:41:45.391 --> 00:41:47.315
<v Nina>i mean imagine getting trained on like,

00:41:48.430 --> 00:41:52.891
<v Nina>imagine you're trying to train a model for ethics or something that that would

00:41:52.891 --> 00:41:58.101
<v Nina>that would stink like like images for what is ethical versus what's not i've

00:41:58.101 --> 00:42:00.031
<v Nina>heard horror stories about it oh.

00:42:00.031 --> 00:42:03.771
<v Matt>Yeah i mean yeah breaks people, right? They have to go to therapy and stuff.

00:42:04.236 --> 00:42:06.935
<v Matt>I mean, you probably end up with AI that has to go through therapy at some point.

00:42:07.932 --> 00:42:08.333
<v Nina>Yeah.

00:42:11.142 --> 00:42:12.411
<v Nina>Well, yeah.

00:42:12.411 --> 00:42:14.991
<v Matt>On that note, shall we end the podcast?

00:42:15.431 --> 00:42:19.471
<v Nina>Yeah, it's kind of a depressing ending note. Yeah. Good talk.

00:42:20.891 --> 00:42:24.434
<v Matt>Thank you all. Alright, I'm going to end this. I'll see you next week.

00:42:24.994 --> 00:42:25.900
<v Nina>See you next week.

