HuggingGPT: one chatbot to rule them all

PLUS: No more blurry Zoom calls

Author

Anthony Castrio
April 04, 2023

Good morning human brains, welcome back to your daily munch of AI news.

Here’s what’s up:

  • Zoom! Enhance! just became a real technology 🕵️

  • One chatbot to rule them all 💍

  • An AI tool that automatically animates CGI characters 🎥

  • A cartoonist from 1923 predicted the singularity in 2023 ✏️

Follow @BotEatBrain on Twitter for breaking news.

P.S. We are hiring researchers and writers, click here for more info.

APPETIZER

ReBotNet: No more blurry Zoom Calls.

ReBotNet can make live streams and video calls look better by using AI to sharpen your image in real-time.

It’s basically the “Zoom! Enhance!” meme from CSI brought to life.

ReBotNet works almost as well as the best available image-restoration models while being much much faster.

The RVRT model is slightly better, but not as fast as ReBotNet.

Don’t forget: AI can’t find information that isn’t there. It CAN make up information if you let it and that can be dangerous. This tech is awesome for making your video calls sharper, but it will never reveal a killer’s face hidden in the reflection of a screw.

CSI: Zoom! Enhance!

BUZZWORD OF THE DAY

SOTA

It stands for State of the Art and is often used in AI research to indicate a new best-in-class technology based on some standardized performance comparison.

FROM OUR PARTNERS

Say goodbye to endless support tickets

SelfService Chat enables your users to self-serve, reducing support tickets. Spend more time working on your business than in it.

MAIN COURSE

HuggingGPT: one chatbot to rule them all 💍

HuggingGPT is a new system that uses big language models, like ChatGPT, to control other AI models and solve complicated tasks.

It connects with a lot of AI models from Hugging Face (an online community for researchers to build and share AI models) and uses language as a way to work with them.

Language models can connect other AI models together to solve complicated tasks.

By doing this, HuggingGPT can help with many different tasks and do well in areas like language, seeing, and talking. Basically, it wants to be JARVIS from Iron Man (and that’s what they’ve named their GitHub repo too).

In this concept, an LLM (Large Language Model) acts as a controller, managing and organizing the cooperation of expert models. The LLM first plans a list of tasks based on the user request and then assigns expert models to each task. After the experts execute the tasks, the LLM collects the results and responds to the user.

From the HuggingGPT paper.

Don’t confuse this JARVIS with the Jarvis from Iron Man, or with this separate (also very cool) project released yesterday with the same name:

Mckay Wrigley uses his voice to direct an AI programmer.

LLMs like ChatGPT, LLaMA, and yes, even Google Bard, are quickly turning into the “glue” that connects all AI models together.

Predictions:

1/ We will soon see fully autonomous AI agents operating on the Internet, running businesses, and pursuing goals set by their human masters with little-to-no oversight.

2/ An AI will commit fraud or a financial crime within the next 6 months and its human creator will be liable (if they can be found).

3/ A platform will be launched within the next 3 months for managing autonomous AI assistants that can handle tasks like coordinating appointments or planning a trip.

MEMES FOR DESSERT

YOUR DAILY MUNCH

Think Pieces

Stanford released their 2023 report on the state of AI.

Could you train a ChatGPT-beating model for $85,000 and run it in a browser?

Applying the Dunning-Kruger effect to ChatGPT.

Startup News

Fourthline raised $54 million to build AI-powered identity verification tools for the finance industry.

Quantexa raised $129M at a $1.8B valuation to help navigate online fraud

Suprema AI launched a tool for better face detection.

Twitter open-sourced the recommendation algorithm for their home feed.

Police relied on data from Clearview AI and put the wrong person in jail.

Research

BloombergGPT: A Large Language Model for Finance

Achilles Heel for AGI: is it possible to implant a weakness in a superintelligent AI?

Objaverse: a Massive Dataset with 800K+ Annotated 3D Objects

Just some of the 3D objects available in the Objaverse dataset.

LLaMA-Adapter: a small modification to Facebook’s LLaMA AI that works almost as well as larger models like Alpaca-7B and GPT-3, while also having multi-modal capacity (it can process images).

Tools

avtrs.ai: generate AI avatars with an iPhone app

senec.ai: create a landing page in seconds, just type your startup idea and click generate.

stockimg.ai: easily generate logos, book covers, posters, and more.

askrails.com: an AI Ruby on Rails Expert at your fingertips.

wonderdynamics.com: an AI tool that automatically animates, lights, and composes CG characters into a live-action scene [closed beta].

Wonder Dynamics Demo

AI EXPLAINED

Why does AI struggle so hard with generating human hands?

AI ART-SHOW

“The kid stood frozen in the middle of the street, his heart racing as he gazed in wonder at the distant flashes of lightning and intense bursts of light, wondering what incredible events were unfolding just beyond his view.” - @Paperwrks3d

Until next time 🤖😋🧠

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.