Favorites February 25th 2023

San Martin California, August 2021

Nabeel Qureshi @nabeelqu

We know the genes for ‘short sleepers’ (5 hours instead of 8, no bad side effects); using embryo gene editing, newborns could have these as default setting. Would increase effective human lifespan by ~20%. Seems like a clear EA cause area. 🙃

I wonder if the scary thing in Gattaca was the level of difference between edited and non edited people. Selecting for people to need a bit less sleep, or for people less likely to get Alzheimer's feels more reasonable, but it seems like a slippery slope…

Simon Willison @simonw

From the paper: "For instance, LLaMA-13B outperforms GPT-3 on most bench- marks, despite being 10× smaller. We believe that this model will help democratize the access and study of LLMs, since it can be run on a single GPU." https://t.co/z3AQuxuS4d

Guillaume Lample @GuillaumeLample

Today we release LLaMA, 4 foundation models ranging from 7B to 65B parameters. LLaMA-13B outperforms OPT and GPT-3 175B on most benchmarks. LLaMA-65B is competitive with Chinchilla 70B and PaLM 540B. The weights for all models are open and available at https://t.co/q51f2oPZlE 1/n https://t.co/DPyJFBfWEq

This is a really impressive accomplishment, it runs on a single data-center grade GPU, previous LLMs require significantly more. They're not actually releasing this to the general public, but to other researchers, the key thing is that this means that LLMs that can run on a phone may be entirely possible. They've proven that you can get good results out of much smaller models than anyone realized. Estimates indicate this cost them like $4 million to train, (Stable Diffusion cost ~$600k).

M. Nolan Gray @mnolangray

It's easy to get cynical if you're an American urbanist. But remember that the sprawl that surrounds you is a canvas for something so much better.