Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

ADSP: Algorithms + Data Structures = Programs

ADSP: Algorithms + Data Structures = Programs artwork

Episode 246: Not High on AI?

Aug 8, 202537 minAudio
ShareXLinkedInFacebookWhatsAppEmail

In this episode, Conor gets Ben's thoughts on AI! Link to Episode 246 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP: The Podcast: Twitter Conor Hoekstra: Twitter | BlueSky | Mastodon Ben Deane: Twitter | BlueSky Show Notes Date Recorded: 2025-08-05 Date Released: 2025-08-08 2025 Stack Overflow Developer Survey ADSP Episode 244: High on AI (Part 1) Discussion Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity Software Unscript Episode 109: GPU Programming and Language Design with Chris Lattner Declarative Style Evolved - Declarative Structure - Ben Deane - C++Now 2025 Intro Song Info Miss You by Sarah Jansen https://soundcloud.com/sarahjansenmusic Creative Commons - Attribution 3.0 Unported - CC BY 3.0 Free Download / Stream: http://bit.ly/l-miss-you Music promoted by Audio Library https://youtu.be/iYYxnasvfx8

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

BenI think why not both? You know, people should be experts in as many things as they can. You know, your your brain is not a bucket to be filled where if you put too much things in it pushes other things out. No, your your brain is a tree where if you grow a new branch, you have more places to hang things. I did.

BenI did listen to those last couple of episodes. I'm not an adherent of AI assisted coding tools. I want to work on things that AI can't do, frankly. And I don't think I'm wondering whether AI is good for the world. You know? It doesn't seem to be doing much good for the world. This is important, right? It is important. So either, you know, either I'm gonna come across as an old fuddy duddy um and you can reject what I say, or I'm gonna come across as someone wise and you can listen to what I say.

BenEither way, I have opinions.

ConorWell, I would like to hear your opinions, and I'm sure the listener, I mean, there was, I think it is the Stack Overflow survey, which admittedly, now that I think about it, I I forgot had come out, but I heard about it on a Python podcast, and the reason I'm bringing it up now is they asked a bunch of AI tooling and AI questions, and one of them was like, What's your general experience? And I think it was 67 or 70% of folks said they were frustrated by them where they felt like they weren't uh meeting their needs for what they were doing.

ConorSo I I think that potentially uh Bryce and I are in the minority, and actually most of folks, at least according to the Stack Overflow survey, are not on the hype train. And so anyway, over to you. I'm happy to hear your opinions, and I'm sure everyone else is as well.

BenYeah, I think, well, I don't know about that. Uh well, I think one thing we can say is that from that survey, um the the average working developer um has less trust in AI this year than they did last year. That was one of the data points I think came out of that survey. Am I remembering that right? Well, I I haven't read the survey, so I'll take your word for it. Um I think one of the points was, you know, after having tried it, now we trust it less as a as a community of developers. Um not not we don't that number didn't go down a whole lot.

BenIt went down by like I'm gonna say five to ten percent. I can't remember the exact figures. But yeah, like many people who work in any kind of corporate enterprise, um there's been a lot of fuss over AI, right? And uh every time I have tr I was gonna say tried using AI, been subjected to using AI is maybe another term I could use, uh it hasn't really gone well.

BenYou know, um I've had AI do code reviews um on my code. Um it turned out to be very good at making work for people and saving no time at all. In fact, the opposite. It it I've seen AI leave so many comments on pull requests that I basically couldn't load the pull request in the browser anymore. That pull request had to die because it could not be reviewed, because the AI had left so many comments, it had basically killed the experience.

BenIt just wouldn't load. Is this on GitHub, GitLab, or? Uh yeah, on Git on GitHub. Wow. I've seen AI leave a dozen comments on one of my pull requests, and here's the thing: the suggestions always sound good on the surface. But when you dig into them, and it takes you a half an hour to go through each one and think, okay, you know, and they suggest writing this test or writing that test or doing this or doing that. And in one particular case, in this case I'm thinking about all of them turned out to be wrong.

BenIt was zero for twelve on its suggestions. Um but it took it took forty-five minutes to figure that out, right? As a human, right? It took, it it was, it's like dealing with, I've heard it described as being, it's like dealing with a brand new developer who you have to train, except they never learn.

ConorRight.

BenRight? You always have to tell them what to do. They never learn. And, you know, I don't really have time for that. I I have a lot of time for teaching people who will learn, but not for teaching AIs who won't.

ConorUh are you I'm it I'm not sure if you're allowed to say what because I know it sounds like you're using like GitHub Agent or CodeRabbit or one of those tools.

BenUh I I don't actually know. Okay. There are internal tools that we use. There are we certain we do use GitHub. I mean, we we use GitHub as for our regular development. Um But as far as the AI specifics go, I couldn't tell you because I actually don't know.

ConorOkay. But anyways, it was some kind of code review tool that was leaving automated. Uh this particular one, yeah. Yeah.

BenAnd have is that the main experience? Well, I you know, I have colleagues who use AI and who find it useful. You know, it's horses for courses. Um the kind of work I do Horses for Courses? What does that mean? Uh it means different things are suited to different jobs. Okay, okay. I guess. Sorry, it's a Britishism. You know, for some kinds of work, like like if you do work that's like summarizing documents or, you know, I don't know, because I don't do that kind of work, but I'm sure there are many areas where AI can be useful to pull things out.

BenBut if you're if you're doing work programming, and in particular, if you're programming and your your aim is to provide lasting value in the code base, then I don't think AI currently helps you. And bonus, it it's environmentally extremely unfriendly as well. So not only does it not help, um, it uh you know, it's very bad to the environment.

ConorI mean, we can definitely agree that it's it's bad for the environment. I mean, most of the uh AI people that are pro AI, I should say, they say that, you know, oh, we get to a certain point and then we can fix all our problems with AI. That is predicated on getting to that point.

BenWell yeah. It's predicated to get to that point. In the meantime, you know, data centers are projected to be the energy cost of of the fifth largest country in the world. And you know, according to many reports, the cost of a simple web search has gone up by 5x in terms of the environmental cost.

ConorOh right, because yeah, Google is prefixing all your questions with a little AI summary answer.

BenSo, you know, I maybe I'm old and curmudgeonly, maybe I'm old and wise, you can decide. But uh, you know, I I I don't think I I don't wanna I don't want to do that. Um also, you know, to my previous point, if our job as developers is to add lasting value, right? I have not yet seen a case where AI really helps with that. Like, I'm sure AI is great at producing code, right? It's great at solving problems if you want to throw away the solution.

BenBut I don't think it's very great at solving problems over time, solving software engineering problems. So let me ask you your experience there.

ConorI mean, I think you might be right in the moment, but I'm also not sure. I definitely can say anecdotally for myself, I find these tools like crazy force multipliers. Even for like non-coding tasks now. Like, there's so many times now I reach for these tools when I just need to solve some problem that is sometimes just in the real world. Like, for instance, over the weekend we were at a Shima and I were at a different uh place, not our own, and we were trying to watch and or season two, fantastic show, if you haven't seen it, and season one.

ConorAnd uh we had it I had it on my laptop and we were trying to um with an HDMI cable display it on the TV that was in this place we were staying at, and it was 4K, so like the rates were super choppy. And I don't know how to fix that. So I went and asked it, and actually I I went and just got cursor because now like I'm I'm on my laptop and I just say, Hey, I I don't actually need you to write any code for me, I just need help. This 4K TV is choppy. Uh can you fix it? And a minute later, it basically has run a couple commands using this tool called XRand, which I've never heard of before.

ConorAnd it basically it it did some XRAND list where it found the sources, it found the 4K TV, and it said, okay, yeah, let's just switch this to I think it was 60 frames per second instead of 120 or something, completely solved the issue. I didn't even know this tool existed. I'm sure I could have, using Google searches, like gotten to the solution eventually, uh, but it definitely would have taken way longer than 60 seconds. And that is like a concrete thing where you know it it is actually solved a like real life problem for me like over the weekend.

ConorUm so I I definitely think that these tools can be useful. Are they lasting or are they creating lasting value? I guess that's TBD. I mean, I know a lot of these companies that are writing these tools are kind of like dog fooding, and there are, I don't know what the numbers are of the number of developers that are using these tools, but um to a certain extent, a a lot of folks seem to be getting value out of this stuff. I will say that I I have a developing view that there are certain types of work, potentially all the code that you're writing, you know, template metaprogramming, Variatic, C, you know, on embedded software, that it's terrible for.

ConorAnd you can throw, you know, writing CUDA kernels in in that category of stuff, that it's just it's not gonna help you when the for for a certain set, like def define some parameters on, you know, very complicated code. It's not good at doing that. But if you need help spinning up some kind of like single-page application that you know is just statically serving you some information, like Shima now is uh is using these tools to like build like web like basically personal websites that are helping her with her job in terms of I mean Bryce and I talked about the the OHIP codes, but she started building another one for basically um medical prescription translations, if you want to get you know someone with uh you know certain medical thing on there on one drug, but you want to taper them off onto some other drug.

ConorUh there's like some complicated like math that goes into doing that. And right now, when you're uh uh an addiction doctor, you basically just do that personally, like by hand, and it's but it's very like you're doing the same thing every single time, but there's like a huge matrix of like are they taking pills? Is it is it Ivy? I don't know all the details, but it's like a very complex matrix of stuff. And she said it would be amazing. And there are actually websites that she's she mentioned MD Calc, which is this kind of website that doctors do utilize, but it's only for a certain set of these calculations. Um and she said, like, if I could have one specifically for my thing, this would save me like over my career, potentially like weeks and weeks of of work, because it's just it's stuff that you could do uh with some kind of medical calculator, it just doesn't exist, right?

ConorSomeone's not gonna go and build a one-off thing for that bespoke thing.

BenI would love to I would love to hear Shima's opinions. Um, and I think you and she should, if she's willing, you should record with her, because you know, otherwise this podcast is in the genre of two guys talking, and sometimes it's you and Bryce, and sometimes it's you and me. But either way, the genre of two guys talking is not what the world needs, I think, sometimes, yeah. So I'd love to hear Shima's opinion, you know, as uh medical professional, as a woman, as as someone who's not in the same demographic.

ConorYeah, I mean we uh I'm sure if you would if you want to be on, um she'd be happy to come on. In fact, we were talking about it back in June and we were gonna record, uh, but then I said, oh, really, we should we should try and get Ben because um I live a little bit more in her world, so probably I'm not the best person to ask like questions. Uh better to be um have some third-party AKAU.

BenSure, I'd be happy to do that. Um because the and the interesting overlap here is that you know the the other big problem with AI is is for me is um it AI cannot take accountability for code, right? That there's a legal problem, right? If I if I'm an engineer writing some code, I need to be able to take accountability for that. If I'm using AI LLMs to write it, where's the accountability? Like, this is why I say it it might be okay for like throwaway things, right? But it it I am not gonna have an LLM write my code, and then I have to stand behind that code.

BenRight? And so if an LLM writes it, I still have to write it, right? I still have to read every part of it. I still have to understand every part of it. Like the typing of the code is not the slow part there. In fact, the more code produced, the less I'm likely to understand it.

ConorRight. There is definitely a um a dichotomy between the vibe coding stuff that's like uh you know personal things versus stuff that you're shipping to production. And I I will say, like, I have used the code-assisted tools to help with writing libraries, but that is much different because I'm actually reviewing all the code that's being written there. It's it's it's still a force multiplier, but it's it's not the same as, oh, I have this dashboard that I want to build and I can just spin it up, and I really don't care the JavaScript that it generates.

ConorI just know what I want it to look like, and I can verify you know what it looks like at the end of the day to know that it's correct, you know.

BenAnd and in the case where you have to verify everything, you know, there is there is one data point, one study that's been done that I'm aware of, which I think someone mentioned on the on the GitHub for one of the last episodes. There is one study so far, and uh it showed that developers think they are 20-25% faster. The data shows they are 20 to 25% slower. But even when confronted with the data, of course, the classic thing is they still believe they're 20 to 25% faster. And I have to say, Connor, your response to that was perfect.

BenYou just said, well, I flat out don't believe the study, I'm gonna call you out on that. You know, so I think there is this, maybe it does reflect this dichotomy of like vibe coding, you can produce a lot of code. But but coding when it when it matters, which which frankly is the code that I want to do professionally, right? I don't want to be vibe coding on stuff that doesn't matter. When it matters, it's not speeding us up.

ConorI mean that's that's where I I mean, like I said, I I do think that my developing view is that there are types of coding that it helps you for, and other types that it is much less helpful for. That I mean, that being said, I like I'll read because Ben is not mistaken. He did he did read my reply, and the first sentence is anything, and so the the um this is well, let's see if we can get their name there. Handle is M. Heyman, uh, which is for Michael Hayman. So uh this was on episode 244. Uh they posted so there's this measuring the impact of early 2025 AI on experienced open source developer productivity that just came out, uh saying what Ben just mentioned.

ConorAnd uh we'll skip over the rest of uh their post. I replied saying anything that says Gen AI tools are slowing people down, I just don't believe. And then I'll skip the next thing I said, but I I pulled out two things. I didn't read the study in full, to be fair, but uh they were using Cloud 3.5 and three points on it. So I understand that when you do a study, you're doing the latest and greatest, and probably that was the latest and greatest at the time. But my my first remark is do the study again with Cloud 4 and see what they say. And the other thing is I wanted to I wanted to see like who are these folks, like what is experience, because it's kind of like the analogy I had in my head is if like if you're gonna go to some race car track or some F1 circuit, and then you get some like moderately experienced, which is what they say in the the paper.

ConorThey say it's 16 developers with moderate AI experience, and you put them in some car, that doesn't really show like what you can do with the car. But if you get, you know, the best drivers in the world or whatever, the people that are the power users of these tools, and see what they can do, that's actually measuring like what the car can do. So it might not be a perfect analogy, but I I kind of thought, you know, that the lack of detail on how much experience do these folks like have with the AI tools, it doesn't it's not uh super fair to say that like, oh, these people, and if your point is like, well, what if we just unleash these tools on the world, like what impact is it gonna have?

ConorOkay, maybe that's fair.

BenYeah, I mean I I I don't think you can dismiss that point because they're not power users. If they're representative of your your regular working programmer, then it's not just you know, what's the cost of unleashing these tools on the world? The question is, like, are the economics even there? Right? Regardless of the other externalities. Economics meaning Like, is it worth it for companies to adopt these things?

ConorWhich which they are trying to do in droves? Right, I see. You're saying that companies don't employ just the power users of whatever tools they're using, there's a distribution of folks.

BenRight. And everything's on a power curve. So your power users are like two, three percent of your users, right? Maybe. I'm I'm ballparking. Right, right. Something like that. But like Yeah. I guess that's fair. I mean, uh I'm just saying, like it's fair to say like this is not my personal experience, but I don't think we get to reject the finding.

ConorYeah, maybe that's true. I mean, I still I still, in my heart of hearts, find it hard to believe. I mean Richard Felbin, who has his own podcast, uh Software Unscripted, he just had Chris Latner on for the second time um talking about mojo primarily. Uh, but obviously they're talking about AI as well. And it at one point, Richard said that he just didn't believe like when people are throwing out like 10x and 100x productivity gains. He had some quote where it was like, you know, you can see 100x from space.

ConorLike there's there's no way anyone's having that kind of like productivity gain.

Richard Feldman (Clip)Well, I think the the my biggest pet pee with the hype is when people start throwing around numbers like 10x, 100x. Like, you know, I appreciate that at the beginning of the conversation, very quick to be like, hey, we're not saying we're 10x faster than Rust, right? Yes, but like with AI hype, it's like the opposite. Someone's like, we are 100x as productive, and I'm just like, wait a minute, wait a minute. Like, you can see the outputs of 100x productivity from space. Like, if you're actually 100x productive, you don't need to tell anyone about that. You just have people walk walking by with their jaws open at like what you're producing, and they're like, How are you possibly outputting this much?

Richard Feldman (Clip)Which of course is not happening. Correct, 100%.

ConorYeah, yeah. And in my head, I in my head I was screaming, like, well, it's I think it's even it's I think it's like for certain people like Shima, who admittedly, she actually at a internship or at some point in her life was actually did do a little bit of MATLAB uh in some research position back when she was, I think, in her undergrad, uh so many years ago. So it's not like she has zero programming experience, but she definitely would not call herself a programmer. But the kind of apps that she's you know, she's a doctor.

BenShe is well qualified in a technical field. Right. Let's say. Yeah. She's done plenty I'm sure she's done plenty of math in her career.

ConorAt one point, yeah, she considered pursuing a a career in in mathematics. Um but the point being is the the apps that she uh is able to build with these tools is something that's like would without going and you know, developing a whole different vertical skill set, she is incapable of doing. So that's like it's zero to like impossible to possible. So like if if we're measuring like productivity improvements, it's like, well, that's something that she would have to go spend days, weeks, like however long getting up to speed where she could build that. So like in that sense, like isn't it greater than 100x?

ConorLike it's and I I feel like people like in this study too, the the thing that they're primarily um measuring is open issues on open source like GitHub repositories. And yeah.

BenWell they tried to measure like the a regular day job, right? I think they tried to take here are regular programmers doing a representative thing they would do during their day, right? And that's different from Shema, right? Shima's main job is not writing code. Um, it's a very, very small part of her job. Now, if she can amplify that, but it then it's great. But in the end, what does that account for? 2%, 5% of her job? I don't know. But but it's not, you know, it's not the focus of her job.

ConorYeah, I guess that's true. I'm guessing.

BenI should say I'm guessing.

ConorWell, the the the tools that she's building are just to uh help her. It's not it's it's yeah, maybe not indicative of the majority I I guess that's a question is what is the pie chart of software development and software engineering and code writing that happens in the world? Because uh That affects, I guess, how people are gonna have their productivity amplified. The cases where I think it's like the best is where you're going from zero to a hundred of some tool that you want or some website that you want, and it's gonna do all the scaffolding, all the setup.

ConorUm like just just this morning I wrote a cool uh a tool that I called STT, which is speech to text, because I'm like I'm spending so much time like dictating to these models that like or or typing, I should say, at the moment. I just I just basically want to speak to it. And I I looked on Linux, there was a couple different options, but Speech Note, the number one recommended one, didn't work for some reason. And then the second option was just to use OpenAI's whisper. Interesting. Anyway, so I just built uh I just asked it like, hey, you you wrote me this script to run this tool, throw it in a little GUI, but the main thing I want is just like enter to record, enter to stop, and then when it's done transcribing, copy it to the clipboard so that I can just immediately control V it into the uh agent window.

ConorAnd like with within like 10 minutes I was done. Um I mean it did have to figure out some TK enter or tickle TK, there was some issue, but that was just like on my desktop and it fixed it. But like it's like, I don't know, 350 lines of Python code. I I don't know how long that would have taken me, but I guarantee you I saved at least like 10x. Uh like there's no way I'm gonna write a 350-line Python script that's making use of TKINTER and Whisper and a bunch of stuff that like I've never messed around with.

ConorLike it's gonna take me at least a day, right?

BenLike, and that's it's also predicated on the idea that you can speak comfortably faster than you can type comfortably for extended if if you're doing text-to-speech. Uh and I think for some folks that's true. Maybe for most folks, it's a great accessibility game, I would say. I think for some folks that that proposition is a marginal marginal win at best. You know, some folks can type a lot faster than they can speak. I you know, I have a friend who uses a uh stenography keyboard.

ConorOh yeah, yeah. We chatted about this once, yeah.

BenYeah, yeah. So it's a dying art, but it's sort of coming back through, you know, the niche the niche internet. Um so it's a thing. If, you know, and if you're an experienced stenographer, you can type at 250 words a minute and you can type all day long with with no RSI. Right. Right? So But that again is a niche thing. I think for most people it's true. Text to speech is uh a productivity gain, an accessibility gain.

BenYeah. Let's say as well.

ConorIt it also means like uh I only use it when when Shima's not around, because I'd probably drive her nuts just talking to myself. Yeah, there's some Yeah, exactly. And you know, i if we move at some point and I have my own room, then okay, I could close the door and I won't irritate her, and hopefully she'll be at a different different corner of whatever the place we're at. But um this all to be said is that like this is an example of a tool that like definitely like it took me like grand total, like 10 or 15 minutes, and I don't know if it's a day, I don't know if it's half a week, but like the product, like my ability to make that tool is like at least 10x faster.

ConorThis is the type of thing where like I said, I'm going from zero to a hundred. If you're working on an existing uh very like old legacy or not even legacy code base, maybe it was built in the last year, but you know, it's tens of thousands, if not hundreds of thousands, heaven forbid, millions of lines of code, um, and you just have a bunch of open issues, potentially the productivity you're gonna gain, you're gonna get there. Like if all the, you know, like I know your workflow, you have all the bells and whistles already set up. You've got your testing frameworks and your your clang tidies and formats. Yeah. And if all that stuff is already done, and the only thing you need help with is basically solving these open issues, whether they're simple or gnarly, you're probably uh what are they, your your mileage may vary.

ConorLike the the mileage that you're gonna get out of these tools is is definitely way less than the zero to one hundred use case.

BenWell, and let's face it, these AI tools are great at spinning up boilerplate, like spinning up a new project. But but let's actually be clear here, it's pretty sad that we need them to do that. Like spinning up a new project, even in C ⁇ , even with dependencies, ought to be I can do it in a couple of minutes. I don't need to use a tool. Right? And if you have a certain workflow and if you have things set up like I do, like for my professional environment, it is that way already.

ConorYeah. I mean I I made that is uh I 100% agree. And I made a comment when I was chatting with Bryce that I said maybe it was because of the painfulness of the C experience that I I like these tools, and there's a ton of people that are are not uh thrilled with it. And I I mentioned Rust and Ruby because those are, you know, off the top of my head, two languages that I know that really, really are tried to make sure that the tooling and the uh ecosystem experience of those languages is really good.

ConorSo potentially if you're in Rust, when all you have to do is Rust up, you know, I don't remember what the cans are, but it's just basically a couple to install the latest Rust, and then cargo init, I think, to and they even initialize your like hello world, and then you're off to the races, and you've got unit tests built into basically like the language and documentation, all that stuff. So potentially the value that you get from you know uh these models if you're a Rust developer is a lot less than if you're in a language like C or that has a lot more tooling and ecosystem ecosystem problems.

ConorAnd don't get me wrong, C has made massive uh improvements over the last decade. But um there's still a long way to go, I think.

BenWell, and it's not just language specifics either. This is where I get to put on my really old curmudgeon hat and uh, you know, bring up people who use command line, people who use Emacs, people who use Vim. Generally, you find as a whole, among those kind of folks, they are much more efficient at doing things on the, you know, well, I say doing things on the command line, obviously that's a tautology, but but they're more efficient at using the tools than you know, folks who haven't learned that and are used to using the more GUI-oriented tools.

BenUm and even, you know, even modern editors like VS Code, they have a lot of bells and whistles, they do have a lot of things in there, but they don't have the kind of focus on editing that has been going on for 40 years in environments like Vim and Emacs, right? Right. And and you know, I know I'm sounding like I know how I'm sounding.

ConorUm I mean there there is uh there's something to be said, you know. Uh people always say, learn your tools and it's gonna serve you for a lifetime, right?

BenYeah, yeah. If you really get to know your editor, uh well, the point about Emacs is it's more than an editor, right? It's like an operating system. It's an environment. What makes it so great is you can look at any part of it. You can ask it about itself. You don't know how something is happening, you can ask Emacs, right? The whole environment is open for you to do that. Um, and you can make it do literally anything you want.

ConorRight.

BenRight? And so if you put that together with some command line now, right? So you know Git, you know CMake, um, you know grep exargs find, all those all the all those other small POSIX utilities. You know, and and it's not just Emacs, it's it's you know, Emacs is a lot easier to program than Vim. But you know, Vim's great uh strength is its um it's its language for manipulating text, the movements and such, right?

BenAnd you can and that can be implemented inside of Emacs. In fact, many people use evil mode. But I think I think among the camp of people who really sort of mould their environment around themselves, they already have a super nice workflow. You know, I've worked with people who um who are just wizards of that kind of thing. You know, you can you come to their desk, you ask them a question, before you've even finished asking the question, they've been typing to find the answer, right? And not asking Google, but just doing some command line query, stringing together some Unix tools, and before you finish the question, literally they have the answer.

BenRight? You're asking about some, you know, some question about a dataset that the company's working with. They can literally work at the speed of thought.

ConorYeah, I mean my thought, and this is gonna sound uh I don't I don't mean this uh uh this is just the thought that popped into my head is that is it is there analogy um because I I definitely agree that learning your tools it's it serves you for life. Like I I completely agree with that sentiment and personally have experienced it. Like I remember when I I did my first um kind of real co-op. I I had a I worked at my university a couple summers, but the first time I had a real internship during university was at an insurance company in Canada called Manulife.

ConorTechnically, actually, it was at the subsidiary John Hancock, which is uh American company. And actuaries use a lot of Excel, and I remember uh standing behind my boss at one point, um, who was a young guy at the time, uh so he's only a few years older than me, and he was like moving around Excel like I had never seen before.

BenRight.

ConorAnd I was like, what I was like, what are you doing? And he was like, What do you mean? I was like, I was like, you just like went down to the bottom of the data in that column. And he was like, uh, yeah. And I was like, how'd you do that? And he was like, what are you talking about? Like this is this is like 101 Excel. Like everyone knows this. But I was in school, I'd never seen anyone do this, and he's like, Oh, you just hit control and use the arrow keys, and if you want to select, you hit control shift. It blew my mind. It was like opening up Pandora's box. Anyways, the point being is yeah, over the next four months of that co-op term, I became like an Excel guru.

ConorAnd I remember I remember in the interview process, they asked me on a scale of zero to ten, how are you in Excel? And I think I said nine or something. Uh, because I was like, ah, there's always room for improvement. I was a zero. I think I've told this story probably on the podcast at some point. I was a negative one. You know, I knew how to make a bar chart, and that was it. And I thought that's what Excel was for, but Excel is this massively powerful tool. The point being, I totally agree with the sentiment, Emacs, Vim, the folks that are, you know, that 1%, it's like it's like magic watching people, people work, you know. They never use their mouse. Yeah. It's fantastic. That being said, I do wonder uh the thought that popped in my head is the folks that are, you know, doing the magic things, is it the equivalent of the person that was like the best at the abacus right before you know calculators came out?

ConorAnd like calculators were the democratization of the ability to do do those types of calculations. You don't need to be an expert to use a calculator. And, you know, it did mean that folks that were experts at the abacus, did it did it obsolete? Is that a verb? Can you what is the verb of uh Yeah, I think that's right. Yeah, to obsolesce that skill set. Anyway, so that is I don't mean it like, oh, if you've become an expert in this stuff, uh it's not important anymore. But at the same time, uh should it be that we're uh encouraging folks to, you know, become an expert in this vertical instead of saying, hey, there's this tool that you don't need to become an expert in it.

ConorAnd if you if you watch the the thinking or the reasoning of these models, they're doing all that stuff. They're grepping, they're using sed, they're using awk.

BenWell, I think I think why not both? You know, people should be experts in as many things as they can. You know, your your brain is not a bucket to be filled where if you put too much things in it pushes other things out. No, your your brain is a tree where if you grow a new branch, you have more places to hang things. Oh, did you just come up with that on the spot or is that a quote from someone? Uh that was brilliant. Uh that I don't know. I don't think it's an original thought. I wasn't quoting anyone, but uh but but that's that's how to look at it, right?

BenRight. Because because that's how knowledge is. It's it's associative in nature. We don't we don't forget one thing because we learn a new thing. No, we we relate new things to the things we already know, and our whole knowledge tree grows.

ConorYeah. I I mean I don't uh that's a br uh a fantastic sentiment. And I don't disagree. I yeah, I'm curious to see what will happen. I like I said, my evolving view is that there's a set of things, and I think what's gonna end up happening though is that for the non-experts, they're gonna veer more towards this kind of AI assisted code writing, and then there's gonna be not a bifurcation necessarily, maybe it'll always be some kind of sliding scale.

ConorBut uh for the folks like you, um, and maybe this is a good transition point to your uh C now talk that is not not public. I'm not sure. Are we allowed to link it to the listeners if they're listening? It's fine.

BenWell, it's unlisted at the moment.

ConorIt's unlisted. Uh but if guess what? If you're if you're a listener, we'll link it in the show notes. Uh but yes, I mean you are showing I I would say some pretty sophisticated stuff in your talk, especially at the end there, where you had the the whole the bonus slides where you had the whole thing. Oh, that was just the Yeah, that's just something silly. Um But the point being is that I I I do think there's gonna be kind of like the this bucket of experts where expertise is really required. That being said, there will be a percentage or fraction of the work that is definitely outsourced to these LLMs.

ConorBe sure to check these show notes either in your podcast app or at adspthepodcast.com for links to anything we mentioned in today's episode, as well as a link to a GitHub discussion where you can leave thoughts, comments, and questions. Thanks for listening. We hope you enjoyed and have a great day. I am the anti Bryce. Um

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Episode 313: C++26 Reflection vs CRTP, AI, Kids & More with Zach Laine

    Nov 20, 202647 minAudio

    In this episode, Conor and Bryce chat with Zach Laine about C++26 Reflection vs CRTP, AI, kids & more. Link to Episode 313 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP:...

    • Apple Podcasts

    Credit the people on this episode

  2. Episode 312: The C++26 Contracts Fiasco

    Nov 13, 202652 minAudio

    In this episode, Conor and Bryce chat with Zach Laine about the C++26 contracts fiasco. Link to Episode 312 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP: The Podcast:...

    • Apple Podcasts

    Credit the people on this episode

  3. Episode 311: Career Changes, GPUs & More with Zach Laine

    Nov 6, 202628 minAudio

    In this episode, Conor and Bryce talk with Zach Laine about his career and the career changes he has made. Link to Episode 311 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials...

    • Apple Podcasts

    Credit the people on this episode

  4. Episode 310: Leaving NVIDIA for... ?

    Oct 30, 202624 minAudio

    In this episode, Conor and Bryce talk about Bryce's next chapter and where he is leaving NVIDIA for. Link to Episode 310 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP: The...

    • Apple Podcasts

    Credit the people on this episode

  5. Episode 309: Default Construction & CppCon 2026 with Jason Turner

    Oct 23, 202629 minAudio

    In this episode, Conor and Ben chat with Jason Turner about default construction in C++. Link to Episode 309 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP: The Podcast:...

    • Apple Podcasts

    Credit the people on this episode