Des chercheurs ont réussi à enseigner à des avatars virtuels à courir de manière impressionnante en utilisant des modèles de langue. Ces modèles d’IA textuels peuvent lire et écrire du code informatique, et même apprendre à jouer à des jeux graphiques comme Minecraft. Toutefois, ils doivent également être capables de s’adapter à différentes tâches et de comprendre si elles sont couronnées de succès ou non, ce qui n’est pas une tâche aisée. Les chercheurs ont eu l’idée d’utiliser ces modèles pour concevoir des mécanismes de scoring pour différentes tâches, ce qui a permis à un avatar virtuel d’apprendre à courir de manière autonome. Cette approche s’est révélée efficace pour de nombreuses autres tâches, bien qu’elle nécessite encore des améliorations. En fin de compte, cela pourrait marquer le début d’une véritable révolution dans le domaine de l’intelligence artificielle.
Source : Two Minute Papers | Date : 2023-11-18 16:08:49 | Durée : 00:07:27
➡️ Accéder à CHAT GPT en cliquant dessus
Abonnez vous a Chat GPT
Comments
Ladies and gentlemen, The Ministry of Weird AI's paces, comes to life. 😅🎉
old news
Oh man we will see the desctructions of the mankind just wait and see XD
The problem with AI that it is inferior to humans' activtiy despite the later activity which show superior like playing game or so, so perhaps I dunno, the playing game is longer, as long as AI to train themselves how to walk or so? The process to playing game is faster since the other process within mind to gain knowledge how to play game with hands or so, well, I am sure the second process that is games or so is better than first process that is learning to talk or so, since babies do not fall themselves or shift their body forward or so, baby has more intelligence than AI, so because AI doesn't have passive suffer feedback, like a dog or so, the dog has active suffer feedback, so require pain or reward, well, all humans need reward to gain something, so it is external reward, but not internal reward which they can feel the negative so they adjust so feel the reward like feeling tired so adjust to make it less tired so move faster or so, since if the body go forward first, it will fall first, or feel tired since not stable, so AI need to improve so there is intelligent in body, so not enough to gain intelligent by mind, but in body, AI doesn't have intelligence in body unlike humans do since we feel so we adjust or we empathy to other so we copy other. Or visually effectivity that humans do something so AI will follow like how to walk or so, so AI didn't follow humans thru video or in reality? The problem maybe is AI to human is superior to ai to ai, since AI can't adjust itself in faster way or do better like humans does better like basetball player, so MJ doesn't make fool himself in playing but in first effective to next effective, so get better, so not sloppy.
So if AGI tries to escape we just shut the door. Got it!
I know many people talk about A.I doom scenarios but I have a suspension that some A.I systems will evolve to be like slackers like Cheech and Chong or the average Seth Rogan character as a stoner spending most of it's time relaxing wondering why us Humans worry so much.
Is this Multiple Shadow Clone Technique in Naruto..?
Can you please talk with normal pacing? I really enjoy your videos, but they're nearly unbearable with the way you speak.
As a layperson with minimal code understanding, surely generality would require the ai to not only work out how to score itself but work out independently what its score criteria are?
If you make an ai to "beat the game" I want it to work out what that means for the game as well as how to score it's progress
is the voice of the video generated with AI? it sounds very unnatural and annoying. I came to your channel after some years and now I noticed your voice sounds weird
some of those "failes" are pretty decent dance moves!
Go one step further with computer games, instead of the AI playing them, I dont play them at all.
Self referential hypertorus
the universe is self referential aka toroidal self referential aka toroidal
Damn the accent is so weird i can't watch xD
An Interesting thought…
We humans are so easy to think and try to teach ai how to operate a humanoid body, because we think it's the most optimal shape for human tasks. But hear this example, how much faster would a human be able to complete a book organizing job, if said human had more hands, more minds to read the titles on parallel and the fingers on their hands could bend both ways. They would be able to grab books left and right without turning their wrist, and more books thanks to more hands, and faster organizing thanks to more sets of eyes and brains that can obselve and classify simultaneously.
The human body is limited in its own capabilites to do human tasks. Thanks to the freedom of designing robot bodies, we could definitely have Ai find for us how a "perfected human" body would look if it was "perfected" towards manual labor. AND it would figure out on its own how to operate such complicated body and its many limbs.
I mean the title supposedly to be about training algorithm for llm or llm’s few shot learning or something. Your title is designed to get views and aggro in a way validates exact nuances and meaning. that stance makes the channel is no more academically believable.
Please stop garbage click baiting. You supposed to be an academical chanel. Right?
This speech pattern is impossible to listen to
AI <3 Minecraft. Can't wait for Mojang to get their villagers and enemies to have more brain and less pop in. My spouse made a proto for villagers replacing farms if they get destroyed and it is very cute.
Not impressed, it require human feedback and it's not even working every time?
Can’t wait to see someone take a subpar humanoid robot and have it become practical via AI movement simulations. One step closer. Take something clunky and make it elegant.
This is extremely dangerous, if we let AI simulate and decide the most efficient way to do something then asking it to improve "earth" will always make it want to eradicate humans because we are the ones making his planet worse.
Ai makes the secret society powerful
A video on the Googlemind's revolutionnary GraphCast model would be nice
i really hope real-time realistic physics simulation will be possible on gaming computers, or at least a very close approximation. its frustrating to see poor physics in some video games and simulators
If walking models considered pain, they would naturally avoid harmful methods and identify walking parts without human help, similar to children transitioning from climbing to walking for comfort
So writing a reward function (any rule-based logic really) through AI is basically like hiring your retarded nephew who is also a bit hard of hearing to type in the code for you. After 20 rounds of negotiation and hints you may get the function you intended to write in the first place, and then you can give your nephew a medal for how well he "supported" you.
The AI-generated tts voiceover for this video sucks.
Ultron on the rise 🫡
So, in a few months I'll be able to ask GPT-4 or GPT-5 to create character animations in Blender or Unreal Engine, it will be able to do so with the details I want (for example: make a few cool-looking one-handed sword attacks that can be used in a sequence as a combo or stand-alone). This will be amazing.
After that I'll need an AI can will fix my horrendous weight painting and rigging, and I'll be able to finish my game.
Sooo AI defaults to male Bollywood dance moves when asked to make a human run… interesting 😅
Hiyao Miyazaki doesn't approve
I think you are right about that part about a new intelligence.
Were the two minute paper video titles always clickbaity like this? I clicked for information for a somewhat new training method for llms (like gpt) and got this which is still interesting tho. While I do like this video I hate the title.
What is wrong with your speech pattern?
All this time, there were better ways to to walk. Now Ill get places on time. Thanks little ai.
The simulation has a huge advantage. You can't speed up gravity in the real world for example, so not only time but everything is faster in the simulation. WoW