Welcome to WednesdAI – Pixel Dreams’ weekly update with top stories from the rapidly evolving world of Artificial Intelligence.
Let’s jump right in and waste no time meeting humanity’s first genuine robot buddy.
Subscribe to the brand new WednesdAI YouTube channel!
The Humanoid Robot Revolution
This section was going to be titled “ChatGPT Gets a Body” but just as we were going to press, NVIDIA released some video at their GTC 2024 presentation that makes the news coming out of Figure AI look like a grade school science project by comparison.
First, watch the demo from Figure AI showcasing the advanced capabilities of Figure 01. This robot can describe its surroundings, use common sense reasoning when making decisions, translate ambiguous requests into context-appropriate behavior, and describe why it executed a particular action. Figure, the company that built the robot, has gone from nothing to an autonomous humanoid robot capable of task completion using an end-to-end neural network in only 18 months.
Watch an analysis of the technology by The AI Grid:
And then days later, NVIDIA gave their keynote presentation at the GTC 2024 conference to show off Project Groot – a lab where where they are training a wide array of humanoid robots to carry out all manner of human activities. Figure 01 could hand its user an apple and put away some trash. Robots coming out of Project Groot are making juice, playing the drums, and learning how to perform complicated actions by simply watching a person do it.
Watch the video:
The AI Grid brings us a detailed breakdown of all the capabilities demonstrated at GTC 2024:
Read the official announcement about Figure 01 at Figure.AI and read NVIDIA’s official announcement of Project Groot.
Consistent Characters in Midjourney
The newly released feature solves one of the biggest gripes with AI-image generation.
Using the new CREF tag, users can apply the same facial features and other characteristics across multiple image generations. It works best when used with images generated first in Midjourney itself, applying distortion effects to likenesses from photographs in order to limit the feature’s ability to create fake images of real people.
Watch the feature in action by Cyberjungle:
Tom’s Guide has the full tutorial.
Your Own Personal Chatbot
Create your own personal chatbot by integrating your own data with advanced AI models.
“Chat with RTX” is an innovative demo application designed to customize a GPT large language model (LLM) with your unique content, such as documents, notes, or other types of data. By utilizing technologies like retrieval-augmented generation (RAG), TensorRT-LLM, and RTX acceleration, it allows users to interact with a tailored chatbot to receive prompt and contextually accurate responses.
Read NVIDIA’s official news release about Chat with RTX.
Grok-1 Open Release
Elon Musk’s xAI released the code for the AI model almost twice as big as GPT-3 at launch.
Grok, an AI modeled after The Hitchhiker’s Guide to the Galaxy, is intended to answer almost anything and even suggest what questions to ask. Grok was officially announced back in November, and the base model weights and network architecture of the 314 billion parameter ‘Mixture-of-Experts’ model Grok-1 was released just this week. The open source nature of the model means that it is uncensored for any purpose.
Prompt Engineering on YouTube gives us the technical rundown:
Read the detailed rundown at Ars Technica.
AI Videos of the Week
Check out this week’s best new Sora videos!
POV of a bee:
A bull tiptoes through a China shop:
Interview segment:
The realm of AI-generated music video continues to get exciting.
Dune-inspired video for “Arrakis” by Kream, generated in Runway:
Tribute to Marshall McLuhan – song generated in Suno, video generated in Pika:
Uncanny Harriet’s first song “Bits of Code”. The song, the video, and even the artist were all generated in AI:
Every week seems to have more exciting developments than the last so come back every week for the latest AI news and our favorite AI-generated videos of the week.
The images accompanying the news items in this article were generated in Midjourney using the following prompts:
Join us every Wednesday for WednesdAI – a Pixel Dreams production
