tyla do be walking around looking h**** as f***
I also spent like 5 mins on this post to make it digestible for you sensitive mfs. Wouldn’t want to set any of yall off
Just storing this for future reference
Clip is supervised and uses image text pairs with contrastive learning to create an embedding space
Medclip improves on this by using sparse labeling to drive parameter inference using similarity between rhe sparse and predicted labels
My idea first uses contrastive learning to pair images and text seperately, then uses the geometry of the resulting manifolds to overlap and produce an embedding space. This can be improved by additionally making a new latent space via factor graph to create the manifold similar to kernel trick in svms
S*** is groovy asf
date friday
date saturday
Win or lose my d*** getting sucked
What’s your fumble rate?
Target audience for streamers is mexican kids under 22 years old. Fair to say?
Their audience is illiterate people
I’ve fumbled so many girls and the one I was supposed to marry
F*** it. They’ll disappoint you anyways