AI in code.org

DragonFireGames

Just a peek...

rockyo_

DragonFireGames yeah nah you are way to good at this holy

gZany

100 coins for the link to this

DragonFireGames

gZany I'll give it to you in a private post.

DragonFireGames

gZany the last hundred

lol

DragonFireGames

A slightly more ridiculous answer:

Phoenix

link

gZany

If he sends a link to the whole forum it may crash his server. He told me in a private chat.

Varrience

gZany is it using a backend as i assume, or is it using CDO's inbuilt AI feature? of which i have been meaning to look into but don't know how to structure the training data to be effective for CDO's models and or don't have any big enough csv data to throw into it for actual results

Captain_Jack_Sparrow

Varrience does Code.org actually have a inbuilt AI feature?

DragonFireGames

Varrience It uses a backend and the image model is https://huggingface.co/spaces/stabilityai/stable-diffusion
the text model is https://huggingface.co/openai-community/gpt2

Letti42

Captain_Jack_Sparrow it does for applab

Binary_Coder

DragonFireGames I've used GPT in projects like this too, don't you have to pay for it?

MonsterYT_DaGamer

H O W

DragonFireGames

MonsterYT_DaGamer I send a request to my server, my server sends a request to huggingface, huggingface returns the data, my server then returns the data as an image which can be parsed by the cdo project.

Varrience

DragonFireGames you may be able to generate results on your server end, from my results they've been pretty good at least while using nodejs

DragonFireGames

Varrience It was mostly a test of my new fetch() implementation in FKEY, pay me 100Ⓒ (the price set by Gabriel) and I can show you

Varrience

DragonFireGames considering i built the fetch library lmao (well a baseline version of it anyways) seems you've vastly improved it for modular usage,
also you may be able to generate results on your server end, from my results they've been pretty good at least while using nodejs so that's most likely what i will do later on rather than key chaining

oh also update on the CDO AI model stuff very weird and can only do predictions with limited data, so generating results will most likely be very tedious with probably needing model chaining to get anywhere near the performance of other stuff

LGM_Productions

DragonFireGames

I would pay, but I’m too poor.

DragonFireGames

Varrience It's a bit different than your fetch library.

DragonFireGames

The main difference is that I can send and receive data easily using the new version of fetch. I also use Owokoyo's method of parsing the image which is like 100x faster. The fetched content doesn't have to be text or json either, you can also just fetch an image directly.

https://cdo-backend.onrender.com/fetch?url=<insert url here>&data=<json for 2nd parameter of fetch>
you can also add &proxy=1 if you want to just see the unencoded data using your browser

Varrience

DragonFireGames ah yea the biggest jump for it is probably using fromCharCode and passing through an array through it though seems like it'd be much faster, though i don't think there's really a need to fetch an image directly because then it just takes longer because it needs to pass through your proxy, unless your storing cached results which can be beneficial.... though i may update my library to include the method sooner or later

DragonFireGames

Varrience fetching images is useful for the ai image gen and probably nothing else