Lugh

joined 2 years ago
MODERATOR OF
 

A team of researchers across 20 separate labs just unveiled Genesis, an open-source physics engine that combines generative AI with ultra-fast simulations, potentially transforming how AI learns to interact with the physical world.

Genesis runs 430,000 times faster than real-time physics, achieving 43 million FPS on a single RTX 4090 GPU.

It’s built in pure Python, it's 10-80x faster than existing solutions like Isaac Gym and MJX.

The platform can train real-world transferable robot locomotion policies in just 26 seconds.

The platform is fully open-source and will soon include a generative framework for creating 4D environments.

Why it matters: By enabling AI to run millions of simulations at unprecedented speeds, Genesis could massively accelerate robots' ability to understand our physical world. Open-sourcing this tech, along with its ability to generate complex environments from simple prompts, could spark a whole new wave of innovation in physical AI.

Commentary from 'The Rundown' Newsletter

I follow a lot of tech news, and one of the most common biases I see in most commentary is its 'Venture- Capital-Centredness'. Almost everybody mostly just talks about VC-funded start-ups. Meanwhile, often the most important trends are happening outside of those spaces.

Open Source's role in AI and robotics is a prime example of this. Players in the AI space are using it to 'poison pill' their competitors. Investors are pouring hundreds of billions into companies like OpenAI, but every time they have a chance to justify that cost with a revenue stream, someone ruins it by open-sourcing the tech and making it free-to-use.

These 20 academic institutions aren't motivated the same way, but they will have the same effect. Former robotics leaders like Boston Dynamics have lost their advantages. Now small companies in China's industrial zones have the same as them, but at no cost.

The result? Married to Chinese manufacturing, future robots will be cheap, ubiquitous, and it's likely no one Big Tech player will own the space. Will there be an Apple version of robotics? A company able to make hundreds of billions from high end products? Perhaps. But even if there is, most robots will still likely spring from open-source and many different manufacturers.

 

Press Release

Youtube Demo

It's only a matter of time before tech like this becomes widespread for grieving families. Eerie to think that in preparing for death some people will create their own afterlife personas to best comfort the people who care about them they are leaving behind.

More prosaically it makes a great alternative to regular phone messaging and emails. If you can trust the AI "You" to say the right thing in responses.

If you thought the proliferation of virtual boyfriends and girlfriends was sad already, prepare to get sadder.

 

Market share data courtesy of yipitdata.com.

There are others, but Waymo in the US and Badiu's Apollo Go in China, now seem ready for take-off with robo-taxis. From now on the only constraints to growth will be how quick they can deploy new vehicles to new markets. When this explosive growth is finished, there will be tens of millions of robo-taxis in every town and city on planet Earth.

The real revolution will be the global displacement of tens, perhaps hundreds, of millions of human driver jobs. We are rushing headlong into this future without anyone preparing for it, yet it's going to happen whether people like it or not, and it's heading straight for us.

[–] Lugh 8 points 1 year ago (3 children)

DK has both the research and manufacturing capacity to make these claims credible. Earlier this year they outlined an improved chemistry for lithium ion batteries that might boost their capacity between 10 and 40%.

The ceramics in these batteries are too delicate for batteries that large, even so small wearable devices that need to charge much less often will be very commercially popular.

[–] Lugh 7 points 1 year ago (1 children)

The human intestine is 6 meters long. It can be useful to locate problems with millimeter precision as this approach claims to do.

[–] Lugh 23 points 1 year ago (1 children)

Edible robots and robotic food — edible systems that perceive, process, and act upon stimulation — could open a new range of opportunities in health care, environmental management, and the promotion of healthier eating habits. For example, they could enable precise drug delivery and in vivo health monitoring, deliver autonomously targeted nutrition in emergencies, reduce waste in farming, facilitate wild animal vaccination

I think this is one of those ideas, that when you first hear about it you scratch your head thinking what on earth could that be useful for?, but then the more you think about it, actually these researchers have a point.

It would be silly to have large edible robots but what if the future is filled with trillions of tiny insect-sized robots? There are already drones being built this size. From that perspective, this makes more sense. For a start they are biodegradable. It gives them all sorts of uses in monitoring health and delivering medicine to animals. Suddenly you can have a whole layer of monitoring tiny robots in the environment and not have to worry about pollution when they come to the end of their useful life span. Not to mention this is a targeted way of delivering food to vulnerable species that may be affected by climate change emergencies.

[–] Lugh 15 points 1 year ago (2 children)

I have zero confidence in the promises of any of the big tech companies when it comes to privacy and AI.

[–] Lugh 5 points 1 year ago

I have a feeling it will be China who gets there first with mass-produced robots. They have the manufacturing base for it, nowhere else on the planet does near as much.

[–] Lugh 1 points 1 year ago* (last edited 1 year ago)

What concerns me is that a lot of these efforts seem to be political in nature and are tied to mitigating the inevitable decline in the fossil fuel industry. More often it makes more sense to speed up the use of renewables and dropping the use of fossil fuels. Fossil fuel use still hasn't peaked. That is mainly driven by China, who are still building new coal and gas electricity plants. However the peak year of fossil fuel use is very near, there is some speculation it may even be this year for oil use. From then on it will be in steady decline, so of course that industry is going to do everything they can to delay.

[–] Lugh 10 points 1 year ago* (last edited 1 year ago) (1 children)

This is interesting as it runs counter to what many people think about current AI. Its performance seems directly linked to the quality of the training data it has. Here the opposite is happening; it has poor training data and still outperforms humans. It's not surprising the humans would do badly in this situation too; it's hard to keep up to date on things that you may only encounter once or twice in your entire career. It's interesting to extrapolate from this observation as it applies to many other fields.

One of the authors of the paper goes into more detail on Twitter.

[–] Lugh 4 points 1 year ago (4 children)

Four months is a long time in 2020's AI development. OpenAI debuted Sora in February this year but hasn't publicly released it. Now a Chinese company called Kuaishou has got ahead of them with a model it calls Kling. Kuaishou is TikTok's biggest competitor in China and has a video-sharing app used by 200 million people. Presumably, that is where all its training data came from. Unlike OpenAI, Kling is available to some of the public.

This tech still doesn't look ready to level the TV and movie industry. It does 5-second clips, but who wants a 90-minute movie made up of nothing but 5-second clips?

[–] Lugh 1 points 1 year ago (1 children)

This technology is still very much at the proof of concept stage. What's fascinating is that they did not expect the neuron tissue to be healed in the way it was, and don't even understand how the stem cell robots did so.

That is a problem though. How do you develop the potential for something, when you don't really know what it may be able to do, or how it may be able to do it?

[–] Lugh 17 points 1 year ago (3 children)

Not sure I consider this futurology,

Yeah, its borderline. But I posted it as the the excuse/reasoning centers around AI. Microsoft's plan for 'Recall' are a huge invasion of privacy and stem from use of AI too. It's the topic of AI & privacy that merits discussion.

[–] Lugh 1 points 1 year ago

Although I wouldn't want the final decision to rest with AI, it makes a lot of sense that this is used for preparatory and research work leading to decisions.

[–] Lugh 1 points 1 year ago* (last edited 1 year ago) (1 children)

I gave you a shout out in the main sidebar as an extra thank you.

view more: ‹ prev next ›