This is crazy

Large language models use computational power to predict the next token, I think this will eventually lead to a machine that can predict the future.

Ok I found a simple task GPT-4V can't do

This is really impressive, CPU inference is becoming more viable and faster, this would drop costs significantly and even allow small devices to run fast inference

Predicting the future is what LLM’s will do soon

New meta AI understands cap 🧢

post your history no cheating

Just pushed out a new version of fastai -- now compatible with the recently-released PyTorch version 2.1. github.com/fastai/fastai

managed to run inference on the Mistral Trismegistus 7B on a M1 2020, thanks alot for the effort! im be posting my prompts

Mistral Trismegistus 7B will help me make a golden ring, ima become a jeweler

using mistral for about 2h and already feel how much better it is in speech alone, it talks like a real person and don't use any odd RLHF dick-fuckery

im starting a collection of rare pepes, post yours

this is pretty good for a local 7B model, im pressing it to see how far it can go in magick theory

Approach to Language Modeling Using Numeric Tokenization Concept Overview:This approach involves a unique method of tokenizing input text. Instead of traditional methods, I employ a numeric tokenization scheme with four distinct parts: Relative Position Encoding: The first part

I wonder how much ram one will need to run llama3

Things to figure out: * How do humans synthesize information * How does language evolve and is created * How to pragmatically access information * Why is the math working in the first place

what is a good way to download someone's tweets? does X have an API still? any other better way?

Libs of TikTok Exactly

Did we achieve AGI already?

nice to hear that your team has already explored this, I don't have much knowledge in physics, just my undergraduate engineering background. What I was thinking is that if we could create an optical network parallel to a conventionally trained neural network—essentially