@mavnn Aw ☺️
"Ostensibly for Risus of all things. Takes me back."
"He wrote Risus as well."
"Ah!"
Front page
Whelp, found out today that someone wanting to improve the clarity and engagement of their #ttrpg writing hadn't heard of 'Toast of the Town' so I was able to give them enough experience to immediately level up.
Thanks, @[email protected]
2
Truly evil lobbying example https://www.politico.eu/article/europe-choose-ai-climate-goals-data-center-chief-warns/
@[email protected] You'd hope tactics like this would be self-defeating, but unfortunately that "hope" is bearing rather a lot of weight I'm not sure it can support.
Applesauce, like its namesake, contains neither carrots nor sticks.
@[email protected] But like a pineapple does it contain neither apple nor sauce?
2
Chatting with someone about #ttrpg writing and was reminded of Nobilis 3rd edition, a game that contains my all time favourate example character and that has writing in its rules that you could quote for days. Like... how the interaction between your supernatural connection to things you value (the Treasure stat, think things like Thor's hammer) interacts with the Aspect stat (your ability to personally perform tasks beyond mortal limits, or opposed to say Domain where you control the thing you're a Power of to create miracles):
If you possess something using Treasure, you can wield your Aspect through its body. This draws on the natural limits of that body, however, rather than your own — if you’re possessing one of your hopes, for instance, you’re not going to be able to lift a great weight very easily because hope has no arms and legs. All you’ll be able to do is coil more tightly about your own heart, or shine in yourself with greater strength.
Thinking about LLMs again because, unfortunately, I needed to think about LLMs again and I think I found a way to say something.
LLMs are a technique for matching input requests with semantically connected outputs from their training set, where 'semantically connected' is determined by how language in different contexts in the training data. That's what they do: find related language to what 'should come after' the prompt.
Which means that with enough input related to anything in the training set, you'll start getting the related training data (mixed and matched at the token level to varying degrees). Bomb making instructions? If they're in the training set, there's a way to extract them. Hacking tips and tricks? Of course. Other customers private data? If the LLM can access it, it can be triggered to share it.
Which means, fundamentally, the only way to control LLM output is... choosing the LLM training data. Apart from all of the odious business practices of the current suppliers, why are we trying to legislate LLM output when what we really need to enforce is ethical LLM training input? The rest is just noise; even banning LLMs makes no sense, at the end of the day they're just applied matrix multiplication and we've seen how 'banning maths' goes historically a few times already. #llm #MathsBan #IllegalNumbers
And synchronisity in full play I'm not the only person thinking about this and trying to get the idea communicated in different ways.
Thinking about LLMs again because, unfortunately, I needed to think about LLMs again and I think I found a way to say something.
LLMs are a technique for matching input requests with semantically connected outputs from their training set, where 'semantically connected' is determined by how language in different contexts in the training data. That's what they do: find related language to what 'should come after' the prompt.
Which means that with enough input related to anything in the training set, you'll start getting the related training data (mixed and matched at the token level to varying degrees). Bomb making instructions? If they're in the training set, there's a way to extract them. Hacking tips and tricks? Of course. Other customers private data? If the LLM can access it, it can be triggered to share it.
Which means, fundamentally, the only way to control LLM output is... choosing the LLM training data. Apart from all of the odious business practices of the current suppliers, why are we trying to legislate LLM output when what we really need to enforce is ethical LLM training input? The rest is just noise; even banning LLMs makes no sense, at the end of the day they're just applied matrix multiplication and we've seen how 'banning maths' goes historically a few times already. #llm #MathsBan #IllegalNumbers
1
Teaching early teens how to create their own game assets for #VisualNovels (and #GameDev in general) and today we were talking about using layers to speed up art creation (alpha locking, filters on layer groups), to give characters different expressions, or even create variations with multiple colours without having to redraw the whole image every time. In talking about expressions we discussed icons and why adding two diagonal lines to an orcs face immediately makes them an angry orc.
We also talked about how you always want clear visual separation between your visual focus (the orc) and the background (the corridor), lowering saturation and making sure that there's a contrast between the edge of the focus object and the background even if it isn't "realistic" - i.e. why it wouldn't work to have the corridor fade to black if we want to use filled line art with black lines.
All after doing some thumbnail sketches to work out what the problems were in the first place. Not bad for an hours work! #krita #VisualDesign #IterativeImprovements
4 media
@mavnn 🟪 do you have five highs?
@[email protected] while I had sort of assumed the five referred to fingers, I suppose there is also the possibility that (the people being represented by) American kids on TV during my childhood took a lot more drugs than I thought.
1
🟪 if your hands are full you can always give a thigh-five instead
@[email protected] ...you have five thighs?
3
Not too surprisingly, we're huge fans here of Inkle studios (headed up by @[email protected] and @[email protected]) which is part of why we chose to use their scripting language to power VisualInk.
We are mentioning this because currently their whole back catalogue is up for a charity supporting sale on Humble Bundle and if you like fun, quirky, narrative heavy games you may want to go have a look. We, er, might have just bought a copy ourselves.
@jonny @geeeero @SnoopJ @davidgerard it is if nothing else an excellent shitpost. It will be an even funnier shitpost if it works though
@[email protected] @[email protected] Funny enough for another star and watcher definitely. I do wonder if (unfortunately) links from places like xitter and the orange place are where the repos get found.
Just testing something. #kumquat
Scripted up a simple data transformation for school today in Google Sheets and App Scripts are slooooow. Like order of magnitude slower than running vbscript in Excel circa 2000. Is this normal?
Turns out that each operation is very, very, slow but the code speed isn't actually too bad. Basically: for must batch updates and then set values for large ranges at a time, or your code will run like molasses.
Scripted up a simple data transformation for school today in Google Sheets and App Scripts are slooooow. Like order of magnitude slower than running vbscript in Excel circa 2000. Is this normal?
1
@mavnn 🟧 sounds silly and fun. mostly commenting on how afaik both gdscript and built-in logic in Godot are interpreted directly and there's no extra compilation step in the middle
might be JIT though, although assuming not
@[email protected] I think you're right, yes.
@mavnn 🟧 from what I understand nothing is compiled at all, just run directly from definitions, which makes it relatively impressive
although generally the hard limit is number of nodes since I believe those don't optimise runtime at all and just run every node's logic every frame, minus events which are properly scheduled
@[email protected] I probably wasn't clear: the spells your character casts are a stack based programming language whose built in functions are implemented as nodes. So gdscript interpreting arrays of nodes as stacks of operations, sort of. Including quotations and subprocesses - sorry, subspells. It's very silly.
2
Lighting starting to get going, and no #godot logo! (No real artwork either, but hey).
Initial performance tests are encouraging though; given the spells are basically a very slowly interpreted language built of Godot nodes I was worried it would be unworkably slow and I'd need to start optimising early to get it workable. Turns out that even with dozens of spells flying around at a time it all still just works. #GameDev
4
One of the students at my wife's school showed great cunning and skill by managing to sneak his mobile phone into a no digital device externally moderated exam.
He showed slightly less cunning when he posted a picture of one of the other students taking the exam to my wife's Google Classroom complaining that nobody with that posture could be taking the exam seriously.
I'm honestly unclear on what he thought would happen next.
