I know MS Teams is awful on every platform. But it's a special sort of awful when you have to run it in a browser on Linux
@[email protected] Procol Harum's lesser known hit, "A special sort of awful"
Front page
@kirtai @whitequark @iris *glances towards profile picture* I think I'll be ok :)
@[email protected] this comment prompted me to go to your profile as up to this moment I'd only viewed your profile pic in feed - where my brain has always interpreted the mini version as a teddy bear with a party hat.
1
xoxo.zone/users/Ashedryden/s...
Oof. This. The combination of #adhd and #chronicfatigue is a life of constant war between "you should rest immediately when you need to" and "here is your reminder of critical thing that if you don't do it right now it will drop into a black hole for ever"
When your laundry timer goes off the second you finally decide to sit down
@grimalkina @mavnn Huh. Well, OK! Practical data beats theoretical concerns, so if it's helpful I shall grumble more quietly. (What, me, not grumble? Hah, as if :)
I suspect this is another facet of "everything everywhere is a mess, nothing is ever perfect including the definition of 'perfect', so you do what you can with what you have at hand."
Life. Just a never ending series of compromises, but beats the alternative.
@[email protected] @[email protected] Yeah, this is fascinating as I'd only hit the reverse issue; people thinking they had 'taught' the LLM something and then being surprised (and sometimes embarrassed while showing off the progress) that the explanation had no impact on the next time they asked the LLM to do the same task.
Using 'learning' as a way to explain how the question you are asking shapes the response is something I'd never thought to try, but is certainly something I'll keep in mind for the future.
Anyway, sorry for derailing the thing you were meaning to spend your time on this morning; this is very much idle curiousity on my end, so please don't let it highjack you.
1
@grimalkina I think what bothers me about the (in my view, which is of course inarguably 100% correct 😊 use of "learning" is that it carries implications that are inaccurate and leads people to making assumptions and decisions based on that inaccurate implication. Not even in a malicious way, pareidolia is absolutely a thing, but secondary decisions are made based on primary misunderstandings, and once those get hardened into a system... it's difficult to fix, because systems are persistent.
@[email protected] @[email protected] The language space is getting pretty overloaded in this area in ways that are unhelpful, it has to be said. I recently nearly skipped over a genuinely interesting piece of machine learning (sorry!) research as it had AI in the title and was something LLMs would obviously be terrible at - but they weren't using LLMs.
> I really mostly saw people correctly take it to just mean "this changes in response to information"
It's interesting to see you (Cat) say this, as one of the issues I have with LLMs is that people assume they will learn while in reality the 'learning' already happened at the model training stage (with added bonus confusion if there is or isn't any context being stored), and the thing you're actually using doesn't change in response to information. Which is true of a lot of machine learning tools in general, not just LLMs, but LLMs have the added disguise of tracking context within a conversation (and frequently now an internet connection), so interface almost implies that longer term learning is happening (despite not having 'learning' in the name :facepalm:).
8
Tempted to write a json-ld library for #zig as a way of learning Zig properly and because it might be a better way of implementing #fediverss (blog.mavnn.eu/2026/04/10/fed...) for low resource environments than dotnet. #activitypub #rss
Me, being clever: so I don't want this RayCast node firing every physics tick, so I'll disable it and only enable it when I force it to update.
adds all the code to enable and disable the ray casting
Me, actually reading the doc comment for the force raycast update function: "Note: enabled does not need to be set to true for this to work." sigh
It's almost like the devs considered this use case! #godot #gamedev
My latest modern art creation: the warming of the soup.
Worth at least an extra ten thousand due to the ephemeral nature of its satisfying beauty.
@katemorley
Did you whack him with it?
@[email protected] @[email protected] Doing this to random people on the internet seems rude and lacking self awareness. Doing this to your employer seems to take a... special something, and maybe limit future life opportunities
bonfire.mavnn.eu/pub/objects...
Could these be... new features related to the new course draft?
Yes. Yes they could. #visualnovel #gamedev
⁂ Article
Figure 1: Smooth style with photos
Figure 2: Pixelated style with pixel art
Crunchier pixels in VisualInk
Read this post in its full formatted glory at https://blog.mavnn.eu/2026/05/01/crunchier_pixels_in_visual_ink.html
Pixel art has a certain charm all of its own, and visual novels in particular often borrow from the older games of yesteryear for their artistic inspiration.
But unfortunately, that means that if you're using a web based visual novel engine like VisualInk your assets can end up looking blurred and smudgy depending on the screen size where they are viewed.
VisualInk now has the pixel_style variable for choosing whether images that need to be resized are scaled using "Smooth" style (the default, good for photos and scanned hand drawn artwork), and "Pixelated" style for low resolution pixel art to be scaled up keeping nice sharp edges between the pixels.
As an example:
Figure 1: Smooth style with photos
Versus (excuse the placeholder images):
Figure 2: Pixelated style with pixel art
Crunchier pixels in VisualInk
Read this post in its full formatted glory at https://blog.mavnn.eu/2026/05/01/crunchier_pixels_in_visual_ink.html
Pixel art has a certain charm all of its own, and visual novels in particular often borrow from the older games of yesteryear for their artistic inspiration.
But unfortunately, that means that if you're using a web based visual novel engine like VisualInk your assets can end up looking blurred and smudgy depending on the screen size where they are viewed.
VisualInk now has the pixel_style variable for choosing whether images that need to be resized are scaled using "Smooth" style (the default, good for photos and scanned hand drawn artwork), and "Pixelated" style for low resolution pixel art to be scaled up keeping nice sharp edges between the pixels.
As an example:
Figure 1: Smooth style with photos
Versus (excuse the placeholder images):
Figure 2: Pixelated style with pixel art
@blog Of course, the only issue here is that the screen shot on the blog is itself going to get scaled, but hey ho!
Ideas for the new course starting to solidify, with the plan being a five week online intro to creating #visualnovel assets of your own for under 18s with no budget (above and beyond having a computer) using tech you probably have around the house and open source software like #krita and #audacity. Official announcement coming soon. #HomeEd #GameDev
Also, if you have less terrible ideas for a course name, I'm all ears
Ideas for the new course starting to solidify, with the plan being a five week online intro to creating #visualnovel assets of your own for under 18s with no budget (above and beyond having a computer) using tech you probably have around the house and open source software like #krita and #audacity. Official announcement coming soon. #HomeEd #GameDev
2
Security Engineers: "Automatically patch all the things right away!"
NPM infection # 683735
"... okay maybe wait a couple weeks on patching things ..."
@[email protected] "There's about five hundred thousand more valuable targets with the same vulnerability, so just think of it like Russian roulette with a 500,000 shot revolver. That shoots like a really fast machine gun. I'm losing this metaphor, aren't I?"
Been a bit quiet on the VisualInk (visualink.mavnn.eu/) front for a few weeks, but a minor release went out today which allows scenes to have names with spaces in. Previously a bug meant that you could upload scene backgrounds and give them a name with a space, but they then wouldn't show up correctly in your visual novel.
Also, look out for some new #visualnovel course news in the coming week!
Anyone else allegedly being a robot ? I've been asked again and again on a loop 4 times and not working
Might be a vivaldi related bug , but this is such a poor experience
@[email protected] I have a couple of websites that gets in this loop on one of my devices regularly. Other Cloudflare checks on the same device work, I can access the site from an other device, but the tick box just loops. The next day it'll be working again. It's been going on for years intermittently at this point.
alright friends, it's Wednesday, and you know what that means!
no? uh yeah, I have no idea either. sorry.
@[email protected] Same thing we do every Wednesday. Try to take over the world!
@mavnn nice, that’s very cool!
@[email protected] I think some of it came from previous experience with character generators disappearing and leaving people unable to access their old (or even current) characters. That said, I've realised this isn't quite what your were proposing: there's no proviso to release products built on top of the source code, I don't think
- all source remains property of the dev, not allowed to release source publicly etc
As an indie dev of both games and apps the idea of actually releasing software this way scares me of course and I imagine it would be very risky and open to abuse. But… *waves hands*… somehow?!
@[email protected] I've no idea how successful it was, but Dan Simon actually did this when he created the HERODesigner software (herogames.com/store/category...) for the Hero/Champions #ttrpg.
You can pay for the app, which is obviously what most people do, but you can also pay for the source code to modify the app with your own house rules etc.
I honestly have no idea at all how many copies of the source code were sold, but it was definitely interesting to see this model being tried out 'in the wild'.
2
I wonder what the stupidest way for a business to fail with LLMs will be
@[email protected] @[email protected] It's going to be interesting seeing somebody beat the opening double tap of "we let AI delete our product" followed by "we let AI delete our product after seeing it happen to the other guys already"
