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.
If you want to get into stargazing in 2025, there’s still a chance to catch some of the best meteor showers of the year. Also known as shooting stars, meteors happen when Earth’s orbital path crosses a path of debris left by a comet and that material burns up in the Earth’s atmosphere. Watching a meteor shower is one of the most accessible ways to engage with the night sky.
The next shower are the Geminids, a busy and bright shower that peaks in mid-December, offering the chance to see hundreds of shooting stars each hour. This is just one of nine major meteor showers that grace skies throughout the year, and details of when they will appear in the northern hemisphere are listed below—so mark your 2026 calendar for these.
The Next Big Meteor Shower: The Geminids
The Geminids are active from about December 4 to December 17, peaking overnight from December 13 to December 14. They have a sharp peak, so the night of the 13th is the best time for skywatching.
The Geminids are the most spectacular meteor shower of the year. In addition to boasting up to 120 or even 150 meteors per hour during its peak, this meteor shower is also the brightest and most colorful of the year.
The Geminids are bright, slow-moving meteors that often have yellow tones, but they can be a range of other colors, including green, blue, white, red, or orange. And unlike most meteors, which are caused by comet debris, the Geminids are the remnants of an asteroid.
The night that the Geminids peak, their radiant, the constellation Gemini, will be above the horizon all night and will reach its highest point around 2 am local time, so meteors will be visible almost the whole night.
That same night, the moon will be about 32 percent illuminated and will rise around 1:30 am in the eastern US, so if you watch this shower shortly after midnight, the moonlight won’t interfere with your viewing experience.
How to Watch a Meteor Shower
You don’t need any special equipment to see a meteor shower—in fact, using devices like binoculars or telescopes actually prevents you from seeing meteors, because they travel too fast to be seen through the lenses of such equipment. All you need are your eyes, a dark sky with little to no moonlight, and a location that’s away from excess light, as moonlight and light pollution can wash out shooting stars.
Note that the moon appears (rises) and disappears (sets) in the night sky at different times depending on what time zone you are in. All moonrise/moonset times in this piece are for the eastern US. You can use tools like Time and Date’s moonrise/moonset calendar or this tool from the US Naval Observatory to check the precise moonrise/moonset times in your exact location.
B&H Photo is one of our favorite places to shop for camera gear. If you’re ever in New York, head to the store to check out the giant overhead conveyor belt system that brings your purchase from the upper floors to the registers downstairs (yes, seriously, here’s a video). Fortunately B&H Photo’s website is here for the rest of us with some good deals on photo gear we love.
Save on the Latest Gear at B&H Photo
B&H Photo has plenty of great deals, including Nikon’s brand-new Z6III full-frame mirrorless camera for $2,196.95 ($300 off). The Z6III (8/10, WIRED Recommends) is huge update for the Z6 line, and a great all-around camera for Nikon fans. There are also great deals like the Nikon Zf (8/10, WIRED Recommends) full frame mirrorless camera for $200 off, or the Nikon Z8 for $600 off.
Not in the market for the latest Nikon cameras? There are plenty of other deals, like our favorite action camera, the GoPro Hero 13 for $70 off (7/10, WIRED Recommends), and Sony’s a7R V full frame mirrorless camera for $400 off.
How Do I Get Free Shipping at B&H?
One of the many perks of B&H Photo is their generous free shipping policy. Most orders at B&H over $49 qualify for free expedited shipping to the lower 48 states. The savings just keep coming, as most items totaling under $49 also qualify for free standard shipping in the contiguous US.If for some reason your order doesn’t qualify for free shipping, you can review alternative shipping options during checkout—full shipping policy details can be found here.
Best Times to Find B&H Promo Codes and Discounts
Like most retailers, B&H Photo offers some of their best deals of the year during Black Friday and Cyber Monday, but that doesn’t mean there aren’t other ways to save outside of this sale period. There’s still time to jump on extended holiday sales, with up to $300 24-hour camera discounts, and featured price drops across all accessories. While there may not be any B&H promo codes available at the moment, there are plenty of other ways to save at B&H.
Get the latest B&H Preorder in 2025
If you’re a photographer, videographer, or creator, you’re not going to want to be left out of some of the most anticipated camera releases of the year, including the Canon EOS R6 Mark III Mirrorless Camera, Canon 45mm f/1.2 STM Lens (Canon RF), and Canon EOS R6 Mark III Mirrorless Camera with 24-105mm f/4 Lens, and more. Make sure you visit the above link to get the full scoop of camera details and releases, as well as pre-order them while you still can.
Exclusive B&H Student Discounts
If you’re a student with an EDU email, sign up for B&H Photo’s student discount program, which offers free shipping on most orders and exclusive discounts.
Score Deals on Used and Refurbished Gear
B&H Photo also deals in used and refurbished gear. Deals (and conditions) vary, and I have never purchased a used item this way, but if you’re looking to save some money, that’s another way to go.
Trade-in Your Old Gear at B&H
The flip side of B&H Photos used deals is that you can sell your old gear. I put in my old Sony a7 II and was offered $210, which is more than I would have thought. Your offer is contingent on it matching the condition you claim, but if you’ve got gear you aren’t using anymore, this is a way to turn it into some extra cash.
Snap Up Limited-Time Deals in the Deal Zone
There are always rotating and limited-time B&H Photos deals at this url, which B&H Photo calls the Deal Zone.
President Donald Trump signed a highly anticipated executive order on Thursday that sets in motion a plan to establish a national regulatory framework for artificial intelligence while undercutting states’ abilities to enact their own rules.
The order, titled “Ensuring a National Policy Framework for Artificial Intelligence,” creates an AI litigation task force within the Justice Department to directly challenge state AI laws the administration finds to conflict with federal policy. It also directs the Department of Commerce to craft guidelines that could make states ineligible for future broadband funding if they pass “onerous” AI laws.
The push for sweeping federal preemption of state AI laws has largely been fueled by AI investors, conservative policy shops, and tech industry trade groups. These groups have argued that a patchwork approach to AI regulation could stunt Silicon Valley’s AI boom and reduce America’s competitiveness on the global stage. White House AI and crypto adviser David Sacks has been one of the most vocal proponents of a light-touch approach to AI regulation.
“The EO gives your administration tools to push back on the most onerous and excessive state regulations,” Sacks told Trump during Thursday’s signing ceremony. “We’re not going to push back on all of them. For example, kids safety we’re going to protect.”
The order is similar in many respects to an earlier draft obtained by WIRED but with a few key differences. The executive order instructs Sacks and Michael Kratsios, the assistant to the president for science and technology, to prepare a legislative recommendation establishing a federal policy framework for AI. One of the new additions is a carve-out within this legislative recommendation asking Congress not to preempt state AI laws that aim to protect children, promote data center infrastructure, and encourage state governments to procure AI tools.
“We want one central source of approval, and we have great Republican support. I think we probably have Democrat support too, because it’s common sense,” Trump said during Thursday’s signing ceremony. “Every time you make a change, and it could be a very reasonable change, you still won’t get it approved if you have to go to 50 states. This centralizes it.”
In the absence of federal regulations, officials from states across the country have pushed through their own investigations and legislation to govern the use and development of AI. Trump’s executive order specifically calls out certain state AI laws—such as Colorado’s SB24-205, which aims to limit “algorithmic discrimination” in AI models—as an attempt to “embed ideological bias.”
Several other state AI laws may also fall in the crosshairs of this executive order. California governor Gavin Newsom signed a law in September requiring large tech companies to publish safety frameworks around their AI models. In June, New York’s legislature passed a bill that would empower the state’s attorney general to bring civil penalties of up to $30 million against AI developers that fail to meet safety standards. That bill is currently sitting on New York governor Kathy Hochul’s desk, awaiting her signature or veto—though she’s reportedly considering amendments that could weaken the bill significantly.
Copra price boost: Govt hikes Copra MSP for 2026 season; farmers to get up to Rs 12,500 per quintal – The Times of India
Vietnam’s textile sector anchors most of its supply chain employment
Arsenal transfers: Mikel Arteta rules out Gabriel Jesus exit
-
Sports7 days agoAustralia take control of second Ashes Test | The Express Tribune
-
Politics6 days ago17 found dead in migrant vessel off Crete: coastguard
-
Entertainment1 week agoSabrina Carpenter recalls ‘unbelievable’ experience with pal Taylor Swift
-
Fashion1 week agoBangladesh’s economic outlook cautiously optimistic: Govt
-
Fashion4 days agoGermany’s LuxExperience appoints Francis Belin as new CEO of Mytheresa
-
Tech1 week agoThe Trump Administration Wants Immigrants to Self-Deport. It’s a Shit Show
-
Tech7 days agoWIRED Roundup: DOGE Isn’t Dead, Facebook Dating Is Real, and Amazon’s AI Ambitions
-
Politics1 week agoGen-Z battles to gain political ground after ousting ex-PM Hasina in Bangladesh