- Bot Eat Brain
- Posts
- Google researchers unveiled Streetscapes.
Google researchers unveiled Streetscapes.
PLUS: Bot ties the knot. Literally
TOGETHER WITH
Good morning, human brains, and welcome back to your daily munch of AI news.
Hereβs whatβs on the menu today:
Bot hits the streets π€ π£οΈ
Google researchers unveiled Streetscapes.
Bot ties the knot. Literally π€ π
University of Singapore researchers unveiled TieBot.
Sponsor Bot Eat Brain and reach over 20,000 readers.
(Now 50% off)π€―
Peep today's βWhat Would You Do?β at the bottom. π
MAIN COURSE
Bot hits the streets π€ π£οΈ
Last week, Google researchers unveiled Streetscapes. It's an AI system that generates realistic street views of entire made-up cities.
How does it work?
The AI was trained on millions of real Google Street View images. It takes three inputs:
A street map
A height map of buildings
A desired camera path
From there, it generates step-by-step video sequences with details like windows, cobblestones, lighting shadows, and more.
Whatβs under the hood?
Motion Module: Ensures smooth movement and consistency between frames.
Temporal Imputation: Each new image considers previous ones for better continuity.
Long sequences: Can generate up to 100 frames covering 170+ meters of virtual travel.
Can I customize my virtual road trip?
Yupp. You can control the city's appearance with text prompts. You can generate Parisian streets in New York style, change the time of day/weather, and more.
SPONSORED BY ROOSTER VANCE AI
AI-driven real estate business in a box
Our Shotgun Wholesaling Business is a multi-million-dollar land business that is fully automated and recession-proof.
The real estate industry is built on leads, and the Rooster Vance team has built the most advanced AI-powered lead mining platform and existence.
Build yourself a wholesale real estate empire by modeling our business inside our Shotgun Wholesaling mentorship program, or blast off by investing in your own clone of our business.
This comes complete with all of our AI bots and our custom CRM for a fully automated turnkey real estate business that will grow into your own empire with a little nurturing and care.
Book a quick call today to discover how your new automated, AI-driven real estate business can be launched and profiting $100k per month in under 60 days.
SIDE SALAD
Bot ties the knot. Literally π€ π
On July 4, University of Singapore researchers unveiled TieBot. It's a framework that teaches robots to learn tasks from human demonstration videos.
What's the big deal?
It allows robots to learn complex tasks like tying a necktie by watching videos online. Most current methods need in-domain videos filmed in the same setting as the robot. TieBot learns from any video.
How does it work?
The framework has three main components:
Real2Sim: Tracks object motion in the video and replicates it in a simulation.
Learn@Sim: Uses reinforcement learning to figure out grasping and placing points.
Sim2Real: Deploys the learned policy to a real dual-arm robot.
Why do I care?
It has the potential to solve the data scarcity problem in robotics and help transfer learned skills to real robots more efficiently.
YOUR DAILY MUNCH
Think Pieces π§
Is AI already sentient? 20% of Americans believe AI is already sentient, but AI currently relies on algorithms and data without true consciousness.
Startup News π°
Apple's released DCLM on Hugging Face. It outperforms Mistral-7B and SmolLM, on the MMLU benchmark using 40% less compute.
Research π¨βπ¬
LazyLLM β accelerates large language model (LLM) inference by selectively computing only essential tokens during both the prefilling and decoding stages.
Shape of Motion β reconstructs dynamic 3D scenes and motion from a single video using a compact set of SE(3) motion bases and data-driven priors.
FURRY FRIENDS
This is Theodora. π
βSheβs always watching.β
Respond to this email with your petβs name and pic for a chance to be featured here tomorrow!
MEMES FOR DESSERT
AI ART SHOW
WHAT WOULD YOU DO?
This band sucksβ¦ π
Who are we listening to instead? π
Let's rock π€π€π€ |
Ideas? Comments? Complaints?
Respond to this email or hit me up on π.
Until next time π€ππ§
What'd you think of today's newsletter? |