iiinova — where your digital self keeps evolving

Here it is with the lights on, and some lights added to the shots. interesting! I like where it is going. Needs some more tweaks of course. I will need to add a wall glow pass to the projectiles so that they illuminate the walls as they go as well.

010

Finally, I added a bounce limit, so after a certain number of bounces they will also return to you. this number could be tweaked, perhaps by the player? dunno.. but different weapons would have different properties of course. Maybe have a 'pull all balls back now' skill? lots of options

100

Since it is a blocky-breaky kinda game, there is lots of bouncing around, and if you can only shoot a certain number of things, then having them get lost in the maze would suck a bit. So if they go offscreen then they will boomerang back to you automatically.

100

More work on the #godot prototype. playing with a limited clip approach where you have a number of projectiles and once they are thrown you have to wait till they come back before you can throw again. If they 'hit' you then you will reclaim them..

121

*automation used to be: symbolic decision making (ie old-school AI). People these days just bolt on an LLM to do their testing, but LLMs are not deterministic so you have to check their work. Symbolic is repeatable. 99% of what people are automating with LLMs can be done with a script for pennies.

111

Can AI Test Thoroughly? Why the Industry Is Sleepwalking Into a QA Crisis

This is good summary of why AI is bad at QA: paulbrodner.dev/2026/can-ai-... I am a HUGE fan of automated* testing, but we always used to joke that automation can do Quantity Assurance, but for Quality Assurance you need humans. LLMs do not change that, they make it more important if anything.

paulbrodner.dev

AI can generate a thousand tests in minutes — and that's exactly why teams are starting to mistake coverage for quality.

131

The same models that people claim can help with ground-breaking scientific reasoning and can discover new mathematical proofs will also tell you that drinking your own piss is good for your health. I don't know how people square that circle, but I guess they are just fucking idiots.

011

This is how those 'one billion daily active AI users' numbers are made: companies keep moving the normal not-AI stuff that everyone uses under the AI umbrella. Google search? Oh yeah, totes AI Reminders? AI now I guess Map Directions? probably AI Email search? fuck yeah AI fuckers

020

* mostly: I no longer see a stuck weapon, but in _very_ rare cases objects will get stuck in a wall and then because they get a frame of free movement sort of 'dance' through the walls. But this is so rare at this point that I am mostly meh about it. Will reinvestigate if it becomes an issue later.

000

Right! I made a super quick fix where I basically just give the objects a single frame reprieve from collision detection after a collision. And that seems to have (mostly*) fixed my getting-stuck-in-geo problem, and also hugely increased perf. So that is probably optimal enough for now!

100

I think that my physics code is allow the axe to penetrate the collider but then not get out, so it spawns a particle animation every single frame. If I can fix this then I suspect my perf will probably double if not more.

100

Tho I actually did this test not for perf, but to replicate this issue with the weapons getting 'stuck' in the geo. (this is going to look like shit on bsky probably, but here is a closeup where I frame step through it, you can see one of the axes gets jammed in the geo here).

100

The perf failing is a combination of the #godot physics system not being able to handle thousands and thousands of collisions per frame and my ECS needing to process tens of thousands of animations all at once. This is with zero optimisation work so far, so: meh, its probably fine for this game TBH.

100

Some stress testing this morning: it fails pretty quickly but holds up actually better than I had thought it would. Those particle effects are not actually particle effects, just a bunch of sprites being animated that I bodged together. if I fix that then this should be plenty performant 😅 #godot

100

Jason Gorman @jasongorman.bsky.social · 30/09/2026

love the smell of vibe coding in the morning

a bluesy error 503 on the discover tab, so like not that important at all. Good job Claude!

010

Reposted by Ben Britten

I think there's a good case to be made that folks using LLMs to do things they already have tools for that are quicker, more reliable and require many orders of magnitude less compute have lost the plot. See also: prompts for code that are longer than that code.

3313

And I added a handy debug panel that you can call up that has a handful of useful info in it so I can stop scrolling endlessly through the logs 😅 Not much there yet, but it is all nice and extensible so I can quickly bodge on things I need to look at in real time to figure stuff out.

000

Did not get too much time to work on the #godot prototype today but still managed to do some small tweaks. lighting: I changed the function that calculated the wall glow from distance to be a proper easing function with some adjustment available so I could adjust it and make it a bit less gloomy.

140

we've got games week at home! (well, in melbs anyway)

000

Makkon @makkon.bsky.social · 29/09/2026

this would only be a problem if the AI was trained on a massive corpus of media that was systemically biased against minorities, women, LGBTQ+ folk, etc... uh oh.

030

Reposted by Ben Britten

I'm immensely proud to announce that my texture library that's been in the making for nearly a decade is now available on my itchio store for indie devs to purchase commercial licenses. makkon.itch.io

screenshot of my itchio store, showing several texture packs available for purchse

542648995

And now that I have spent a few days on this, the code is an absolute disaster and I will spend an hour or two tomorrow cleaning it all up and getting it put into proper classes and taking out all the expensive debug stuff 😅 (well, probably not all of it, some of it is still handy)

000

Once we have that all sorted out, the corners are in and it feels much more like it _should_ feel. Here we see me going through a narrow corridor without corners and with corners. with corners feels so much more 'correct' (even tho, algorithmically it is not correct! stupid math!)

100

So we add yet another edge case pass to our line of sight algo. This is specifically looking for that edge case where we have a corner that is not technically visible, but should be! Here we can see the LoS pass with and without the new corner edge case code.

100

natemare before christmas @natescape.bsky.social · 29/09/2026

More tweaks on my #godot roguey-likey-blockey-breaky prototype. Now that my lighting is 'done' (lol) I will proceed to tweak the shit out of it forever. First up: fix the corners! The algos to calc the visible walls always leave out corners because they are technically not visible! Boo! Bad corners!

131

Reposted by Ben Britten

Do you know how much has to have gone wrong for me to not like the internet

141310368

Reposted by Ben Britten

Aftermath @aftermath.site · 28/09/2026

These AI companies should be responsible for everything they do, and they fact nobody is holding them accountable is driving me insane: aftermath.site/ai-agents-resp...

8909283

So! all the pieces are there now, but I will need to tweak on them a bit more as it still feels a bit too gloomy :) Also apologies for the initial version of this post, the vids were a bit too small, they look great on my widescreen, but not so much on bsky 😬

100

Once you have all the visible tiles, then you can use a second tileMapLayer that renders just above the main one, and you can draw the wall tiles without illumination, and modulate their brightness based on the player's light cone. I bumped up the wall brightness a bit to make it a bit more readable

120

So, I am also going to do a traditional tile-ray-cast to calculate the visible tiles. Usually you would use a Bresenham line algo, but since I can move through fractional tiles you need to use something like the DDA Amanatides + Woo algo. (which is what I am using here)

100

Turns out adding lights to a tileMapLayer in #godot is easy! Just add a PointLight2D and then set the tile occlusion in the same way you set collision. easy peasy. The problem of course is that the occluded tiles are in their own shadow. Makes for a very stylized look, but a bit hard to see. 🧵

180

gonna repost that last one with better videos 😬 sorry to whomever just liked it as i deleted it!

000