ai girfriends maybe?

[endoftheskies]

as an mathcel, im pretty confident in my ablity to code ai
i mean it would be pretty easy to code your own ai girlfriend from python(using torch or tensorflow) or rust(candle/etc)
imma try it and if i do it
i will be open-sourcing the code
if yall are interested what features do you want to see in it..
i will also be open-sourcing the weights lmao

 

[WastedPotential]

you know that feeling when you just woke up and you just stretched yourself

 

[endoftheskies]

you know that feeling when you just woke up and you just stretched yourself

yeah

 

[WastedPotential]

I tried learning neural networks before it became mainstream in 2017-2018 ish i think. I was too dumb to do it. I could only code a single perceptron model that could learn when a combination of two numbers represented an OR gate or not.

 

[endoftheskies]

I tried learning neural networks before it became mainstream in 2017-2018 ish i think. I was too dumb to do it. I could only code a single perceptron model that could learn when a combination of two numbers represented an OR gate or not.

its math-heavy, it was also hard for me when i was starting out also, but arxiv is an godsend on this world bro

 

[WastedPotential]

its math-heavy, it was also hard for me when i was starting out also

Nice that you persisted.

I have no motivation. I maybe would have learned more and been able to code something cool if I kept going, but I never stick long to anything

 

[RandomGuy]

you know that feeling when you just woke up and you just stretched yourself

View: https://youtube.com/shorts/qjdNeRymRns?si=_O51f9mRVoGGrozr

 

[LUCK]

When the elites replicate and monetize human companionship, the men of the world will become eternally docile. No hope for revolution or freedom will ever exist again.

Anyways better quality please

 

[endoftheskies]

as an mathcel, im pretty confident in my ablity to code ai
i mean it would be pretty easy to code your own ai girlfriend from python(using torch or tensorflow) or rust(candle/etc)
imma try it and if i do it
i will be open-sourcing the code
if yall are interested what features do you want to see in it..
i will also be open-sourcing the weights lmao

chat, i have some news.
i got multi-head attention working
i got rope with yarn working
i got word2vec.
its kinda unfinished but i still need some work.
for byte-pair encodings imma use tiktoken with gpt-4 merges..
i think i will include an 3d conv network for the visual component. i don't want to use slow-fast networks, visual transformers and all that shit.
once im finished with all the architectural stuff, imma give it as an onnx model.
of course the big question is the weights/how to train them, but i guess i will download mistral's weights or some other open source llm's weights and adjust them for this purpose

 

[Smiggles]

Lmk how it goes.