Algorithms + Data Structures = Programs - Episode 249: AI, Podcasts, Scandinavia Trip and More!
Episode Date: August 29, 2025In this episode, Conor and Bryce chat about some open source projects, podcast recommendations, our upcoming trip to Europe and much more!Link to Episode 249 on WebsiteDiscuss this episode, leave a co...mment, or ask a question (on GitHub)SocialsADSP: The Podcast: TwitterConor Hoekstra: Twitter | BlueSky | MastodonBryce Adelstein Lelbach: TwitterShow NotesDate Recorded: 2025-08-21Date Released: 2025-08-29Astro Bot VideoADSP Episode 176: 🇺🇸 prior, deltas & Dinner with PhineasThrust Github Search Vibing ProjectPaddlePaddle/Paddle RepoUber AresDB RepoLatent Space PodcastBig Technology PodcastCheeky Pint PodcastDwarkesh PodcastTraining Data PodcastADSP Episode 39: How Steve Jobs Saved Sean ParentRoku Engineering SymposiumCopenhagen C++ MeetupCasey Muratori – The Big OOPs: Anatomy of a Thirty-five-year Mistake – BSC 2025NDC Tech Town CUDA Python WorkshopNDC Tech Town CUDA C++ WorkshopIntro Song InfoMiss You by Sarah Jansen https://soundcloud.com/sarahjansenmusicCreative Commons — Attribution 3.0 Unported — CC BY 3.0Free Download / Stream: http://bit.ly/l-miss-youMusic promoted by Audio Library https://youtu.be/iYYxnasvfx8
Transcript
Discussion (0)
There's vibe coding.
That's what Andre Carpathie came up with.
You don't look at the code and literally you're just building some kind of visual thing
or maybe it's even a non-visual thing, but it has to have some kind of output in order
for you to verify.
And you never look at the code.
Then there's rage coding.
That's when vibe coding or this kind of coding goes poorly.
And then you're just asking it to do something.
This happened all the time with Claude 3.5 and 3.7.
You'd ask it to do something.
It would spin its wheels for a while and then it'd be like, you know what?
The solution is we're deleting all those tests.
We're going to get the test to work by deleting them,
or we're just going to remove the file.
The file's not there, now it works.
And that's rage coding, because it is so frustrating.
It's like gaslighting you.
It's like, oh, the test works.
I know it doesn't, because you literally just output like it doesn't work.
Anyways, the middle is guide coding.
Guide coding is when you're pair programming with it.
You really need to guide it where it's going, and it goes well.
So we've all been there, except for Cloud 4.
Cloud 4, it's amazing, folks.
It's amazing.
Welcome to ADSP, the podcast, episode 249 recorded on August 21st, 2025.
My name is Connor, and today with my co-host, Bryce, we chat about a plethora of topics,
including some open source libraries, some podcast recommendations, our upcoming trip to Europe,
and the upcoming trilogy of episodes with Sean Parent.
All right. What should we talk about? Should we talk about the three bullets I sent? Or should we talk about AI things? Your choice, dealer's choice.
We can talk about everything, but first, I need to open up. I actually wrote down a list of things to bring up.
Things to bring up.
Number one is I made a reference to the Dunkie GTA6 video in our last recording where they, I worked
the clip graphics. Look at the graphics on these graphics.
Look at the graphics on this graphics.
It's not actually from the GTA6
donkey clip. It's from the Astrobot
video. So if you went and looked for it
and you were like, what's this guy talking about?
The Astrobot is one of his best.
Anyways, that's number one. Number two,
Phineas, past guest on the pod
when we went to that vegan restaurant in New York
and then we went to the escape room
afterwards and luckily he did not have to leave in the middle of the escape room
his wife going into labor he pointed out that when we I was talking about the
repetition memory you know system I knew it had a name I couldn't come up with it
he messaged me when he was listening to it and he said it's called I might
actually mispronounce it Anki A-N-K-I that's the system of like you've got flashcards
you discard the ones you know right away and then you keep on
feeding to yourself the ones that you have trouble with so it had a name i knew that i just couldn't
come up with it perhaps it says something about our natures but um i do and i do feel like every
time we chat about i i encounter greater frustration than than you do like the latest thing was
like i wrote some curse rules from a project like i told this thing like that by thing i mean
i mean uh clod i told it you know if you want to run the the tests in this
repo. You got to use this script because the script like launches a container on a remote node
that has the right hardware to test this thing. And you know what it kept doing when I asked it to make
a change? It's just like I wrote this very clear cursor rule. I was like if you want to run the
tests, you have to do this. And then like it just kept repeatedly not doing that when it wanted
to run the tests. And I'm like, you don't understand. Why didn't you listen to the cursor rules? And
It was just like, oh, I'm so sorry.
You're absolutely right.
I did just ignore what you told me.
Wait, did you test that the cursor rule worked before you made it a cursor rule?
What do you mean?
I mean, cursor rules, in my experience, I don't use them much anymore because Cloud 4,
it's Clot for Sonnet, to be precise.
I know the Opus model's bigger, folks, but you need to be able to iterate quickly and interrupt
the model.
You can't do that with Opus and the big ones.
Sonnet's the best.
People keep asking, what do I use?
It's Claude for Sonnet, best in the market.
And in my experience, you shouldn't make a cursor rule until you find the golden path of what to say that gets it to do it correct all the time.
Once you find that golden path, then shove it in a rule.
This is like, I listen to like, oh, yeah, remind me to list off like six different new podcasts that I listen to all about dev AI tools and AI because I'm going through like a, that's not a midlife crisis.
We just, I was an actuary at one point, then I was an engineer.
I guess now I'm a software engineer still, but it's all AI.
Anyways, it's all about context engineering.
So you have to figure out what is the,
and that's literally the special sauce of all these frontier labs
that are making these frontier models, foundation models,
is they all have this like, you know, special incantation of words and whatnot
that they whisper to these huge LLMs,
and then based on how good that kind of context that they give it
or the algorithms that they have for compressing that context
so that you don't end up with like an ever-growing context
and the models go crazy.
So the point of meaning,
you can't just write a cursor rule
and expect that it works.
You've got to test out which one works best,
find the golden path,
then throw it in a cursor rule.
So that's what I've been doing.
Because the only reason I've written cursor rules
is like I've repeatedly run into it like not
like every other chat or so.
You know, it tries to run the tests in the wrong way
and then I just tell it, no, no, no,
you have to do this.
And then like at your five times,
I just put it into the cursor rule.
And it's just, it's like two seconds.
sentences. Like, it should be able to figure it out. But, but it, uh, and then like a lot, I feel like
my impression is a lot of times it does dumb shit. And you know what? When it does dumb shit, I get
pissed. I understand. That's called rage coding. You know my three different categories, right? There's
vibe coding. There's vibe coding. That's what Andre Carpathie came up with. That's the, you don't look at the code
and literally you're just building some kind of visual thing or maybe it's even a nonvisual thing, but it has to
have some kind of output in order for you to verify and you never look at the code then there's rage
coding that's when vibe coding or this kind of coding goes poorly and then you're just asking it to do
something this happened all the time with cloud 3.5 and 3.7 you'd ask it to do something it would
spin its wheels for a while and then it'd be like you know what the solution is we're deleting all
those tests we're going to get the test to work by deleting them or we're just going to we're
going to like remove the file the file's not there now it works and that's rage coding because
it is so frustrating it's like gaslighting you it's like oh the test works I know it doesn't
Because you just, you literally just output like it doesn't work.
Anyways, the middle is guide coding.
Guide coding is when you're pair programming with it.
You really need to guide it where it's going and it goes well.
So we've all been there, except for Cloud 4.
Cloud 4, it's amazing, folks.
It's amazing.
Speaking of that, I have to make a pitch to you.
I don't, I'm actually, I don't believe that I need to make this pitch because I believe
that it is the inevitable outcome of this.
arc of like our career and life journeys and and i think i mentioned this to you like last time i
chatted but uh i think inevitably you and i are going to have to uh build uh a small large language
model uh like we're going to like it is the next step in our path down here which is like
we're going to we're going to build one and we're going to do it live from the podcast
I mean, sure.
There's nothing we can't do, folks.
You know why?
We got Cloud 4.
You know what?
It doesn't matter what we need to do.
We can do it now.
Like, what did I?
I'm not sure how closely you looked at that link that I sent you.
But one of the things I've been trying to do is find real world thrust examples in the wild.
And then, well, anyways, we'll just stop there.
That's what I'm trying to do.
GitHub search, infamously terrible.
Well, we'll make this very long story short.
But I spent the whole weekend.
I was up till 630 on Saturday
because I got back from a late 5K at 930
didn't go very well
had a couple drinks with my buddy
and then we got back at one
and I had this idea as like you know what
I bet I could get some Claude4 scaffolding
around GitHub search
and find you know basically the top repositories
ranked by GitHub stars or whatever
and long story short
you know come Monday
I've got like three different versions of this
it's absolutely beautiful
like I don't know anything about GitHub search as API
I know all about the rate limits now.
We got $5,000 an hour if you're doing metadata on the repository.
We got 10 a minute if you're doing code search.
It's all garbage.
And you know what?
You can just start like, you know, scraping that stuff by hand.
But the point being is now link in the show notes.
Another vived project.
We've got 2,000 repositories.
Do you know what the number one repository?
I mean, if you looked at it, you know.
But do you know what the number?
I looked at it briefly.
I don't remember.
The number one with 92,000 GitHub Stars.
This is just the one that it found.
is Pie Torch.
After that, it is
phase?
F-A-I-S,
which comes from Facebook
research.
Then, actually, wait,
this is the
thrust colon-colon one.
Did it update?
Drop the Facebook
research one
because that was looking
for thrust without the
colon-colon.
And it has references
to thrust
in mark-down files and whatever,
but it doesn't actually use it
anymore.
So anyways, number two,
for actually using it
is DMLC-S-X-E.
G boost, and then number three, which is what I've been working on.
Wait, wait, wait.
I'm supposed to be looking at the thrust colon colon, colon, or the...
Thrust colon, because that one is more accurate in terms of...
Because thrust by itself comes up with like a lot of rocket or airplane, anything that involves
the verb actually like thrust, takeoff, et cetera, will come up as a repository.
Wait, wait, wait, wait, there's a bug here because when you go to it by default, it shows
data set thrust colon colon, but the second result is the Facebook one.
Yes, I noticed that.
So yeah, it says thrust colon colon, but doesn't actually load the correct data set.
So you need to toggle on, toggle off.
Listen, it worked for my purposes.
I'll get Claude to fix it.
You know, I'll get, Claude was doing a lot of work this weekend for me.
We've got to give them a break.
It's not perfect, but that's okay.
We love you, Claude.
Anyways, number three is paddle paddle, which is a parallel distributed deep learning
framework. I have a theory about some of these, Connor. I have a theory about the one below
cutlass. The one below cutlass, which is... We'll see if you... Oh, yeah, yeah, yeah. But that's definitely
correct. It's gonna... Wait, okay. It's called fly-by-wire sim slash aircraft. I just explained,
though, that yes, that if you have like an airplane, you know, rocket ship one and they use the
verb thrust, guaranteed the... I already looked it up. The namespace is auto-thrust.
which has functions such as leg filter, washout filter, lead leg filter, time since condition.
So it's clearly like some kind of engine thing.
So it's not perfect.
It's not perfect.
But you know what?
It's better than GitHub.
I'm using GitHub, but GitHub itself is broken.
Anyways, the point being, it's fantastic.
So one of these I knew about, which is Uber's AresDB.
Yeah.
I've known about that for a couple years that they use, that they use thrust.
It's a real-time like analytics storage and a query engine.
It's pretty cool.
Yeah, I think you brought it up once on the podcast that Uber had their advanced technology group and they were making use of thrust.
Yeah.
But now we got a top 2000 list.
And it's still not perfect.
I haven't gone through and taken this all the way.
But it was good enough to get me some repos that I had never heard of.
And maybe we'll talk about the example that I'm working on right now is a.
rank two array so a matrix and you want to do row wise mode computation so most frequent element
row wise interesting problem we'll save it for another date uh we've only got a couple minutes left
and then we're going to be interviewing sean so i will rattle off my podcasts actually i probably
need my phone for this and uh and then we will be back afterwards because right now i think
this is only going to be like 15 minutes.
The podcast, if you are
Claude Pilled or AI-pilled like me,
and listen, folks, we all should be,
is R.
Actually, I should go in order.
The Latent Space podcast is probably the best.
It's called the AI Engineer podcast.
I discovered them because they had Chris Latner on.
Anyways, that's a great.
You should also listen to the big technology podcast.
I don't actually know
the guys are. I think they had someone on, and YouTube recommended to me, and so I added it to the
Q. Cheeky Pint, which is hosted by John Collison, who I believe is the brother of Patrick Collison,
who I believe is the CEO of Stripe, which is the payment company. I think he has some like
C, insert O role at Stripe, but I don't actually know. So that's another one. The Dwar Keshe
podcast, apparently this has been like a massive podcast that I just had never heard about, and the
Open AI podcast, they have a podcast now.
Obviously, they're going to be pretty biased.
Training data, that's a Sequoia Capital
podcast. They've had some bangers.
And then I just added
the,
it was like the A16Z
or something. Yeah, A16Z.
This one might
be leaving the queue because
they're another venture capital, but they're a little
bit too venture capital for me, you know.
I don't actually care about, I don't actually care about
venture capital and, you know,
private equity investments. I care
about the, honestly,
latent space is the best.
Anything close to latent space.
So anyways, there's a list of podcast, folks.
It's an amazing time to be alive.
And my life is crazier and busier than it's ever been.
I'm talking and traveling more than I ever have.
And I'm also writing more code than I have the past two to three years.
Like, in like, maybe I'm spending like a tenth of the
the time writing code and producing
like the output that I used to when I was like
20 and writing code like 12 hours day.
Yep.
It's because you're not writing code anymore.
You're managing AIs that write the code and you step in every once in a while.
I'm an angry, I'm an angry AI manager.
You're a happy one.
Yeah, you got to switch from rage coding to guide coding or vibe coding.
I do a lot of both of the first two, not too much of the rage coding.
Anyways, we've got to hop over.
We're going to be interviewing Sean.
so you got to wait a week.
Well, actually, I guess you have to wait a week plus 10 minutes or half a long.
Bryce and I record after this.
But, yes, a week from now, you'll be hearing from Sean.
I got to say, I got to say, it's a pity that Sean has left.
Sean is the only person I know of his generation who is such a forward-looking take on, like, AI in the software space.
So yes, let's pause here.
So what's happened now?
let me let me go get the numbers on this um
ADSP so we you listener currently are listening to oh my god it's and that is great
we didn't even plan that maybe I'm okay let me pause let me pause let me pause I'm too
excited here folks so um you are currently listening to episode 249 you were just listening
to us before the record scratch that you heard a minute ago before talking with Sean then we
then we had a chat
an interview with Sean
it's going to be split up
into three parts
it is I mean
I'm not sure if it tops
I don't think anything can top
the Steve Jobs
you know
was it a Porsche
whatever car it was
I think it was a Porsche
I thought it was a Porsche yeah
nothing can top that story
but this is
it's fantastic
you know I think the first episode
it's going to be
mostly about Rust
and then a little bit about AI
the second episode's going to be
about AI.
And then the third episode is going to be about, like, the societal impacts and what we
have to look forward to and not look forward to.
It is fantastic, folks.
I'm excited for you.
And the reason I was saying is this is super exciting is because that's going to be episode
250, 51, and 52.
So I might take, what do you think?
Should we release the societal one?
Because that, I mean, technically it would be out of order, but it would be kind of great
if our 250th episode is, it would be kind of great if our 250th episode is just getting Sean's
thoughts on the societal implications of what's happening right now.
I think, I think it'd be great.
It's kind of a little bit out of order because it's going to start with me asking the
question about whether his book, because he mentions he's working on the book,
Better Code, and he makes an offhand remark whether, you know, it'll be necessary now in the age
of AI, TBD.
Anyway, so I filed this question in the back of my head.
And then, yeah, just listening to his answer is, it's just very thought-provocative.
And anyways, yes.
It's interesting because I think his take was that he had a more optimistic take on the adoption.
I'm
maybe I'm just jaded
after 10 or 15 years
of being in the industry
but I feel like
change takes time
and I feel like it won't be
as rapid as
people think
because there will be
lots of resistance
I don't know
like I'm starting to see
now
I think people are understanding
that it
for certain tasks
is just like
it's just rocket fuel
like I was talking to a guy
in a meeting on Monday
who had bought
what do they call it
like one of those
etching
sketch kind of ink things where it's a low energy panel and and anyways had had like promised
his wife a couple years ago that he was going to make a weather app out of it or whatever and anyways
kind of it only gotten it half working and then boom he sick he sicked clot on it
clod went and learned all the APIs he was like oh make sure you do it for free got a bunch
of free anyways it's up and running i don't think it's mounted on his wall yet but uh anyway
the point being and and just listen it's just so refreshing i i i didn't realize sean was
using it as rocket fuel. I knew he used it, but I didn't know to that extent. And anyways,
we should stop spoiling the upcoming three-part interview with Sean. Yes, and you said you wanted
to announce, which I agree, this is where we should do it, because by the time we mentioned
in that episode, we're probably already, already in Europe at that point, we have, it's not
a road trip per se, because no cars are being rented. But it is, hey, we can change that.
No, no, no, no, we're not driving to, we're not driving to Collinsburg.
There's a ferry.
There's a ferry that you can take.
I'm pretty sure there's a bridge you can take from, uh, oh, I mean, what ferry you're talking
about?
You can't take a talk, uh, there's a ferry from Copenhagen to Oslo.
Oh, that actually wouldn't be bad.
Yeah.
It does take like eight hours.
That's all right.
If you take the night fairy, when, uh, Shima and I, uh, I don't know if I've ever said
on the podcast, I know what, have I?
Anyways, I proposed to Shima in Norway.
and we did a big trip while we were there
and at one point we were in Helsinki in Finland
and we needed to go to Stockholm.
I had mapped out the driving instructions
but then when we pulled into Helsinki
there's a ton of what looked like cruises
and then I was like I bet you can get cars on those
sure enough a bunch of the cruises
are just like overnight one shot things
and then you skip and save like a ton of time
and we didn't know what kind of cruise it was going to be
if it was just going to be like a boring fair
or something else.
But anyways, it ended up being an absolute blast
because it's kind of like,
it's like, I've never been on a cruise.
And they've got like little...
You've never been on a cruise?
Never been on a cruise?
And I think this one is not like the crazy Disneyland,
it's a pretty old cruise,
but it still had like a bunch of fun activities
and pubs and restaurants and karaoke and whatnot.
Anyways, I think the World Cup was on at the time
and we ended up watching a Netherlands game.
I can't remember if they won or lost,
but it was a great time.
Anyways, we are going to be,
in Europe, Bryce can explain the details.
We will be in our hars, near Copenhagen.
You think it's our hars?
I think it's Arhus.
Arhus.
Arhus.
In Denmark, near Copenhagen.
One of these.
We'll be in a town.
The town actually where Biana went to school, we learned.
We learned that a ton of programming language icons are from this little town in Denmark.
And there was this article that I sent.
kind of the other day about the Arhars
Mafia, the Arhuis programming
language mafia. What's crazy is I was
watching a talk today by Casey
Muratori and it was called
the Big Oops as a play on
like object-oriented programming
and it mentions, I'm only like
10 minutes into the talk, but it mentions it like the
seven minute mark that
Bjarna Strewstrup was learning
simula from the inventor of simula
whose name I will not recall
and it literally says
Arhus on the slide. I'm like, what are the
I guess it's one of those things where probably I've seen a talk where that city name is on a slide and you just don't notice it because you don't know where it is.
But I wouldn't be surprised if actually this is the first time I've ever seen it in a slide deck and it just happens to be a month before I'm going to the city.
Well, and so we are going to the place where it started, the place where presumably C++ start.
I don't know.
He probably was.
I'll finish watching this talk and I'll get back to you listener.
It probably was because if that's where he learned simula.
Yeah, you're right.
If it didn't start, you know, typing-wise,
it definitely started Genesis idea-wise.
Because I didn't know he went to Bell Labs,
which is, I think, where he started working on it.
Maybe not, though.
I think maybe it was his dissertation.
We'll send him an email.
We'll send him an email.
We'll ask him for Ar-Hars recommendation.
Ar-Hus.
We'll ask him how to pronounce the name.
But we will be there for the Roku Developer Summit,
which is part of Roku does this, like,
or this Roku Engineering Symposium.
It's an engineering symposium.
They do this thing, and they bring in all their developers,
but this part of the engineering symposium is open to the public,
and Connor and I will be giving a talk,
and people are welcome to come.
There's a website.
The link will be in the show notes with the details.
And then the next day, we will be at the Copenhagen C++ Meetup,
where we will also be giving talks.
and then the day after that
and all of this is going to be
September I want to say
18th 19th
18th yeah that's correct I mean I guess that
we should get the actual dates right
it seems like it is actually correct
18th is a Thursday that's Roku day as in my calendar
and then 19th is Copenhagen meetup
and then from there we go to Norway
to NDC Tech Town where if you
dear listener want to learn
from the experts
how to use
GPUs, how to write Kuta
programs. We have not one but
two full day workshops
at NDC Tech Town.
On
September 22nd,
day one, we will teach you
how to write
modern Kuta C++.
That'll be me and Connor.
We have one of our colleagues, Ashwin,
who will also be joining us for September 23rd,
when we'll be doing a full day workshop
up on Kuta Python. So you can come to IndyC Techtown, sign up for the tutorials, and learn
how to be a GPU wizard in two days. You can come to one. You can come to both. I think they're
both going to be great. We've got amazing content. Both workshops are very interactive. You're
going to do a ton of exercises, but they're both no install. So we're going to do everything
in this nice little web-based UI where you'll be connected to
some GPU server that's on somebody's cloud somewhere, and you're going to get to write some code
hands-on. It's a great course. We've taught both of them before a couple times. We've got great
feedback, and I'd love to have a nice big audience for this. So please come join us. We'll have
a ton of fun. And then on September 24th, I'll be giving a talk at IndyC. Techtown. And then I'm
going to go home to sleep for a long period of time because that's like four or five things
one after another so and uh there'll be some there'll be some podcast recording probably and uh
we should have ADSP stickers I was just going to say Bryce sometimes we are it is crazy
I've said this the last time we recorded too where you were saying one thing and I was like
that's literally what was in my brain not only was in my brain they're in my hand they weren't
just sitting on my desk I had them in my hand ready to say and not
only will you be able to chat with us.
I'm sure there'll be some socials and stuff outside of the training.
But for the first time ever, thanks to Bryce, it's actually been on my list of things to do for
like three years.
I think we brought it up probably 10 times on this podcast.
Multiple times.
It's come up so many times.
But the problem is I actually did talk to a sticker guy.
And anyways, but we have, I think there's nine here.
One, two, three, four, five, six, seven, eight, nine, ten, eleven.
All right.
They said they sent me 10, or maybe it was 12.
But so I have 11 at least.
I'm imagining Bryce is going to order more, but...
Yeah, I'm going to do a full production.
But if he doesn't, folks, I will be bringing 11 stickers.
I have the 12th one on my laptop right now.
So if Bryce fails to order these or they don't show up in time,
there are going to be only 11.
And I guess if there's 11, we'll have to, like, come up with some kind of quiz.
And that's...
No, no, no, no, I already ordered them.
They'll definitely be here.
However, those are like the pre-production ones,
so maybe we can give away,
we can give away those ones specifically.
Oh, yeah, yeah.
I just have to remember to find the, right.
They're probably going to look the same, though, no?
Or did you make a slight change?
The thing is, we did multiple different...
You said you'd be able to tell, though,
because you had it on file.
Bryce ordered a bunch of different styles and then sent me one,
and then I was like,
ah, it looks pretty good.
It's a little bit dark for the red,
but you know what this is why we didn't have stickers for four years because we're like
oh you know it needs to be right but uh it's close enough but then bryce was like oh so what's the
number on the package and i was like i tossed that like the second i got them man or recycled it
it would have been like a last little sticker though but let's see i i i think i'm getting
pretty close to i'm trying to find i got multiple order numbers here too many order numbers
yeah anyways folks we've got it's going to be a great time because when i feel like there
was uh go back to the episode lists oh yeah oh yeah that's right we were both at c plus plus under the sea
so we kind of do have like an annual we had the the slovenia road trip in 2023 and then which will be
probably hard to ever be topped then we had the c plus plus under the sea uh last year in 2024
and now we're going to be at nDC tech town in norway um in september uh this time we're
going to have stickers and it's going to be a oh yeah and that's the other thing is that we are
working towards open sourcing uh the library that i've been working on for the last while yeah i
don't want to say anything about it i think it's safe enough to say it's a actually i won't say
anything but anyways we're working towards open sourcing it on the mvs um github and maybe by the
time this is actually airing in a week or two uh it might already be out there and so hopefully i'll
be able to talk about it in the presentations that I'm giving.
You think you can say the name?
I don't want to, because the name could change at like the 11th hour.
And so I don't want to say that yet.
But I will say, actually, let me quickly share my screen before you go, Bryce.
It is because I had, well, I have to be very vague about this.
I had a emoji as the logo of the library, but then there are some copyright rules that you
could run into by using the.
emoji itself because this emoji
I preferred the Microsoft one
and so in fact
if I go get you the
what do you call it emoji
and then we can't say what emoji
because that gives away the name of the library
but if we go
But the name of the library is something that is an emoji
yeah so look if you go to this one
that's clearly not the emoji rendering I want
because for folks that know
there's like a bajillion different
that that is an ugly version
of the thing that the library is.
The one that I like
is the Microsoft
2021 flat.
That is a much more attractive
one of the thing.
But unfortunately,
you can't use these emojis
as like logos and stuff.
Anyways, Michael, my boss was like,
yeah, you got to change the logo.
So what did I do?
I just went to Gemini
and then said, hey,
I want the same?
It's not a logo.
It's not a logo.
It's a graphic signifier.
And I said,
I really like this.
emoji
but I
want it to be different
so I said
just make it
with it
oh shit
you're gonna have to
bleep that one
just make it
slightly different
and sure
I'll bleep that out
and I mean
actually I did this
with both Gemini 2.5
I think Flash
and then
chat GPT
5 and the chat
GPT one
I wonder if I can find
it in history
it is terrible
Yeah, I know it's so bad
Let's see
Dun dun dun
Oh yeah
Yeah yeah I do have it
Can you uh
I've lost
This one
This is so wait
I've lost oh yeah
It looks like a
It looks like a haunted version
Of the thing
Yeah
And that's the thing is
It's they both
They both technically like
For what I requested
They both met the goal
But just like
Look at this one
it's yeah like it's haunted because it's not using white it looks dead and uh whereas this one
this one looks cute as hell like it does it makes me but it also doesn't it doesn't really
look like that's all right thing it just people can tell what category of thing this is correct
people can tell what category of thing that's all we need that's all we need that's all we need
although it looks like it looks like a thing that you would have at an
I'm going to, we got a, we got a, I'm going to have to peep, beep that out because that's too,
that's too much information.
Anyways, folks, I'm going to stop presenting, um, look forward to the Sean parent trilogy.
Or did they call it a trilogy?
Yeah, we'll call it a trilogy.
Trilogy of episodes.
Any final thoughts before you go, buddy?
Oh, no.
Well, um, yes, yes.
Oh, he does have a final thought.
I have, I have, I have the three top.
that we'll talk about on a future podcast.
Oh, yes.
And to interrupt Bryce, because he's about to mention some algorithm topics.
The reason I said, it's one, it's going to be exciting to be able to talk about the
library I've been working on.
But two, there's a bunch of algorithmic conversations that are going to come out of
open sourcing this and being able to talk about it.
Yeah.
But the thing that I was talking about, I'm not sure if it was actually on air or if
it was before air where, yeah, I was mentioning the thrust GitHub repo search and then
the algorithm that was in paddle, paddle, paddle or whatever that.
repo was. It was a, yeah, a matrix row-wise mode. There was a bunch of interesting stuff behind
that and also this algorithm called replicate. Just put that on your list of things that we
have to talk about. And yeah, I'll explain what replicate is and we can dedicate a whole.
But anyways, there's been a bunch of like, if folks are worrying what happened to our algorithm
podcast. It's now just an AI podcast. The algorithms are going to make a comeback. Your screen is
frozen. Are you still there? Bryce is back. He left. He's back. He's back. He's back.
but anyways, replicate algorithms, the three things that you have on deck to talk about.
Yes, teaser for future episodes.
These were just three notes that I wrote down like six months ago,
and every month I've been getting like an email reminder.
And so I'm going to mention them so that we'll be forced to talk about them.
The first, parallel rotate and parallel rolling reduce are related.
Some of you might remember my ramblings about rolling reduce a few years ago,
which we're going to have to talk about again because I've discovered some new use cases for rolling reduce.
I've come up with an even better algorithm for rolling reduce.
So stay tuned.
And Phineas,
and Phineas,
we are going to bring you back on to talk about the thing we were going to talk about.
He posted,
he sent us a message,
maybe he was on GitHub or something.
Anyways,
we're going to have Phineas back on.
There's much to talk about with him.
Adjacent difference is the inverse of a scan.
We've talked about that on the pod.
You have forgot him.
But we'll talk about it again.
Did we?
Did we?
Okay.
We'll talk about it again.
Christabella,
if we forget to bring up Christabella when they talk about that,
That'll be a sin of omission because his name should be mentioned.
This last one is just a, it's really more work because I need it for something in NVIDIA,
but can you shuffle with a scan?
I thought you could, but then I looked at some more.
I'm not so sure.
Maybe with an exclusive scan, but not necessarily with an inclusive scan.
Those are the three things, the three things.
So if anybody stay tuned, but in an infinity,
Anybody who's got any thoughts or ideas about those.
You know how to communicate with us on the interwebs.
Yeah.
All right, I got to go because if we don't leave for the U.S. Open now, we're not going to be on time,
and then Ramona's going to throw me off the train or something.
All right.
And watch out, folks.
I've been inspired by some AI podcasts I'm listening to.
I think we're going to start dipping our toe in the shorts world.
And that's not the shorts you wear.
That's the shorts on the inner webs.
We're going to potentially start an Instagram, ADSP, a key.
account, a TikTok ADSP account, and maybe even a YouTube ADSP account.
YouTube shorts. That's what I watch. I watch the YouTube shorts.
And yeah, so, you know, we're trying to, it's a new, it's a brave new world. We're trying to, we're trying to grow our reach.
Mixed media formats. Mixed media formats. Yeah. And the goal is that in the future, the AI is just going to do it all for us.
And just so we're going to get the ball rolling. And, yeah, anyways, we're letting Bryce go.
Enjoy the tennis. And we will chat soon.
I'll chat with you later.
Be sure to check these show notes, either in your podcast app or at ADSP the podcast.com for links to anything we mentioned in today's episode, as well as a link to a get-up discussion where you can leave thoughts, comments, and questions.
Thanks for listening.
We hope you enjoyed and have a great day.
Low quality, high quantity.
That is the tagline of our podcast.
It's not the tagline.
Our tagline is chaos with sprinkles of information.