https://www.youtube.com/watch?v=1cllCVK-9lo>One-shot learning via in-context prompting. The model learns new tasks in seconds when prompted with 3 to 12 seconds of a single demonstration, no training required. We refer to the use of sensorimotor examples in the context window as “physical prompting.”>Compositional generalization. Given two different physical prompts in context, the model chains them into a single longer-horizon behavior.>Zero-shot sim-to-real transfer. A demonstration recorded in simulation works as a physical prompt for a real-world task, even though pretraining contains no simulation data.>Human-to-robot imitation. In some cases a person can demonstrate a task with their own hands, in view of the robot’s cameras, and the model reproduces it with the robot’s hands.>Few-shot adaptation via gradient descent. The model can be fine-tuned to a new task in 1–10 gradient steps on 1–5 minutes of data (~10–50 demonstrations).>Improvising new strategies and tool use. The model generalizes at the level of behavioral strategies: forming entirely new trajectories to reach a goal, using unseen tools (e.g. brush, dustpan, etc.) to create new solutions to tasks demonstrated with other tools, and working ambidextrously even when prompted or fine-tuned to perform the task with a specific hand.
>The shouts of joy when the thing makes it into the bowl>Who cares that's a zero value operation.Legit. That's real science going on there, you have no idea.
>>541104005Imagine you don't need import millions of thirdies to do slave job
>>541103736isnt this purely explained by task simplicity Vector control is already hard solved so its like an ai reskinning some wireframe engine.
>Who cares lol it's just boxing and unboxing items, opening and closing containers bottles and lids, sweeping items into other places and stacking cups.https://www.youtube.com/watch?v=mhfleCK_IAIhttps://www.youtube.com/watch?v=geqQSuFHFfkOnce you've got a foundation and a proof of concept, the path to better than human manipulation, is just big compute.These problems are like chess go and jeopardy, the hardest part is often just the structure to iterate. They're not being forthcoming about which algorithms they're using to do this work as that is trade secrets. We would have to get inside as the model developer and say cut the shit with the PR moneybug impressing propaganda, show me the code, show me the pdf of phd whitepapers. What is the architecture, how is the human on computer training event getting transmitted to code, how is it consoldated, how does it flatten to wanted directives, and how does that bubble back up to directions on motion of robot hands.None of this information is in the video.
https://www.youtube.com/watch?v=YRNSJw4HWoUhttps://www.youtube.com/watch?v=VjIdoK0xM5Ehttps://www.youtube.com/watch?v=BFVe2vKhwRwhttps://www.youtube.com/watch?v=R3m4z9CoXEYhttps://www.youtube.com/watch?v=JMTSH1U44cwhttps://www.youtube.com/watch?v=aF77hGie_K0&t=522
>>541103736
Streamlining the gradient ascent step which is typically a 3 week to typically 3 month round trip operation to under five minutes, is a contribution to the field of strong artificial intelligence. My envy is a calling.https://www.youtube.com/watch?v=xUwIyy3PP24
>>541104143So why are they still importing them? We are getting the worst of all worlds both in AI and immigration
>>541105467Jews
>Real Time LearningWhere, exactly, will the camera observer looking over the shoulder of the drone to have learning happen to, reside and platform, in order for this so called: "one shot learning" event to occur real time, be seen by the observer, transfer to the data center for the learning event, and then back to the pilot to send to the drone for modificaation of behavior? The drone in the field doesn't have a mobile camera crew to watch as the work is being done for measure of correctness.https://www.youtube.com/watch?v=co02QtJP8Cg
>One shot learning.>no phd whitepapers with 141 iq noises in it.>no code>no description of architecture or approach.Fail. Fresh Fruits and Wildberry deficiency in the Ukrainian Donbass. This front line isn't moving at all. Also the nordstream pipeline is growing barnacles on the inside, you have any idea how many billions of dollars it's gonna take to clean that shit?https://www.youtube.com/watch?v=to8yh83jlXghttps://www.youtube.com/watch?v=RubSLGTrdOAhttps://www.youtube.com/watch?v=7FwDMzIbZrg
>>541104005today it will just stack things into a bowl but tomorrow it will give you an hand job AND tickle your balls at the same time. you have no idea.
>>541107470They've even got one for the Muslims at 0:29.
>>541103736basedfellow ai criticsbefore the computer god kcan teach YOUI stuff, YOU ( everyone) must TEACH the computertrus tthe plan, UBI in two weeks
>>541107470creepy
>>541108749what the fuck lol xDhow can aIIah allow this mockery xDDDD
>>541103736So what happens when context fills up?
>>541103736>actual interesting stuff instead of flashy nothingburger.Thanks anon.
>PHD in cs and ml getting excited about one robot hand removing a stuck lego from its other hand and it falling into the bowl as well.Legit. None of you dumbasses understands.https://www.youtube.com/watch?v=Vzd3zyZcves
