The world has witnessed the government of Israel’s genocide against Palestinians in occupied Gaza. Month after month, the government of Israel has treated Palestinians in Gaza as a subhuman group unworthy of human rights and dignity, demonstrating its specific intent to physically destroy them. Israel's government and state officials responsible for the crime of genocide must be held accountable. We are calling on all States to use every legal avenue to bring suspected perpetrators to justice. No one should be allowed to commit genocide and remain unpunished. https://t.co/lyfB55V7Y5
GPUs were created to rotate vectors. Vectors defined the shapes of objects in 3D games. To rotate a vector, you multiply it by a matrix. So, GPUs were designed to multiply millions of vectors by millions of matrices so that the player sees a dynamic scene with a high level of detail.
Then neural networks came along. They didn't need to rotate vectors, but they needed to multiply a lot of matrices representing both inputs and the parameters of a neural network. Again, neural networks required the ability to multiply millions of vectors by millions of matrices, enabling them to take numerous inputs and pass them through countless transformations.
But this wasn’t as fun because nothing was being rotated as originally intended. Then came the LLMs. In LLMs, the input vectors representing tokens are rotated by the RoPE matrices! The rotation angle for a token determines its relative position to all other tokens in the input sequence. Things have come full circle!
(The illustration is from my upcoming #theLLMbook.)
Extremely difficult scenes:
An Israeli drone targeted a child with live ammunition in one of the streets of northern Gaza.
When citizens gathered to rescue and assist him, they were hit by a direct strike, turning them into scattered remains.
This is how a father in #Gaza carries his child, wounded by Israeli airstrikes, in hopes of getting them medical treatment.
Every detail of this scene is a crime‼️