You can try to make a neural network.
It will be fun little challenge for you on the backpropagation step.
i keep seeing people on snap regularly talk about neural networks for all sorts of complex tasks, but i don't think snap is anywhere close to being able to run a neural network that can even remember words, let alone understand and properly answer questions. how would you even train the neural network and fit all the data into a snap project?
around the time dame da ne memes were popular and all the deepfakes were going around of it (still with the pretty obvious visual issues), the net that people were using was 1.4GB ZIPPED. and that's just for a bad video of a clearly distorted and low resolution face. people waited a good amount of time to just see the video generate using google's cloud services, imagine the time it would take to actually train the ai in the first place, and doing it through snap.
Scisnap
The proof of the AI-pudding is a contest between guessing algorithms. The “choose a number between 1 and 10” format is not the most suitable I can think of though; the odds of guessing right are about 10%, so it will take many, many rounds to decide how good a guessing algorithm actually is.
(BTW paradoxically a smart player or algorithm may consistently beat other contestants, and thus outperform e.g. a random generator, but when playing a duel against an actual random generator all players and algorithms are equally strong: the odds of winning are equal to those of loosing).
Rock Paper Scissors
So … how about good old Rock Paper Scissors? It’s dead simple, and interesting events have high odds. I built a (rather basic) competition platform:
For demonstration purposes two really stupid algorithms (they are sometimes too slow, or generate invalid values) were implemented as Player 1 and Player 2, respectively. To watch a demo, press the green flag.
Challenge
If at least two Snappers can develop a smarter algorithm and are willing to put them up for testing, I will gladly insert them into the “Player x” scripts. For a realistic outcome of the competition (duel) I propose to configure at least 100 rounds. The time limit for responding is 0.5 seconds now (in Turbo mode; please take into account that the competing algorithm is also using processor time). Look into any of the Player-scripts if you want to know what is expected of algorithms (it’s super easy).
Like I wrote, it’s a really basic platform; an underlying assumption is that competing RPS algorithms will not misbehave (like manipulating the history). Also it’s just for duels, not for a multi-algorithm competition.
I’m looking forward to contestants - who dares?
return random(1,3)
Well to be honest you can connect repl.it servers to Snap!, and along with Snap's Matrix Multiplication (That's what defines Snap) you can make neural networks.
You just need time. Because Neural Networks take INCREDIBLY slow to train.
EDIT 1: Besides training, you need to account for cost. I mean, repl.it is free and stuff, but repl.it is paying for the servers and how much electricity it uses. Using servers and Matrix Multiplication can take lots of energy. And that energy comes with a cost.
What?
100% im in
round 20 0% correct it just guesses my number -1
Great. If anyone else would like to compete we can have a first contest.
if you use repl.it, why do you not care that they're paying for the servers? if they weren't paying for the servers you wouldn't be able to use repl.it
Why be so benovolent?
Somebody is paying for us,and they won't stop paying if we don't care,so why should we?
they aren't paying for servers for you. they're paying for servers because they make money off the people who do pay, and they hope that by offering free services they can attract more people, which does a few things:
- causes people to tell others about it, which brings in more people
- causes people to move away from other services, which might get the services shut down and bring in all those people
- could convince all those new people to start paying for the extra features
in the long term, they want to have the most paying users and the fewest non-paying users (non paying users are only useful if they create paying users). this means they could do all sorts of things that you wouldn't like, such as removing features you previously had, banning you for no reason, changing rules so that you get banned, making it more complicated to use, etc.
nothing is free except out of goodwill, and if you don't have goodwill, why should they?
yeah we shouldn't be those people,we should get the free services
they shouldn't,ideally
but reality is not ideal,so we can use their free stuff
they aren't doing this now,and if they do,i can switch to another online code-deploying(dk how you call that) platform
this reply available for 50¢ CAD
this reply available for 50¢ CAD
posting doesn't cost anything
ideally they would charge per post
lool
btw,we are a bit off topic and the op might be unhappy
my whole point is that slate does not need to calculate how much he costs replit on servers
i still agree that matrix multiplication costs energy,although a negligible amount if not causing computer to compute 24 hours a day
ahem
payed money lol
how would you feel if you had your own servers to pay for and keep up
Nothing in this world is free, I hope you know.
not nice,thats why i use replits free ones !
maybe discuss this somewhere else?