site banner
ArjinFerman  Tinfoil Gigachad  6d ago (text post)   469 thread views

Tinker Tuesday for July 7th, 2026

This thread is for anyone working on personal projects to share their progress, and hold themselves somewhat accountable to a group of peers.

Post your project, your progress from last week, and what you hope to accomplish this week.

If you want to be pinged with a reminder asking about your project, let me know, and I'll harass you each week until you cancel the service.

1
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

Update on the motor pool:

The $500 Civic project has ground to a crawl, aside from replacing the battery (with Walmart's cheapest 51R) and confirming that the car does indeed run and drive, and the A/C works along with three of the four windows. The clutch pedal feels like shit (engagement is too low), but I'm confident that bleeding and adjusting the master cylinder will fix it. The old owner will be back in town this weekend at which point we'll do the needful with paperwork and see if he has to get a replacement title (The title exists and is legible but looks like a dog ate it.).

Speaking of titles, and actual work, I had to replace the starter on the other Civic. It was...not as bad a job as anticipated. I didn't have to pull the bumper, did have to pull the knock sensor, and did have spend $15 on washers to make the slightly different (case is thinner where it bolts to the engine) Amazon special fit the car, but at $60 from Amazon versus $80-$300 from Rock Auto and $325 from the parts store I can afford washers and the job was easy enough that I wouldn't hate myself if I had to do it again. I have a strong prospective buyer from this car, an old friend who fell in love with it (I let her drive it.) and is going to need a car in a few months.

The rational, but unwanted car may have a buyer (probably not, but I can probably sell the thing for five grand) if I choose to go all out and ditch all three for what I think I want (test drive pending).

What I think I want is a 10th gen Civic Si, preferably blue like this but I'd settle for silver, since I've seen one in person, thought it looked nice enough, and it looks like silver is easier to find than blue. Is it a Type R? No, but I don't have Type R money and these are allegedly easy enough to tune without much work as long as you're okay with settling for ~250HP before stuff starts breaking, and what the L15 lacks in top end power or K-series potential it makes up for with torque and gas mileage (My K24 Civic is fun but gets atrocious fuel mileage on premium by the standards of a Civic.).

Didn't get a lot of work done this week. I mostly fell into the optimization rabbit hole after I noticed that the bespoke background texture copy system, that I set up to enable infinite scrolling, does get a bit choppy in terms of performance.

Splitting the background into more textures of a smaller size should speed it up enough, and another option I put in the backlog is to lower the color depth by packing the texture into 16 bits instead of 32 (less data, less time necessary to copy it between the GPU and CPU, which was the bottleneck). I don't want to spend too much time on that as the original Alien Phobia games had a screen-sized arena, with no scrolling, and that's what I'm reproducing as a first step.

I did learn something interesting though. It turns out when you read back data from the GPU you have to wait for all scheduled work to finish, so if you do so after you dispatch the compute shaders, instead of before, you incur a penalty. In the end I could shave off an entire millisecond of each physics frame just by changing where the GPU reads are called from.

How have you been doing @Southkraut?

Thanks for asking.

I tried modelling 3D characters in the Unreal Engine, but found the tooling very cumbersome to use. I'm sure there's a way, but I did not stumble across it yet. Maybe a tutorial would help. Keep in mind that I'm a newbie at manual modelling; I've only ever constructed meshes procedurally through code, vertex by vertex by edge by triangle. So I switched back to Blender for now, which I found more approachable. I'll probably regret it when the time comes to import the models and animations into UE, but so be it for now.

Wanted to try out the new UE5.8 features, Mesh Terrain and MetaHuman. Mesh Terrain is fully 3D terrain, as opposed to a mere heightmap, which seemed like a straight improvement to me - but sadly, the tooling it purely manual, there's no API for procedural generation, so it's straight outside of my use case. MetaHuman was advertised to me as trivializing the animation of humanoid character meshes, doing all the rigging, skinning and animating, so I figured - great, no need to learn those skills, God bless technology. But as it turns out, MetaHuman is just horrendously overbuilt for my modest plans, and I would probably spend more time trying to shut off parts of it (like the 600 facial bones, or the meticulously simulated hair) for ease of use and for performance. As it turns out, it's for cinematic animation, not baby's first low-poly.

Wrote a chunky bit of design documentation on a simpler game, as opposed to the simulations I normally try to build. I'm fairly eager to try it, I like the idea. But "simpler" is a bit of an overstatement, large parts of the document are still "Just make the rest of the game here.". Still, I think it's a decent design overall, since it has iterative development from a very simple minimum-viable-prototype onwards built in.

Then, because I felt like I hadn't done anything sufficiently derpy yet, I went on to pen an atrociously oversimplified pseudocode algorithm for collision resolution tuned towards simulating weapon impacts, leaning onto an earlier attempt I had made for my Unity project that never quite got finished. Like so much.

And then I wrote down a little more lore, thought to myself - "Wait, didn't I write the same thing twenty years ago?", and started wondering where I might find those old documents. It's fun to read my own old writing, sometimes. Old me often surprises new me. Haven't found them yet, but they might still lurk in some old zip files.

I want to get the first 3D model done ASAP so I can get to coding, but it's a busy week at work. Maybe the weekend.

Bazzite Redux

How it started

How it's going:

The good:

  • BTRFS seems... fine. Only occasionally does it feel slowish and while I'm sure BTRFS itself certainly doesn't help in that department, neither does it help that my NVME drive is generic.
  • HFS is it a relief to have a working goddamn desktop. Get your shit together, System 76! All of your shit! So it's together!
  • Calling all AI boosters, I've had the ChatGPT moment. All of the running down of the local LLMs set some pretty low expectations for me, so I was surprised when I found lots of utility there. Yes, I'm sure it hallucinates way more often than the better online stuff, and buddy, does it like to hallucinate here and there, but it's usually blatantly obvious that it's off the reservation anyway. Yes, mode collapse can be a Thing, especially in certain situations, as can verbose mode, and yes, there are only so many tokens that can be processed locally, and yes, the token limit really sucks, comparatively speaking. But with all that said, this ain't ELIZA with extra steps, and running llms on an RTX 5080 seems to be thoroughly... mid? And I'm still more than a little shocked that I can get half-decent images out of a local instance of Stable-diffusion. I was not expecting that.
  • In fact, once I got back into playing around with AI, which took a back seat both because the wife went in her own direction on her work needs and because Cyberpunk 2077, I had the most fun playing with tech that I've had this decade.
  • I'm so used to Ubuntu that I keep stubbing my toes on Bazzite's immutability. This is a Good Thing because as much of a PITA as it is for me to have to set up containers so that I can run shit that demands older versions of Python (don't get me started on fucking Python) or what have you, the payoff is that I shouldn't ever run into my old normal of pretty much expecting to need a fresh OS install every time the new LTS Ubuntu drops, not to mention smegging up the OS over time with All Of The Things.
  • I can run the GOG version of Cyberpunk 2077 on Bazzite/Lutris and it... just works. Score one for killing Windoze as the dominant gaming platform.
  • AppImages have, thus far, also Just Worked. /me nervously knocks wood.
  • I haven't succumbed to Steam's siren call.
  • The learning curve wasn't anywhere near what I thought it was going to be, container conversation above notwithstanding. Did I mention fuck you, Python? I did, didn't I?
  • I am (theoretically) not training anyone else's models for free.

The Bad:

  • The glazing. The sycophancy. The loops of glazing>sycophancy>glazing>sycophancy ad nauseum. Not the worst thing in the world, but that's like >1 week of my life that I'll never get back at this point, not to say that it isn't occasionally fun and/or hilarious, but still.
  • I should have wasted more money Splurged on the full 5090.

The Ugly:

  • Odin help me, but I'm actually entertaining the idea of saving my shekels for a 5090 and the power supply to run it in tandem with the existing 5080. Not seriously enough to check prices just yet, mind you, but I could see it happening.

So that's my update on my original post. How are you doing, South... wait, no, that's not my line...

I had to stop using LLMs because I couldn't take the glazing anymore. I want to step onto the turbolift and say "deck 9" and for it to reply "deck 9" not "sure, buddy, I'll get right on that. And might I say, deck 9 is a chef's kiss choice, it has great Starfleet Energy. Whatever deck you want to be on, I'm here for you, pal."

(Shudder)

You gotta try Claude in voice mode. I read the glazing in its sanded down Manchester accent now and it makes it extra special.

I feel you. I'm not there yet, but it seems like it's only a matter of time...

I missed it, but what does Bazzite Linux mean to you?

Also

Calling all AI boosters, I've had the ChatGPT moment. All of the running down of the local LLMs set some pretty low expectations for me, so I was surprised when I found lots of utility there. Yes, I'm sure it hallucinates way more often than the better online stuff, and buddy, does it like to hallucinate here and there, but it's usually blatantly obvious that it's off the reservation anyway. Yes, mode collapse can be a Thing, especially in certain situations, as can verbose mode, and yes, there are only so many tokens that can be processed locally, and yes, the token limit really sucks, comparatively speaking. But with all that said, this ain't ELIZA with extra steps, and running llms on an RTX 5080 seems to be thoroughly... mid? And I'm still more than a little shocked that I can get half-decent images out of a local instance of Stable-diffusion. I was not expecting that.

IMO local LLMs are a 100% waste of time compared to Fable, if you have access to Fable. Thing is freakishly smart, and solves problems Opus routinely struggles with. Which is wild because 2 months ago I was impressed with Opus, and now I'm so goddamn bummed if I run out of Fable credits and have to use Opus.

So basically it's mostly a fresh new aesthetic after a decade plus of living in various Ubuntu-based distros, with all of their advantages and annoyances and thus far, I'm finding that I like it much more than I expected I would. As I said in my previous post, just NOT having a half-baked desktop experience is actually a big win after waiting (and waiting) for Cosmic to mature. The juice has been worth the squeeze.

IMO local LLMs are a 100% waste of time compared to Fable, if you have access to Fable. Thing is freakishly smart, and solves problems Opus routinely struggles with. Which is wild because 2 months ago I was impressed with Opus, and now I'm so goddamn bummed if I run out of Fable credits and have to use Opus.

I hear you, I don't doubt you, and while there's a decent chance that I'll end up wanting to play with the big boys, eventually, it's really just that hearing so much of this sort of talk here made it easy for a local LLM to exceed my basement-level expectations of it.

Over the past couple months I finished up three new books for my Public Domain Pulp book-to-visual-novel project.

The first was Conan Doyle's A Study In Scarlet, where my main goal was to finally fully address the sprite-gen quality inconsistency issues. This ended up mostly being a success, with the revamped partial-body generation approach producing more cohesive and cleaner character results overall; and, as a bonus, the sprite background removal cleanup was made much much easier by both this and a couple further process improvements around alpha masking and boundaries. (Saving my time and my wrist.)

Second was Chekhov's The Duel, which made some further marginal visual improvements on the new sprite-gen approach, though the bigger goal was finally tackling improving the color normalization process: making the dark areas darker and the light areas lighter and the off-color areas less off-color. The improvements here were again somewhat marginal, but I think still probably worth it for making the images more visually striking and less washed-out.

And then most recent was Hammett's The Maltese Falcon, which ended up mostly being about process-efficiency improvements: automatic regex fixer-upping and parallel image generation/iteration support and easier expression variation batching and what-not. Aesthetically, the images came out strong, though that was less about process quality improvements and more about 1920s detective noir having a good visual vibe.

So on the quality front overall then, things are starting to look adequate now. There are definitely still shortcomings that I’d like to keep improving on, but even if the quality were to end up stalling out at current levels, I think that would be okay. (Something I was not feeling after Pride and Prejudice.)

I think what matters much more at this point is just getting more books out. The conversion process has still been painfully slow, and the rate of starring on Github has also been disappointingly slow; and those two facts seem related, what the project having so few titles to its name (and fewer ones yet of acceptable editing-quality).

But I think I can start picking up the pace from here, putting the process efficiency improvements to better use, getting more books out faster, and especially getting some big titles like Wuthering Heights and Crime and Punishment out by the end of the year. And with each new book, the chance of virality should increase. Hopefully.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats