We’ve raised $8 million to accelerate AI’s ability to solve critical problems in the physical world! What if AI could learn new skills as fast as Neo learns kung-fu in The Matrix? We're building the tools to make this possible for AI & robotics teams. 🌐 https://t.co/HvvptvPTwk

🚀⚙️ Today we’re launching the Bifrost Post-processing Pipeline. A single click creates rules that align datasets with your labeling policy, so your models train on clean signals for maximum performance. Next: boost realism and diversity with image augmentation too

From one scene to a thousand variations - in minutes. Building diverse datasets for computer vision is no longer a bottleneck. Train and test faster. Control what matters. https://t.co/aA0MTQTZVK

We’re excited to advance this frontier of Physical Intelligence together with our partners @CarbideVentures @AirbusVentures @peakxvpartners @Wavemaker_VC @Techstars

Shoutout to @TechCrunch & @kateparknews for covering the fundraising annoucement! https://t.co/0mZcBgy9JU

AI has only begun to scratch the surface of what it can truly achieve in these critical industries. We’re here to bridge the gap and accelerate AI towards its biggest frontier—the physical world. Read more here:

Think about industries that keep our world running - those that build, move and protect. They’re facing significant challenges in labor shortages, supply chain breakages and geopolitical volatility.

AI has had extraordinary breakthroughs in the digital realm. Yet the biggest and most pressing challenges aren’t on our screens; they’re in the physical world:

So awesome to see Animall, @astronauts_id, @bijnisdotcom, @bifrost_ai, @gokwik, @kargo_techno, Momos & @parkplus_io in @ForbesAsia's 2022 list of 100 startups that are addressing real-world challenges with fresh thinking & innovative products & services.🤩 https://t.co/vZKeutPALc

Create richer tags 🏷️, features 🖼️, and metadata 📍, and you'll create better training and testing datasets.

Data labeling still sucks. 👎 Human labeling workflows are fraught with errors and inconsistencies - but great tools and platforms are being built to make it suck less! https://t.co/6FWf3WJzHV Here's the lowdown on the labeling landscape in 2022. #labeling #syntheticdata

Find out how Bifrost AI Intern Shao En generated 20+ datasets in a week, beat the state-of-the-art in aircraft detection 👊🎨📈 and still had time for 🏎️ Mario Kart 🏁 at the end of it. https://t.co/8OrijSoI2Y

In our latest post, we explain how data collection impacts AI behavior, and discuss the promises and perils of each strategy. Understanding these influences helps us design collection strategies to build better performing, less biased AI systems. https://t.co/RfMwjykgR6 (2/2)

If the AI training process is schooling for machines, then data collection is curriculum design. 🤖📚 A quality curriculum tends to produce great students. This extends to machine learning too. (1/2)

Excellent work by the @bifrost_ai team to help their customers build 3D worlds in super quick time. great article on why this is relevant and useful to folks in as diverse areas as gaming , autonomous driving and space exploration @AugieIlag @Sequoia_India @charleswongzx

Great demo from the folks at @nvidia, illustrating the outrageous variety of scenarios that are used to train robust AI vision systems! https://t.co/rWs7rt7Bym AI models trained on wild datasets tend to be more robust, and better generalize to unexpected scenarios.

Andrew Ng: The dominant paradigm over the last decade was to download the data set while you focus on improving the code [...] it’s now more productive to hold the neural network architecture fixed, and instead find ways to improve the data. https://t.co/D28hhv63JZ

What I’ve learned about making synthetic data work for training ML models:

Getting enough quality data to build AI is tough. That's why we're on a mission to supercharge AI development by building 3D worlds 🪐 and synthetic data 👾 - enabling teams to build diverse datasets in hours instead of months. Read more:

Boomer generation this, millennial generation that...what about the procedural generation?