Hey, Good Game

Hey, Good Game Trailer Bonus Episode 50 Season 1

Building a Better Chess Brain: The Story Behind Blitztactics' Fast-Paced Learning Platform

Building a Better Chess Brain: The Story Behind Blitztactics' Fast-Paced Learning PlatformBuilding a Better Chess Brain: The Story Behind Blitztactics' Fast-Paced Learning Platform

00:00
Linmiao Xu is the creator of Blitztactics.com. Lin, a software engineer and avid chess enthusiast, shares his experiences developing Blitztactics—a fast-paced online chess puzzle platform. They discuss Lin's 'blitz chess' addiction, his forays into charcoal art inspired by Dark Souls, and the creative processes during his travels in Asia.

Lin also delves into his involvement in training neural networks for Stockfish, a leading chess engine, and his professional role in developing generative image models at Playground. The conversation wraps up with Lin's future aspirations in game development and his philosophy on creativity and progression.

Check out Linmiao's Resources:

https://blitztactics.com/
https://lichess.org/training
https://github.com/linrock/blitz-tactics

https://robotmoon.com/
https://github.com/linrock
https://unsplash.com/@linrock

https://www.instagram.com/linrock/
https://www.linkedin.com/in/linmiao-xu/

--

  • (00:00) - Introduction to the Podcast and Guest
  • (01:25) - Linmiao Xu's Favorite Games
  • (03:39) - Art and Travel Inspirations
  • (06:31) - Biohacking Vision
  • (10:04) - The Birth of Blitztactics
  • (12:52) - Challenges and Popularity of Blitztactics
  • (15:33) - Community and Traffic Insights
  • (26:03) - Stockfish and AI Involvement
  • (31:44) - Future Projects and Closing Remarks

--

Check out our brainy games:

Sumplete - https://sumplete.com
Squeezy - https://imsqueezy.com/
Kakuro Conquest - https://kakuroconquest.com
Mathler - https://mathler.com
Crosswordle - https://crosswordle.com
Sudoku Conquest - https://sudokuconquest.com
Hitori Conquest - https://hitoriconquest.com
Wordga - https://wordga.com

Creators & Guests

Host
Aaron Kardell
Husband. Father. Founder & CEO @HomeSpotter; now working to simplify real estate w/ our acquirer @GetLWolf. Striving to act justly, love mercy, and walk humbly.

What is Hey, Good Game?

Hey, Good Game explores the stories behind your favorite brainy games. Each week, we interview game creators and dig into what it takes to build a successful indie game, how to monetize, and how to get traction.

Aaron Kardell: [00:00:00] Do you have any, you said you had a little bit of an addiction to blitz chess. Is that behind you or do you find it comes back and in waves or what does that look like for you?

Linmiao Xu: Oh man, like every year or so, I feel like there's a point where

Nate Kadlac: Welcome to the Hey It Good Game podcast, where we chat with the creators of your favorite games that you secretly play in the cracks of your day.

Aaron Kardell: I'm Aaron Kardel and today we're excited to speak with Linmiao Xu the creator of Blitztactics. Lin is a software engineer by trade and has worked on all kinds of projects ranging from fashion platforms to AI driven creative tools.

He's also an avid artist and chess enthusiast with Blitztactics being his signature game. Blitztactics is an open source online chess platform focused on fast paced tactical puzzles that is designed to help players [00:01:00] sharpen their chess skills in Blitz style timeframes. Filled with different game modes and puzzles sourced from WeChess.

org, it helps players analyze faster, become more familiar with patterns and tactics, and improve their techniques via different scenarios. Lin, we're thrilled you're here.

Linmiao Xu: Thank you for having me.

Aaron Kardell: Lin, as you probably know, we always like to start out by asking a hard hitting question. What's your favorite game to play?

Linmiao Xu: Oh man, lately I've been playing this, I guess you call it a collaborative online factory simulation game where, you're one of a team of scientists working on building a robot that battles other robots in tournaments every couple months. And you have this limited resource called cores which your team can allocate towards researching upgrades that make this robot stronger or faster.

And these experiments, most of the time they, don't work out, but the few times that they do, it's, very exciting because you can see that the robots getting [00:02:00] stronger, the ELO is going up in the graphs and the tournament results become, stronger over time. and while it's, so actually that's not really a traditional kind of game, but it's what I just mentioned is how I see chess engine development is a stockfish development is a hobby of mine in my spare time recently.

And I gamified it for myself so that. It's been very fun and addictive for me in the past couple of years.

Aaron Kardell: Just, iteratively keep making the robot better, huh?

Linmiao Xu: Yeah, pretty much. Just, stronger and faster. And its computer chest is still, going up. Still getting better.

Aaron Kardell: That's right. That original game, does it, have a name?

Linmiao Xu: The original game? Oh, it's, oh, Stockfish.

Aaron Kardell: Oh, I see. So you were saying your improvement of Stockfish is the game that you like to play. Is that, I was a little slow there.

Linmiao Xu: Oh, that's right. Yeah, that's, that's what I've been doing the past year or two. That's been my primary game. But if we're, talking about traditional games, [00:03:00] Other than that, recent favorites of mine have been like Baldur's Gate 3.

Liked it a lot. Played a lot of Elden Ring and Bloodborne. So I've been really into the really difficult first person or like Really difficult kind of games.

Aaron Kardell: Got it. I, think you've got one of our most unique answers so far on the original answer to the favorite game. Love the, creativity and kind of spoken like a true developer, I'd say.

So that's, awesome. So maybe just going through a little bit of your background before we, dive into Blitz, 2. Tactics. Did I read right? You made art of Dark Souls three and you love that challenge and that you like difficult games in general. Is that right?

Linmiao Xu: Yeah, that's right.

Like I, I made some art for Dark Souls or Bloodborne. One of them was a charcoal drawing. It was actually my first charcoal drawing. I was, so I'm not an [00:04:00] artist by trade or anything. It's more like a hobby that I do maybe a week or two every year, like on rare occasions. And then one time I was just visiting parents for, winter holidays.

And I was sketching something, trying to darken in some area of shadow for this other sketch I was working on. And then it couldn't, I couldn't get it very dark. So I asked my sister, how do you get this even darker? And then she suggested using charcoal because with charcoal, you can get some really dark images.

And then somehow I connected that to dark souls where. There's like a, Ash and Cole is like a theme of the game and I just took it upon myself as a challenge to draw the cover Art of Dark Souls with charcoal. And that was a fun, difficult challenge and then zone too kinda difficult, like the game itself to try to make that look good.

Aaron Kardell: That's cool. That's great. I think we share fin an affinity for travel. What memory or experiences around travel most, stick out for you? [00:05:00]

Linmiao Xu: Oh man. So one of them actually. regarding Blitztactics, some of the game modes I actually built while I was traveling, and so I worked in tech for a while, based in San Francisco, and at some point I just wanted to get out of tech.

I was tired of being there, just wanted to travel and then start fresh and not code or anything. So I was traveling in Asia. I was in Taiwan at the time for a couple of months. So I was just like cruising around having fun and it was great. It was amazing. But then after about two months, I felt like something was missing.

And that missing piece was creation. Just like creating something, feeling productive about that. And that's what led me to. Pick up Blitztactics, like a second version of it and create the, countdown, speed run, those game modes were, game modes I created around that time when I was traveling in Asia.

Aaron Kardell: That's great. It's, I found too, that just from an entrepreneurial perspective and a product development perspective, [00:06:00] some of my, Best ideas over time have come just by mixing up environments and being in a different spot. And it's, yeah, really, I don't know, it can be a bit of a, an unlock, even if you're trying to decompress in other ways, like sometimes your best ideas come out that way, right?

Linmiao Xu: Yeah, for sure. That's how creativity strikes. I find it's a By changing up the environment, these ideas are connected in unexpected ways. And that's how a lot of new ideas come about.

Aaron Kardell: For sure. And, also intriguing from our research, it seems like you've been, biohacking your vision, so to speak, if I'm categorizing that right.

How's that journey gone for you?

Linmiao Xu: Oh, that's right. So, the way that journey went about was, back when I was in San Francisco years ago. I realized I needed new glasses because my glasses were getting weak. My things were still blurry when I was looking through them. [00:07:00] And I realized that this just kept happening.

every couple of years I would just need a stronger prescription. And I felt like that was just the way things went. And then every automatist you will talk to you will tell you that's just how it is. there's no way that you can really change that cycle. So then I took it upon myself to experiment with ideas around there.

Cause I had read some anecdotes online, like no real science, but anecdotes of people saying that they were trying these other ways to avoid having to get stronger prescriptions by I guess, like practicing, looking at things on the edge of blur and pushing your eyes that way. So I was very skeptical because, everyone had been telling me, that's just not a thing, like you can't do that.

So then I figured, I'll, try it. I might as well try it. Cause I was at a point where I could not look at my computer screen without glasses. Like I could not read the text because it was blurry. So that's, that was my starting point. And I didn't want to just keep depending on glasses forever.

And. my eyes get dry if I [00:08:00] wear contacts. So didn't want to do that. LASIK sounds cool, but it's permanent. what if it doesn't go well? So I decided to try it out, try out not wearing glasses and trying out, just like looking at things that are just on the edge of blur to do that long term and see how that went.

And I found that actually I was able to reverse some amount of my nearsightedness. Like I'm not 20, 20, but I got to the point where I went to the DMV. Okay. To renew my license. And then they had me do the vision test without my glasses and I passed, and now I'm able to also read texts on the screen without glasses.

So it's, my vision has improved a bit, like a couple, like one diopter maybe in both eyes, but not 20, 20 hit, hit a plateau there. And I guess I have not pushed myself to try to go further than, where I am. But it's better than when I started for sure.

Aaron Kardell: That's really interesting to me. I've always, I've had the good fortune and I'd have to wear [00:09:00] glasses thus far in my life.

I'm starting to get closer to the point where it feels like I might need them, but everybody that I know that does have them or has contacts or whatever it I've noticed that same pattern of it always seems like things are getting worse and you need. More and more corrective action. And it just seems like it's a self perpetuating thing.

So I've always been skeptical. is, that, is just getting updated prescriptions or whatever, is it, just making things worse. And the first person that I've, heard that has done something like this, it's a pretty interesting takeaway and cool results.

Linmiao Xu: Yeah. It worked for me.

At least one of my friends has also, was also interested and he tried it out for a while. And he's seen some. Very slight improvements too. So it's, I would say it's, within possibility. It's definitely not, it's not impossible to reverse declining vision, but. But if you talk to any apometrist, you're always going to say it's not possible.

Aaron Kardell: That's really cool. [00:10:00] jumping ahead. So why did you create Blitztactics?

Linmiao Xu: Oh, so this was back in late 2015. I was, I had an addiction and this addiction was. Blitz chess. And so I was playing a lot and I was playing a lot of blitz like hours per day. And I got to a point where I was just not getting any better just from playing.

Like I was playing blitz, I was playing on lead chess. I think my rating was like stuck around 1800 and, I played blitz. I would just blunder pieces and lose on time and just keep making the same mistakes over and over. And then I figured, I just want to get better at the game, but obviously it's not working to just like play.

A lot and analyze a bit after the game. So I looked up different ways to improve at chess. And one way that I guess a lot of top players when they were young, they would read books and get into chess that way. So I tried that, but I found that reading chess books was not really my thing. It's a, I find it a bit dry and [00:11:00] hard to get anything from it.

So I wanted to find a fun way to, to improve. And I figured the drilling patterns into. into memory is one possible way to improve just by getting better at quick pattern recognition so that these can be implemented in actual play. So I tried to find a website where I could do this. There's definitely chess puzzle websites online, but None of them had the kind of experience that I was really hoping for, like a fun, fast paced experience that I could use to train.

And because it didn't exist, I had to set about creating it on my own. I figured there was a chance to create something new here. So that's how that started.

Aaron Kardell: Necessity is the mother of invention, right? Do you have any, you said you had a little bit of an addiction to blitz chess. Is that behind you or do you find it comes back in, in waves or what does that look like for you?

Linmiao Xu: Oh man, like every year or so I [00:12:00] feel like there's a point where it's behind me, where I'm just retired from playing chess, but always some wave hits me at some point where I get back into it. But then, with that said, back at the time I had this goal to hit a 2, 000 Blitz rating on Lichess. And after training a bunch of tactics for a while, I did hit that goal, and after that I felt like it was a good time to just retire from rated Blitz chess.

So that part I I stopped. I have not really played Blitz since then, since I reached that milestone. But definitely, every year there's, I do play occasionally. But for the most part, I'd say I transformed my Chess playing addiction into, engine development. Like I work on Stockfish now.

That's my new addiction that I personally find to be more gratifying than playing way too much and. Yeah.

Aaron Kardell: Seems like effort well placed. As you were creating, Blitztactics, what was the most difficult part of that?

Linmiao Xu: The most difficult part? I would say [00:13:00] there's a couple of elements that needed to be there for it to work.

Like for one is the puzzles themselves, like getting access to good puzzles. And for another it's the gameplay elements. I think the data part was less difficult because Leach has had these puzzles already and I was able to download a bunch of them and then filter them down to like reasonable difficulties for Blitz type puzzles.

And then I guess the gameplay was the harder part for me to work on. Because back then, I was much earlier in my career, I would say it was more difficult for me for it to design and create a good experience, especially from scratch. So I'd say I spent a lot more time just trying to dial in the gameplay elements for the initial version of the site.

And yeah, I guess that, would say that was the harder part is to try to nail that down and I still think there's room for improvement, but a lot of my time went into that.

Aaron Kardell: Yeah. And amongst all the game modes in the game, which have you seen are most, used?

Linmiao Xu: the most. Popular one is [00:14:00] haste, which is the one where you start with a three minute timer and then you just try to rack up the highest score you can.

And if you make a mistake then you lose time, that kind of thing. So I'd say that's the, one that people play by far the most compared to the other game modes.

Aaron Kardell: Do you have a big puzzle database that you're working off of? Like with, starting points or how does that work? Part of the site work.

Linmiao Xu: So the puzzles are originally sourced from the chess, the first version of Leach has puzzles, which at the time I think there was like 120, 000 puzzles or something like that. And then, so I downloaded all of them and back then, some of them had problems in them as in you could make a perfectly valid move.

That's a valid solution, but it would think it's wrong for, some reason. So I had to try to deal with that in some way. Yeah. And because this metadata came with user votes, that was a quick heuristic to filter out the bad puzzles because ones that were bad would get downvoted. [00:15:00] And then ones that are good are more likely to be upvoted.

And then I was able to roughly sort the puzzles by rating because the puzzles have their own, innate rating points based on how difficult they are. So that was a quick way to both filter out invalid puzzles and also get the easier puzzles. And a lot of them, I actually went through myself to manually hand pick, because I figured that was, that is an important part of the process is to.

manually curate to some extent, but for the most part, it's, most of the puzzles are automated by now.

Aaron Kardell: Makes sense. And have you had any kind of big bursts of traffic either around the COVID 19 pandemic or under, or with Queen's Gambit taking off, did either of those have a big impact on visits to your site?

Linmiao Xu: Oh, Queen's Blizz Tactics at the time. So any spikes there, I totally would not have been aware of. But before that, there was a time [00:16:00] when actually Lichess was recommending Blitztactics when people were just asking, Hey, we need building Puzzle Rush and all that. So around then there were occasional bursts of traffic because I guess someone would post it on, Reddit or, something like that.

Like someone would post the site somewhere and get occasionally, occasional spikes. But yeah, around Queen's Gambit, I was just totally not paying attention. I have not actively worked on the site for a while. So if there was a spike, it might've happened, but I just Was not aware.

Aaron Kardell: Fair enough. And I could posit a guess based on just some of the other things that you've shared, but it looks like you haven't ever put ads on the site or chosen to, to monetize in that way.

Is there a reason for that?

Linmiao Xu: Yeah, for me, it's the emphasis is on the gameplay and just having fun learning. I feel like ads would. Detract from that and probably not even make enough money for me to really care about. So that's one reason to not do ads. And yeah, I've never really thought about monetization either, because for now, it's just a site project is open source.

[00:17:00] And if it goes down, then, I don't have to care so much about it, but then if people are paying for it, then, they might expect with good reason that there's some level of reliability and, but that all said, I have not paid that much attention to the site for. a few years now and it's just still working.

So it's, even though I'm not spending much time on it, it still works. So I have not had to worry about it, but if I did have paying users or such, then it would be something I would need to think about out of, responsibility to the users, I'd say.

Aaron Kardell: How did you first get out the word on the site?

how did people discover it?

Linmiao Xu: Yeah. So the one place I ever promoted it was on a lead chess forum post. At that point I had created. The first version of the website, which was just the repetition mode. And then I made a forum post. I was like, Hey, I made this fast paced chess puzzle sites. I, what do you guys think?

And. That was it. [00:18:00] And I actually got a lot of feedback back then there was a, I think there was a surprising amount of positive feedback around this concept. So that definitely validated that there was a gap in the market for this kind of thing. And then, this was back in early 2016. So a couple of years later.

I guess chess. com made Puzzle Rush and then soon after Lichess made their own fast paced game modes. yeah, the one place I ever posted it was on Lichess on a forum post way back when.

Aaron Kardell: Got it. And you may not be tracking this closely, but do you have a sense of how many people are playing on your site every month?

Linmiao Xu: Yeah, so these days it's a lot less than before. Like I was actually, I put up the stats just a few days ago. I have not looked at it in a very long time. Like nowadays it's just like a couple thousand per month is where it's at now, but I'm actually surprised it's even at that point because I just don't even look at the site anymore.

But then, back in the [00:19:00] day it was, the peak was closer to 10, 20, 000 per month in that range. So it was never huge or anything, but it was, a reasonable amount of traffic given that I never tried promoting it or anything like that.

Aaron Kardell: Looks like you've got a discord channel. Do you still spend much time in there?

Linmiao Xu: Oh, the discord channel? Not recently. it was, I put it on there as an experiment to try to see how, see if it could grow into a community. This was years ago. And I would actually say that bliss tactics in this current form is really just a, collection of prototypes. Like nothing is fully flushed out.

They're just game concepts that have potential to become something more. Yeah, it's not something I actively manage or grow or anything, but I think when the day comes that I spend more time on it again. Like maybe, if it grows bigger, I'll actively spend more time on it.

Aaron Kardell: maybe as you're continuing to iterate on Stockfish, maybe that'll spark some other ideas for things you want to come back to on [00:20:00] this.

Have you had any, maybe surprising opportunities come your way from building out Blitztactics, things you wouldn't have anticipated otherwise?

Linmiao Xu: Surprising opportunities? not really. I guess I would say this podcast interview is one of them, but, I've had offers to work together, offers to buy the site, that kind of thing.

But I never really took up any of those opportunities. So it was, yeah, back when it was getting more traffic, there were a lot of random opportunities that came my way. But nothing, tremendous or anything like that.

Aaron Kardell: you may or may not know that HeyGoodGame, part of our business model is we, do Grow by building our own games and by acquiring other games.

And so out of curiosity for our listeners, we run into people all the time that get offers like why. Why was it not interesting to consider selling before in, in your case?

Linmiao Xu: if it's because I, none of the offers would be [00:21:00] any big number that I would, that would make a meaningful difference. So to me, it's valuable just to have this creative outlet basically.

cause I see it as that it's an, opportunity for me to just dabble with coding. Like I, there's times when there's days and I might spend just like you're writing part of the code for. For no reason other than trying to just practice some new code or something like that. So just having that outlet is is valuable to me.

I think it's valuable more so than small monetary amounts that might come my way. I think it's, yeah, mostly about that.

Aaron Kardell: Yeah, makes sense. Looking back, is there anything for other game creators that you might suggest, you would do differently if you were doing this again?

Linmiao Xu: I would say it depends on the motivation for creating a game.

if I were doing this again and trying to build it into something bigger, like definitely I would have put more time into it back when [00:22:00] like, fast paced puzzles were, picking up or like paying attention to Queen's Gambit and seeing if there's a traffic spike to capitalize on, because I do think that seeing those numbers is a valuable motivator to if I see like a 10 X traffic spike, then that alone could be motivating for me to invest more time into it.

So I would say if the motive is to grow. So I think if you want to grow a game into something bigger, there's definitely opportunities to pay attention to and to capitalize on for both motivation and just like marketing opportunities. So those are the opportunities to pay attention to I think are valuable, but if the motivation is just to build something for fun, then a lot of things change about how one would approach building a game.

So yeah, I would say it really depends on, the why behind the game.

Aaron Kardell: there are a couple of threads here that I want to pull on a little bit. And I think they both come back to data in some fashion and just algorithms around data. But. [00:23:00] The first is you talked a little bit about just the puzzle selection process for Blitztactics.

Any other insights that you'd share for other game creators? I know puzzle generation in general, chess or otherwise, is always a kind of a, pain point, but there's, often that's half of the fun for some of our game creators. So any interesting takeaways from your time thinking about puzzle selection?

Linmiao Xu: Yeah, for sure. So for one, the difficulty curve is very important. Because if it's meant to be like a fast paced website, then it needs to be very manageable by people playing, the game. and then there's so many possible chess puzzles that one could use that really, like if one person is working on it, then there needs to be a very easy way to filter that down to a manageable pool of candidate puzzles.

and then beyond that, I think having the ability to just try out puzzles manually to decide whether to use it or not is an [00:24:00] effective way to, to manually curate the puzzles too. So I think it's a valuable part of the process is just the content of the game itself. for example, so Duolingo is an app I use.

I've been learning Japanese for a few months, and it's, learning a language is very difficult. And it's, could be a very dry and boring process. But what they've done is they've taken, basic game concepts, and then they turned learning a language into something fun, with a very, shallow curve that's not too intimidating.

And they have these other game mechanics that they're bringing back, of course. But just that the content is a very core part of how they're successful because well, now they're a public company. So they, obviously they've been doing something very right in that regard of content selection. So yeah, I would say that manually going through and then curating is important, but there needs to be an automatic process that one goes through because there's so many possibilities and only so much time to spend on it too.

So it's really a [00:25:00] balance of. Spending the time the right way to, to choose the right puzzles to use.

Aaron Kardell: Yeah. That curation is really key in terms of the grading. Did you use anything off the shelf or did you? Build any algorithms to help determine how difficult the puzzles were.

Linmiao Xu: So initially I tried out a few different ideas and I actually, for one, I built my own puzzle generator, using Stockfish to just find possible candidate puzzles for out of games and such.

But I realized going down that path that. There was still the problem of whether the puzzles are fun or not. So I realized eventually that Lichess, because they have so much data already on the difficulty of the puzzle, and then the voting system to somewhat determine whether the puzzle was fun or valid and such, like just using that metadata to, to filter down the puzzles was a very quick way to get a reasonable selection.

So that was a very valuable shortcut in [00:26:00] getting a reasonable pool of puzzles to use for the site.

Aaron Kardell: Makes sense. Speaking of Stockfish, it's come up a couple of times so far in the podcast. Sounds like you've really enjoyed your time helping to refine the, Stockfish engine. How'd you get involved in that community?

What's your participation in that look like today? And yeah, where, do you see yourself going from here on that?

Linmiao Xu: Yeah. So I've been involved in various ways since a few years ago. I guess a few years ago, I was helping build out the distributed testing framework behind all the tests, the power of Stockfish, but in recent times, I guess almost two years ago, I got involved in trading the neural networks.

So the way I got involved there was startup company had recently pivoted into AI and we had a, an AI week where we were to spend time learning about fun AI topics and learn something new just for fun. So I just went to the Sockfish Discord and I was like, Hey, what's up guys? I'm here to learn about AI, anything I can do to, learn.

And then one of the devs was like, [00:27:00] you could try training a neural network for us. And I was like, all right, yeah, sure. let me try it out and see how that goes. and this, I didn't realize it at the time, but by that point, progress on training the neural networks, that kind of hit a plateau and stopped for four or five months.

like the lead developers couldn't really make it any stronger. And I, here I am just like showing up and trying to learn AI. And then, what could I possibly figure out? But I think just the, it just felt to me like I was playing dark souls or something, right? Like I'm playing a game. I get to a boss, I get stuck at the boss for hours and hours.

And then the boss is just saying, you gotta get better to, to beat me and move on with the rest of the game. So I saw the neural network training process as a very similar process, where it's like I gave it one data set doesn't work, make some tweaks. That doesn't work. And then have to learn and make small adjustments to the process to finally figure out.

What actually works. So [00:28:00] what ended up, requiring for me to train the first network was reading years worth of discord history, reading all the GitHub commits around the Sockfish neural networks, and just consuming all this information I could about what had been tried and what had worked and not, and eventually I was able to come up with something.

to train that first network. And because it was so difficult, I got hooked into it too. So that's how I got involved, training networks for, Sockfish.

Aaron Kardell: That's great. And, it's gotta be a, a big community working on that,

Linmiao Xu: right? it's, I think at any given point there might be one or two dozen active developers.

So there's, people that jump in and work on things for a while and take a break and some people that are much more active, more regularly. yeah, it's not a huge community. It's a very, I guess it's a very niche activity to try to build chess engines and push it to its limits, but it's, people are there [00:29:00] to, for open source and just have fun and, learn and just.

Push the frontiers of chess, engines. So it's a fun community to be a part of for the most part.

Aaron Kardell: Right on. So it sounds like small enough that you can still have a really big impact, which is cool. And then finally, just working on neural networks, it sounds like you've got some professional interest in, AI.

Are you able to talk about your. Your day job at Playground at all?

Linmiao Xu: Yeah. So at Playground, it's, the products around generative image models. So by this point, we train our models from scratch, image generation, foundation models that, take a lot of VRAM to run. So they're pretty big models that are 60, 70 gigabytes of VRAM.

So they need to run on data center GPUs, but that's, that is a lot of it is just the architecture for one and needs to work, but a lot of it is the training data too, so just like you have this. Architecture that is capable of generating images. [00:30:00] What kind of data do you give it such that it's able to generate what you prompted?

So that is where a lot of the current a lot of where my current efforts go into is Preparing the training data and working on the infrastructure that we use for training these massive models which Take dozens of high power gpus to for any one training run So it's a very, it's a very fascinating process, very expensive process, but it's fun because it's also on the edge of AI research also, where, everyone is trying to figure out how to, AI to generate hands properly or spell text correctly.

Like these are still open problems that are still being worked on. yeah, a lot of my work goes into helping with that, helping figure that, those problems out.

Aaron Kardell: Sounds like you all have been working on this, at least far longer than it's been in the public consciousness, perhaps. Are there certain aspects of image generation that Playground is maybe better at than some of [00:31:00] the, known AI implementations that are out there.

Linmiao Xu: Yeah. I'd say in the recent times, with the latest model, Playground V3, a prompt following is one of the main points where you can give it very, specific detailed prompts about something. And it's able to follow these prompts to a stronger extent than other players in the space. So that's one aspect, and it's gotten a lot better at spelling text correctly as well.

So these are both aspects that are, we're still pushing those to their limits. But yeah, one big thing was just text comprehension, using, open source language models like Llama. To comprehend the text and then guide the process of generating the image such that it matches well with what you, want it to create.

Aaron Kardell: And then, last up, perhaps just back in the context of gaming, do you see yourself building any games in the near to midterm?

Linmiao Xu: Yeah. So game development is something I daydream about quite a lot, actually. Like I [00:32:00] play with these concepts in my mind about games that I would want to build. And want to try.

So it's something I've been wanting to dabble with and occasionally I'll open up Godot and try to make something, but I'm realizing that an actual fully full fledged game engine is its own. It's learning programming from scratch. So I need to invest a lot of time and effort to really create something valuable there.

But I was also thinking, I've been thinking for the past couple of years about new concepts for Blitztactics. because when I first built Blistactics, it was new. There was nothing else that was like it out there. And then now that there is, I feel like if I were to work on it again, I need to bring something new to the table once again.

So I've been thinking about, new concepts that could bring that to the next level. So I think, one aspect that's missing is a sense of progression and advancement and just like looking forward to what's next beyond where you currently are. Basically, it's bringing some adventure game [00:33:00] concepts into the fun of playing Blitz puzzles.

I have some concepts in mind. I've been actually designing them out a bit in Figma and working on these ideas. I think one day we'll see the light of day. But, until then, it's something I'll be, I'll think about and I'll work on. Hopefully get to that sometime soon.

Aaron Kardell: good luck in, those endeavors and, yeah, as we wrap up, if you want people to find you online, Lynn, where should they look?

Linmiao Xu: Oh, I have a website, robotmoon. com, which lists a bunch of random side projects of mine. Otherwise for open source type work, I'm on GitHub. github. com slash linrock. So yeah, we'll just go with

Aaron Kardell: these. Excellent. thanks so much for your time today.

Linmiao Xu: All right. Thank you.