Tech
DeepMind introduces AI agent that learns to complete various tasks in a scalable world model
Over the past decade, deep learning has transformed how artificial intelligence (AI) agents perceive and act in digital environments, allowing them to master board games, control simulated robots and reliably tackle various other tasks. Yet most of these systems still depend on enormous amounts of direct experience—millions of trial-and-error interactions—to achieve even modest competence.
This brute-force approach limits their usefulness in the physical world, where such experimentation would be slow, costly, or unsafe.
To overcome these limitations, researchers have turned to world models—simulated environments where agents can safely practice and learn.
These world models aim to capture not just the visuals of a world, but the underlying dynamics: how objects move, collide, and respond to actions. However, while simple games like Atari and Go have served as effective testbeds, world models still fall short when it comes to representing the rich, open-ended physics of complex worlds like Minecraft or robotics environments.
Researchers at Google DeepMind recently developed Dreamer 4, a new artificial agent capable of learning complex behaviors entirely within a scalable world model, given a limited set of pre-recorded videos.
The new model, presented in a paper published on the arXiv preprint server, was the first artificial intelligence (AI) agent to obtain diamonds in Minecraft without practicing in the actual game at all. This remarkable achievement highlights the possibility of using Dreamer 4 to train successful AI agents purely in imagination—with important implications for the future of robotics.
“We as humans choose actions based on a deep understanding of the world and anticipate potential outcomes in advance,” Danijar Hafner, first author of the paper, told Tech Xplore.
“This ability requires an internal model of the world and allows us to solve new problems very quickly. In contrast, previous AI agents usually learn through brute-force with vast amounts of trial-and-error. But that’s infeasible for applications such as physical robots that can easily break.”
Some of the AI agents developed at DeepMind over the past few years have already achieved tremendous success at games such as Go and Atari by training in small world models. However, the world models that these models relied on failed to capture the rich physical interactions in more complex worlds, such as the Minecraft videogame.
On the other hand, “Video models such as Veo and Sora are rapidly improving towards generating realistic videos of very diverse situations,” said Hafner.
“However, they are not interactive, and their generations are too slow, so they cannot be used as ‘neural simulators’ to train agents inside of yet. The goal of Dreamer 4 was to train successful agents purely inside of world models that can realistically simulate complex worlds.”
Hafner and his colleagues decided to use Minecraft as a test bed for their AI agent, as it is a complex video game that contains infinite generated worlds and long-horizon tasks that require over 20,000 consecutive mouse/keyboard actions to be completed.
One of these tasks is the mining of diamonds, which requires the agent to perform a long sequence of prerequisites such as chopping trees, crafting tools, and mining and smelting ores.
Notably, the researchers wanted to train their agent purely in “imagined” scenarios, instead of allowing it to practice in the actual game, analogous to how smart robots will have to learn in simulation, because they could easily break when practicing directly in the physical world . This requires the model to learn object interactions in an accurate enough internal model of the Minecraft world.
The artificial agent developed by Hafner and his colleagues is based on a large transformer model that was trained to predict future observations, actions and the rewards associated with specific situations. Dreamer 4 was trained on a fixed offline dataset containing recorded Minecraft gameplay videos collected by human players.
“After completing this training, Dreamer 4 learns to select increasingly better actions in a wide range of imagined scenarios via reinforcement learning,” said Hafner.
“Training agents inside of scalable world models required pushing the frontier of generative AI. We designed an efficient transformer architecture, and a novel training objective named shortcut forcing. These advances enabled accurate predictions while also speeding up generations by over 25x compared to typical video models.”
Dreamer 4 is the first AI agent to obtain diamonds in Minecraft when trained solely on offline data, without ever practicing its skills in the actual game. This finding highlights the agent’s ability to autonomously learn how to correctly solve complex and long-horizon tasks.
“Learning purely offline is highly relevant for training robots that can easily break when practicing in the physical world,” said Hafner. “Our work introduces a promising new approach to building smart robots that do household chores and factory tasks.”
In the initial tests performed by the researchers, the Dreamer 4 agent was found to accurately predict various object interactions and game mechanics, thus developing a reliable internal world model. The world model established by the agent outperformed the models that earlier agents relied on by a significant margin.
“The model supports real-time interactions on a single GPU, making it easy for human players to explore its dream world and test its capabilities,” said Hafner. “We find that the model accurately predicts the dynamics of mining and placing blocks, crafting simple items, and even using doors, chests, and boats.”
A further advantage of Dreamer 4 is that it achieved remarkable results despite being trained on a very small amount of action data. This is essentially video footage showing the effects of pressing different keys and mouse buttons within the Minecraft videogame.
“Instead of requiring thousands of hours of gameplay recordings with actions, the world model can actually learn the majority of its knowledge from video alone,” said Hafner.
“With only a few hundred hours of action data, the world model then understands the effects of mouse movement and key presses in a general way that transfers to new situations. This is exciting because robot data is slow to record, but the internet contains a lot of videos of humans interacting with the world, from which Dreamer 4 could learn in the future.”
This recent work by Hafner and his colleagues at DeepMind could contribute to the advancement of robotics systems, simplifying the training of the algorithms that allow them to reliably complete manual tasks in the real world.
Meanwhile, the researchers plan to further improve Dreamer 4’s world model, by integrating a long-term memory component. This would ensure that the simulated worlds in which the agent is trained remain consistent over long periods of time.
“Incorporating language understanding would also bring us closer towards agents that collaborate with humans and perform tasks for them,” added Hafner.
“Finally, training the world model on general internet videos would equip the agent with common sense knowledge of the physical world and allow us to train robots in diverse imagined scenarios.”
Written for you by our author Ingrid Fadelli, edited by Sadie Harley, and fact-checked and reviewed by Robert Egan—this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive.
If this reporting matters to you,
please consider a donation (especially monthly).
You’ll get an ad-free account as a thank-you.
More information:
Danijar Hafner et al, Training Agents Inside of Scalable World Models, arXiv (2025). DOI: 10.48550/arxiv.2509.24527
© 2025 Science X Network
Citation:
DeepMind introduces AI agent that learns to complete various tasks in a scalable world model (2025, October 25)
retrieved 25 October 2025
from https://techxplore.com/news/2025-10-deepmind-ai-agent-tasks-scalable.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.
I’ve always approached taking melatonin supplements with skepticism. They seem to help every once in a while, but your brain is already making melatonin. Beyond that, I am not a fan of the sickly-sweet tablets, gummies, and other forms of melatonin I’ve come across. No one wants a bad taste in their mouth when they’re supposed to be drifting off to sleep.
This is where Onnit’s Instant Melatonin Spray comes in. Fellow WIRED reviewer Molly Higgins first gave it a go, and reported back favorably. This spray comes in two flavors, lavender and mint, and is sweetened with stevia. While I wouldn’t consider it a gourmet taste, I appreciate that it leans more into herbal components known for sleep and relaxation.
-
Photograph: Molly Higgins
-
Onnit
Instant Melatonin Mist
Keep in mind that melatonin is meant to be a sleep aid, not a cure-all. That being said, one serving of this spray has 3 milligrams of melatonin, which takes about six pumps to dispense. While 3 milligrams may not seem like a lot to really kickstart your circadian rhythm, it’s actually the ideal dosage to get your brain’s wind-down process kicked off. Some people can do more (but don’t go over 10 milligrams!), some less, but based on what experts have relayed to me, this is the preferable amount.
A couple of reminders for any supplement: consult your doctor if and when you want to incorporate anything, melatonin included, into your nighttime regimen. Your healthcare provider can help confirm that you’re not on any medications where adding a sleep aid or supplement wouldn’t feel as effective. Onnit’s Instant Melatonin Spray is International Genetically Modified Organism Evaluation and Notification certified (IGEN) to verify that it uses truly non-GMO ingredients.
Apart from that, there may be some trial and error on the ideal amount for you, and how much time it takes to kick in. Some may feel the melatonin sooner than others. For my colleague Molly, it took about an hour. Melatonin can’t do all the heavy lifting, so make sure you’re ready to go to bed when you take it, and that your sleep space is set up for sleep success, down to your mattress, sheets, and pillows.
There’s a lever on the back for this compression mechanism that you manually press down and a separate button to open the dustbin at the bottom. You can use the compression lever when it’s both closed and open. It did help compress the hair and dust while I was vacuuming, helping me see if I had really filled the bin, though at a certain point it doesn’t compress much more. It was helpful to push debris out if needed too, versus the times I’ve had to stick my hand in both the Dyson and Shark to get the stuck hair and dust out. Dyson has this same feature on the Piston Animal V16, which is due out this year, so I’ll be curious to see which mechanism is better engineered.
Bendable Winner: Shark
Photograph: Nena Farrell
If you’re looking for a vacuum that can bend to reach under furniture, I prefer the Shark to the Bosch. Both have a similar mechanism and feel, but the Bosch tended to push debris around when I was using it with an active bend, while the Shark managed to vacuum up debris I couldn’t get with the Bosch without lifting it and placing it on top of that particular debris (in this case, rogue cat kibble).
Accessory Winner: Dyson
Dyson pulls ahead because the Dyson Gen5 Detect comes with three attachments and two heads. You’ll get a Motorbar head, a Fluffy Optic head, a hair tool, a combination tool, and a dusting and crevice tool that’s actually built into the stick tube. I love that it’s built into the vacuum so that it’s one less separate attachment to carry around, and it makes me more likely to use it.
But Bosch does well in this area, too. You’ll get an upholstery nozzle, a furniture brush, and a crevice nozzle. It’s one more attachment than you’ll get with Shark, and Bosch also includes a wall mount that you can wire the charging cord into for storage and charging, and you can mount two attachments on it. But I will say, I like that Shark includes a simple tote bag to store the attachments in. The rest of my attachments are in plastic bags for each vacuum, and keeping track of attachments is the most annoying part of a cordless vacuum.
Build Winner: Tie
Photograph: Nena Farrell
All three of these vacuums have a good build quality, but each one feels like it focuses on something different. Bosch feels the lightest of the three and stands up the easiest on its own, but all three do need something to lean against to stay upright. The Dyson is the worst at this; it also needs a ledge or table wedged under the canister, or it’ll roll forward and tip over. The Bosch has a sleek black look and a colorful LED screen that will show you a picture of carpet or hardwood depending on what mode it’s vacuuming in. The vacuum head itself feels like the lightest plastic of the bunch, though.
Tech
Right-Wing Gun Enthusiasts and Extremists Are Working Overtime to Justify Alex Pretti’s Killing
Brandon Herrera, a prominent gun influencer with over 4 million followers on YouTube, said in a video posted this week that while it was unfortunate that Pretti died, ultimately the fault was his own.
“Pretti didn’t deserve to die, but it also wasn’t just a baseless execution,” Herrera said, adding without evidence that Pretti’s purpose was to disrupt ICE operations. “If you’re interfering with arrests and things like that, that’s a crime. If you get in the fucking officer’s way, that will probably be escalated to physical force, whether it’s arresting you or just getting you the fuck out of the way, which then can lead to a tussle, which, if you’re armed, can lead to a fatal shooting.” He described the situation as “lawful but awful.”
Herrera was joined in the video by former police officer and fellow gun influencer Cody Garrett, known online as Donut Operator.
Both men took the opportunity to deride immigrants, with Herrera saying “every news outlet is going to jump onto this because it’s current thing and they’re going to ignore the 12 drunk drivers who killed you know, American citizens yesterday that were all illegals or H-1Bs or whatever.”
Herrera also referenced his “friend” Kyle Rittenhouse, who has become central to much of the debate about the shooting.
On August 25, 2020, Rittenhouse, who was 17 at the time, traveled from his home in Illinois to a protest in Kenosha, Wisconsin, brandishing an AR-15-style rifle, claiming he was there to protect local businesses. He killed two people and shot another in the arm that night.
Critics of ICE’s actions in Minneapolis quickly highlighted what they saw as the hypocrisy of the right’s defense of Rittenhouse and attacks on Pretti.
“Kyle Rittenhouse was a conservative hero for walking into a protest actually brandishing a weapon, but this guy who had a legal permit to carry and already had had his gun removed is to some people an instigator, when he was actually going to help a woman,” Jessica Tarlov, a Democratic strategist, said on Fox News this week.
Rittenhouse also waded into the debate, writing on X: “The correct way to approach law enforcement when armed,” above a picture of himself with his hands up in front of police after he killed two people. He added in another post that “ICE messed up.”
The claim that Pretti was to blame was repeated in private Facebook groups run by armed militias, according to data shared with WIRED by the Tech Transparency Project, as well as on extremist Telegram channels.
“I’m sorry for him and his family,” one member of a Facebook group called American Patriots wrote. “My question though, why did he go to these riots armed with a gun and extra magazines if he wasn’t planning on using them?”
Some extremist groups, such as the far-right Boogaloo movement, have been highly critical of the administration’s comments on being armed at a protest.
“To the ‘dont bring a gun to a protest’ crowd, fuck you,” one member of a private Boogaloo group wrote on Facebook this week. “To the fucking turn coats thinking disarming is the answer and dont think it would happen to you as well, fuck you. To the federal government who I’ve watched murder citizens just for saying no to them, fuck you. Shall not be infringed.”
Budget 2026 Live Updates: TCS On Overseas Tour Packages Slashed To 2%; TDS On Education LRS Eased
Trump says India will buy oil from Venezuela, not Iran
Budget 2026: Cabinet gives green signal to Union Budget 2026–27
-
Business1 week agoSuccess Story: This IITian Failed 17 Times Before Building A ₹40,000 Crore Giant
-
Fashion1 week agoSouth Korea tilts sourcing towards China as apparel imports shift
-
Sports1 week agoTransfer rumors, news: Saudi league eyes Salah, Vinícius Jr. plus 50 more
-
Sports5 days agoPSL 11: Local players’ category renewals unveiled ahead of auction
-
Entertainment1 week agoThree dead after suicide blast targets peace committee leader’s home in DI Khan
-
Sports1 week agoWanted Olympian-turned-fugitive Ryan Wedding in custody, sources say
-
Tech1 week agoStrap One of Our Favorite Action Cameras to Your Helmet or a Floaty
-
Tech1 week agoThis Mega Snowstorm Will Be a Test for the US Supply Chain