Clay-Institute-Level Question Brainstorming & Discussion

Like these?

Or do you want it to be done with a tag zone?

no no i was saying that solves a question i had (on kill, add counter, damage bonus et cetera) but not what i was trying to solve, which would be hit tracking based on tags (so a melee system thats not one-shot

You could make a tag system with no auto ko, but it would mean you couldn’t use gadgets or lasers unless they are auto ko.

Here's how

Basically, you just need to turn off automatic respawn on the tag device. Then just connect it to a counter, with player scope, to increment when you are tagged. Say that it does one-fifth damage. Then just have your target be 5 on the counter. Then connect it to an auto respawn so that when it reaches whatever target the counter is, it respawns you. Also, make sure to have the counter reset when it reaches its target goal.

Getting boxed 200 pumped in gimnite battle royale

alright.

I’m a ai expert. (only with text generation)

gimme a few days to let me cook

or maybe not Im too lazy :skull:

dragon wat ru cooking up???

2 things.

3d rendering technically has been solved by multiple people. BH actually made like a good version of it.

Also infinite loops having not been solved? What are you talking about? Wire a lifecycle to a trigger. Then that trigger to another trigger. Then the second trigger to the first. That theoretically goes on forever. But the server for the game using this will shut down after a while since it was on for a while but still possible and a thing already. I used it for my map.

3 Likes

i had an original version of this like way before i posted it (before clay institute tag too) and i just kept everything on there for a while then posted it here

1 Like

No, it won’t go forever. That’s cause of recursion.

it’ll stop at 300 loops

I’m still surprised the only thing we’ve solved as a community is AUO.

Well, technically, 3D rendering’s completed (sorta).

Still. There’s so much on the list :eyes:.

well, they’re in clay-institute for a reason.

hey custom keybindings isn’t on here, maybe ill do it later

1 Like

Give me about a month. There are a few questions about this one though, like the number of previous moves shown, or how the player could interact with said moves. This kind of sounds like an advanced use of save states, which is good, I need to learn how to use them.

1 Like

sorry smol brain moment rn I just woke up like an hour ago and my minds still foggy can you simplify what ur suggesting? cuz its feeling like a randomizer connected to a endless tree of randomizers to me

oh wait… recreating 5d chess oof ._. I don’t even play chess much less Understand it mostly checkers for me

Just take a look at 5D chess.

2 Likes

I’m on my school chromebook rn its blocked oof ._.

The two challenges I see here are the size of the AI and the speed of the program. From my math, it should be possible to import about 7 million unicode characters of text into gimkit using all 100,000 memory. We’re also only able to run 300 trigger’s worth of code a second. Keeping this in mind, and assuming we are able to import 7 million characters and then just magically add more space for the AI system itself, here are the two solutions I can think of.

Training an AI outside of Gimkit, and importing it.

This would probably be the easiest solution. We’d take an AI made outside of gimkit, and import it. The big fight here would be compressing the AI to the point where it can fit within the 7 million character limit. If we even want a decently good result, then we’re going to need to import something massive, like GPT-2. However, this AI, which gives only moderately good results, contains some 1.5 billion parameters, so some pretty insane compression would be needed.

Training an AI inside of Gimkit

This is the solution I’m personally leaning towards, at least for a large model capable of fooling a human, because with this method, I think it is possible to implement an AI with the scale of Chat-GPT, the only problem becomes training it. Now we once again run into the compression problem- importing enough data to train the AI takes insane compression. While we could import the data at runtime, doing this would be a pain and would most likely require the use of mods.

Speed

Unfortunately, while the memory size is a big problem, speed is an even bigger one. Since as of writing this, we are only able to perform 300 computations a second, a simple Neural Network (Three layers, 800:50:10 neurons) would take upwards of 2 minutes to run (800*50/300)! Now let’s think about the 1.5 billion parameters of GPT-2.

So… those are the two big problems and the two approaches that can be taken to solve this problem. Thank you for listening to my TED talk.

4 Likes

You can make this faster using parallel channels, right?