Highlights • Demonstration of an intelligent control algorithm running on hardware suitable for spaceflight. • Proposed method combines optimal control demonstrations with reinforcement learning. • Q-network updates its weights online using a novel update mechanism called Extreme Q-Learning Machine. • Reinforcement learning agent applied to a simulated Mars lander environment.
Introducing GAIA-1: A Cutting-Edge Generative AI Model for AutonomyGeneration of multiple plausible futuresWhen we prompt GAIA-1 with a few seconds of starting context, it can imagine several possible futures. This shows that based on the same video prompt, GAIA-1 understands that many different things can happen in the future.This video shows that GAIA-1 can be prompted with a couple of seconds of the same starting context. Then it can unroll multiple possible futures. For example, - turning left at the intersection (top left frame) or - continuing straight (top right frame). - Or even continuing straight along a different imagined road (bottom left frame). - It can also imagine a pedestrian crossing the road ahead of us (bottom right frame).
This was weird and trippy*snip youtube vid*As Musk recently mentioned on Maher's show, people like Kurzweil believe AI could give us immortality, by improving upon the limitations of our biology.Could AI help us to plan and manage terraforming?Could AI give us Grand Unification Theory? Could AI find cracks in the laws of physics, could it expose flaws in the Standard Model and then come up with the answers?Could AI give us FTL?
Quote from: sanman on 05/20/2023 09:06 pmThis was weird and trippy*snip youtube vid*As Musk recently mentioned on Maher's show, people like Kurzweil believe AI could give us immortality, by improving upon the limitations of our biology.Could AI help us to plan and manage terraforming?Could AI give us Grand Unification Theory? Could AI find cracks in the laws of physics, could it expose flaws in the Standard Model and then come up with the answers?Could AI give us FTL?AI as it currently is (highly-specialized-task-performing algorithms) No. Current AI are only as good as the databases they are trained on. They don't actually understand the things that are in that database.
Industrial software provider EOT has launched a generative AI tool for manufacturing. The company’s new Twin Talk GPT product is designed to allow manufacturing and energy companies to train and test predictive maintenance models at the industrial edge...The operational data generation allows training of predictive maintenance machine learning models that perform real-time predictions and anomaly detection at the edge of operation centres, using closed-loop, event-response operational action to instantly avoid expensive downtime and increase production output.“We are thrilled to introduce Twin Talk GPT to the market,” said Matt Oberdorfer, CEO of EOT. “Our new product is a fundamental game-changer for the industrial sector, enabling companies to leverage the power of generative AI to improve predictive maintenance systems with the help of generative pre-trained transformers.”
Quote from: whitelancer64 on 06/19/2023 05:48 pmQuote from: sanman on 05/20/2023 09:06 pmThis was weird and trippy*snip youtube vid*As Musk recently mentioned on Maher's show, people like Kurzweil believe AI could give us immortality, by improving upon the limitations of our biology.Could AI help us to plan and manage terraforming?Could AI give us Grand Unification Theory? Could AI find cracks in the laws of physics, could it expose flaws in the Standard Model and then come up with the answers?Could AI give us FTL?AI as it currently is (highly-specialized-task-performing algorithms) No. Current AI are only as good as the databases they are trained on. They don't actually understand the things that are in that database.More generally, the meaning of "AI" evolves more or less continuously, so older predictions by pundits tend to be out of date and often meaningless. If "AI" ever becomes capable of designing and implementing improved AI, then the result is unpredictable and you can speculate all you want about the outcome, from making all humans immortal and unimaginably wealthy, to killing off all humans. https://en.wikipedia.org/wiki/Technological_singularityNote that the demand for computer programmers is currently decreasing for the first time ever, as GPT systems have radically increased the productivity of programmers. So what happens when GPT models get better at optimizing existing code? Coding is the easiest part of the software development process, but what happens when the GPT models start handling more of the complex parts?
I've actually used Chat GPT to generate code. It's okay, but not terribly great at it.
Quote from: whitelancer64 on 06/19/2023 07:21 pmI've actually used Chat GPT to generate code. It's okay, but not terribly great at it. Let us know when you level up.*yeet video*
Take a look at this:*yeet video*They again mention using the "random perturbations" to essentially explore various permutations and combinations of action, to pursue brute-force discovery of best modes of action.Can we use such techniques to help us design off-world machines that will operate more robustly and efficiently, and be more fault-tolerant?Can we design and test a machine virtually, perhaps subjecting it to virtual testing that vastly exceeds its actual intended operating lifespan.
A more comprehensive overview of some of the self-taught skills from Google DeepMind's robot soccer experiment:*yeet youtube videos*When AI & Machine Learning are now being touted far and wide as the tech-du-jour, then why aren't we hearing more from NASA on this front?Is it because NASA has been on the bleeding edge for so long, that from their viewpoint, this is all just old hat? Been there, done that?How is NASA currently making use of Machine Learning, and what are its future plans for making use of it?
https://twitter.com/Tesla_AI/status/1671589496947712005
...[Tesla's new v12] foundation [vision world model is] shared among the two applications, [Tesla autopilot and Tesla Optimus bot.]
Foundation models:We define foundation models as models trained on broad data (generally using self-supervision at scale) that can be adapted to a wide range of downstream tasks. These models, which are based on standard ideas in transfer learning and recent advances in deep learning and computer systems applied at a very large scale, demonstrate surprising emergent capabilities and substantially improve performance on a wide range of downstream tasks. Given this potential, we see foundation models as the subject of a growing paradigm shift, where many AI systems across domains will directly build upon or heavily integrate foundation models.
I only speculate this work applies to the use-case being discussed here, but still, more generally I think the next big step with these large language models is for them to incorporate some level of self-supervised learning, in the style of Deepmind's alphaZero.
And now we see that Tesla AI also has a generative prediction engine, similar to GAIA-1. Given Tesla's commitment to Dojo AI supercomputing, @Tesla_AI is a space to watch. Who would be surprised if SpaceX someday buys time on Dojo for cutting-edge AI space applications?
I certainly would be surprised, since the data Tesla is training their A.I. with is visual data that comes from the car cameras they can access on the cars they have sold.What data is SpaceX supposed to be using to train an A.I., and for what purpose?A.I. is a tool, but it isn't the right tool for every problem that needs to be solved...
Using Reinforcement Learning to figure out from scratch how to land Falcon boosters:<snip>
A while back I made a project where Reinforcement Learning was used in order to land SpaceX booster rockets. How the rocket works, and how the AI was trained is explained here: To summarize, in the previous project, the rocket had 4 cold gas thrusters on the top in order to orienate itself (right, left, forward, backwards). Additionally, it could utilize thrust vectoring (10degrees in each direction) for more directional control.I got an interesting suggestion from user "sanman" to completely disable one of its key components, in this case the cold gas thruster places on the right of the rocket, and see if the AI could overcome this limitation. To increase the challenge, I also limited the thrust vectoring to 5 degrees in each direction.After ~6 hours of training, starting with the previous model, it seems that the AI was able to handle the situation pretty well. It looks like it tries to alter its incoming trajectory so that the 3 remainig gas thrusters are more effective. Some of the "harder" starting positions result in a rocky landing, however when observing the relevant graphs from the training showed that there was still improvement to be learned. Meaning that letting it train for longer, the landing velocity, as well as the distance to the center, would be lower. (Let me know if you want to see the graphs)Also important to note is that I thought it would be fun to add a negative reward for each timestep, meaning that the rocket ideally wants to land as fast as possible.Probably goes without saying: this is not a realistic simulation, more or less a fun side project. Might improve the realism in later additions.If you have any questions, the video I mentioned covers more details, otherwise just let me know
Quote from: sanman on 05/20/2023 09:06 pm...As Musk recently mentioned on Maher's show, people like Kurzweil believe AI could give us immortality, by improving upon the limitations of our biology.Could AI help us to plan and manage terraforming?Could AI give us Grand Unification Theory? Could AI help to map the contours of the laws of physics, scour them for chinks and cracks? Could it expose flaws in the Standard Model and then come up with the answers?Could AI give us FTL?Hypothetical super intelligent general AI could help with any task that more intelligence could help with.
...As Musk recently mentioned on Maher's show, people like Kurzweil believe AI could give us immortality, by improving upon the limitations of our biology.Could AI help us to plan and manage terraforming?Could AI give us Grand Unification Theory? Could AI help to map the contours of the laws of physics, scour them for chinks and cracks? Could it expose flaws in the Standard Model and then come up with the answers?Could AI give us FTL?
