iiinova — where your digital self keeps evolving

🐁 @ghostopal.transgender.horse · 01/10/2026

theres a strong correlation between knowing more about sam altman and wanting him to die

0200

🐁 @ghostopal.transgender.horse · 30/09/2026

I opened a patreon. There's only one thing on it right now (concept art & project files for my MSPFA) but I will add more soon www.patreon.com/c/ghostopal

patreon.com

ghostopal — drawings and comics and stuff

0104

im trying my hardest so imagine a funny skeet you'd like and repost ok?

0100

everyone who likes this post will get 100 bucks

3150

i still suck at drawing people in perspective lol

I need to start practicing drawing more

120

Sometimes you just have to watch random shit you haven't seen before

080

in 2030 coding assessments will be "what was the answer to leetcode problem 5302? No googling"

000

like there are some patterns you can reuse like dynamic programming or traversals but sometimes they're written like "ok how can we hide the application of this pattern as much as possible". and then its often just rote memorization?

120

🐁 @ghostopal.transgender.horse · 27/09/2026

a lot of problems on leetcode/hackerrank/etc feel like they're just trying to trick you into thinking a problem is solved in one way when it's actually completely different, which doesn't really feel practical imo.

150

Gundam is like if they put a lot more money into My Little Pony

0102

🐁 @ghostopal.transgender.horse · 27/09/2026

had bad nightmares twice in a row this weekend

010

🐁 @ghostopal.transgender.horse · 27/09/2026

perhaps some things are best left forgotten...

010

🐁 @ghostopal.transgender.horse · 26/09/2026

Let's make "bucks" $$$$$$$$$$$$$$

2101

🐁 @ghostopal.transgender.horse · 26/09/2026

fimfictions reader is so nice its a shame the search engine is awful

010

🐁 @ghostopal.transgender.horse · 25/09/2026

1979 Gundams ending theme fucks

0110

🐁 @ghostopal.transgender.horse · 25/09/2026

there's another character i made that's like if Vriska was worse

180

🐁 @ghostopal.transgender.horse · 24/09/2026

check it out, i am a mouse like hamster

0222

🐁 @ghostopal.transgender.horse · 24/09/2026

terezi: 1M BL1ND SO 1M GO1NG TO TYP3 1N 4 W4Y TH4T COMPL3T3LY FUCKS UP SCR33N R34D3RS

4100

🐁 @ghostopal.transgender.horse · 24/09/2026

i keep inventing girls that act like sandy cheeks

060

like, i can explain it, but you wont think about and apply it like i do while doing it. the brain can only jeep track of so many not-fully-learned things at once

141

i don't think i could explain my rendering process very well in words and get it across to people a lot of it is just building intuition and muscle memory

180

its also just more efficient for well lit scenes. But even if your scene is mostly dark, you can treat an ambient-shaded color as "light" and occlusion shadows as "dark", and just add some highlights after for the very light parts.

since you also have to account for like, the transition in the shadow. whereas if you start in light, adding the shadow means you're also thinking about the transition, so you can simultaneously do the blush/normal shadow, then blend it or something.

130

i think light to dark rendering is easier to iterate on because, lets say you want to render skin with a blush part. if you start in shadow youd probably add a bit of a blush, but then to lighten it youd have to go over it with the main skin color then add the blush again which is really finicky

130

you'd think dark-to-light rendering is better because it's like real life but ive consistently had a much better time with light-to-dark rendering

250

I need to look into the blender overlay plugin for krita and also. how the fuck it works

(Of course the better solution is to treat Java like the plague)

050

AI autocomplete annoys me when they try to invent algorithms for me, but I don't mind when it shows a ghost image of all the repetitive boilerplate i was gonna write anyways. Which was honestly really helpful when I had to write Java in intellij for a few classes

150

I just think vibe-coding is like, the worst and laziest possible approach to LLM-assisted code. What happened to just using that kind of annoying AI editor autocomplete that's in every IDE now or like asking it to find gaps in your test cases? Or just saying "can you explain how this library works"

1120

I once saw someone describe it as "boiling the ocean to pick up a coin without getting wet" which i think is funny

040

vibe-coded code has a consistent pattern of being completely fucking unreadable because they thought documenting it would be a waste of tokens. Also LLMs have a habit of bending over backwards to create shitty "optimal" solutions to a poorly designed problem rather than re-evaluating the design.

130

just asked the LLM *how* to implement it and implemented it yourself, or something? It's kind of crazy that they turned programming into a total black box where you have zero clue how it works, just that it does. Sometimes. Kind of like LLMs.

140

Joachim ✊️ UVW-CWA @jdespland.bsky.social · 20/09/2026

I saw a project that I think was an attempt at an improved version of Lean, which the developer said was hundreds of thousands of lines almost all vibe-coded, and that they were "wading through the slop" to human-audit the code. Which made me think like, wouldn't this have been a lot easier if you

170

Reposted by 🐁

We don't live in slop world because of AI, we live in slop world because the purpose of everything is to generate a return on investment for capital. We evaluate everything along metrics designed to serve that purpose over any other. AI is just the best way we found of optimizing for these metrics.

9837308