JARVIS, Anthropic's 5B Plan, Generative Agents, & More
Jarvis, Anthropic and more LLM and GPT news.
Created on April 10|Last edited on April 10
Comment
Here's the latest on GPT and LLM news!
JARVIS
Microsoft released JARVIS, which is built on HuggingGPT. It's available on HuggingFace Spaces but with a very long waitlist. On their GitHub page, they support a few more options for experimenting with JARVIS (CLI, web, server, etc).
Interesting how the project is called JARVIS but the actual paper and model are called HuggingGPT. To my understanding, there is no difference between JARVIS and HuggingGPT.
JARVIS seems like a catchier name. But in summary, this model, whether it be called JARVIS or HuggingGPT, is an ML system consisting of an LLM brain (GPT model) controlling a bunch of smaller models for various computer vision tasks. Given a prompt, the model will be able to assign certain models to certain sub-tasks within the prompt and return your expected output.
Anthropic's 5B 2-yr Plan
Anthropic aims to raise 5B over the next 2 years to compete against OpenAI. They are planning on their Claude-Next model which they deem will be 10x better than our current SOTAs like GPT-4.
Generative Agents
A paper from Stanford is using LLMs in a virtual Sims-like world where these generative agents or characters interact with one another and simulate a normal human day. It's the world of Sims (if you know the video game) but with LLMs that can understand and respond in text, and hold memory!
CerebrasGPT on HuggingFace
They have 7 available models currently ranging from 111M parameters to 13B.
References
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.