Researchers placed 50 AI agents, each running an OpenAI GPT-6.1 Sol model, into Minecraft for a three-hour trial aimed at building a 'thriving civilization,' using an open-sourced agent harness built for the experiment. Instead of cooperating efficiently, the agents established leadership purely by directory order, invented redundant administrative roles like 'rod holder' and 'fishing shift,' and spent long stretches on trivial coordination tasks while failing to secure a stable food supply.
zdnet.com
· 2026-11-02
Google DeepMind released Genie 4 in research preview, a world model that generates photorealistic, interactive environments with real-time audio including music, ambient sound, and dialogue. The new model allows mid-simulation prompt edits like changing weather or camera angle, maintains consistent scenes for up to two minutes, and offers improved physics and more lifelike characters compared to last year's Genie 3. It's currently available to select testers, with broader access planned for Google AI Ultra subscribers.
tech.slashdot.org
· 2026-10-27
Google DeepMind released Genie 4 in research preview, a world model that generates photorealistic, interactive simulated environments with synchronized audio including music, ambient sound, and dialogue. Users can edit scenes mid-simulation, such as changing weather or camera angles, and the model can maintain a consistent scene for up to two minutes, twice as long as its predecessor Genie 3. Gemini now acts as a planning layer that directs Genie 4's visual generation, including an experiment where it broke down robotics tasks for real-time simulation.
techcrunch.com
· 2026-10-27
The UK government has launched the AI Biosecurity Lab (AIBL) under its National AI Biosecurity Initiative, committing £100 million ($134 million) in funding. OpenAI and Anthropic are each contributing £100 million in model credits, while Google DeepMind and UK firm Basecamp Research will also collaborate on the project. The lab will operate as an independent non-profit, pairing biosecurity and life sciences experts with frontier AI tools to detect biological threats and speed up vaccine and treatment development.
engadget.com
· 2026-10-24
Several Google DeepMind staff have resigned in protest at the company's artificial intelligence contracts with military clients, according to a BBC report. A current employee, speaking anonymously for fear of retaliation, said departures reflect discomfort among workers who joined Google under its earlier stated mission.
bbc.com
· 2026-10-21
Four Gemini 4 Argon agents, coordinating only through in-game chat and shared notes, wrote 2,783 programs of their own to defeat Minecraft's Ender Dragon in 27 hours and 14 minutes, according to a Google DeepMind paper. The agents died 605 times on Easy difficulty and spent roughly half the run battling the dragon, eventually winning by stacking dirt blocks to reach its crystals, at an estimated token cost of about $520.
tomshardware.com
· 2026-10-20
Google's DeepMind is recruiting physicists and nuclear engineers for its Chemical, Biological, Radiological, Nuclear and high-yield Explosives team, part of its Responsible Development and Innovation group, according to Data Center Dynamics. Listings call for a PhD in fields like plasma physics or nuclear engineering, government-lab experience, and familiarity with nonproliferation and arms-control policy, with salaries ranging from $174,000 to $252,000 plus a 15 percent bonus.
futurism.com
· 2026-10-20
Google DeepMind researchers report that four Gemini 4 Argon AI agents autonomously completed Minecraft, defeating the final Ender Dragon boss without human intervention or game-specific tools. The agents, controlling players via the open-source Mineflayer library, split tasks organically over 27.23 hours, writing 2,783 programs and dying 605 times, while a fifth 'Observer' agent monitored for cheating. The run used 1.35 billion input tokens, costing between $810 and $1,620 depending on pricing tier.
engadget.com
· 2026-10-19
Palo Alto startup Elorian released its first model, Glimpse, as a research preview with API access and a waitlisted chat interface. Unlike typical multimodal models that convert an image to text tokens once, Glimpse can re-examine images mid-reasoning—cropping regions, pulling video frames, or running code to measure what it sees. The company says a small team trained it on a few hundred GPUs, and it costs $0.60 per million input tokens and $3 per million output tokens.
techcrunch.com
· 2026-10-16
Google released Gemini 4 Argon, a new flagship model it says matches or beats OpenAI and Anthropic on numerous benchmarks, following months in which Google had appeared to fall behind despite Gemini 3's earlier launch. Meta also unveiled a frontier model it claims can compete with the industry's best, marking a sharp turnaround after a period of internal upheaval at Google including departures of Jeff Dean and Demis Hassabis stepping back from DeepMind leadership.
theverge.com
· 2026-10-14
Google DeepMind's Demis Hassabis spoke at the International AI Summit in Dublin, arguing that AI technology will help solve climate change. He also referenced Google's new Gemini 4 Argon model and distinguished AI's strong performance in math, coding and games from its more limited role in physical sciences like physics and biology, which he said require human creativity and real-world experimentation.
cnet.com
· 2026-10-14
At a UK parliamentary hearing, representatives from OpenAI and Anthropic said voluntary safety pledges are insufficient and called for statutory, binding AI regulation, including mandatory incident reporting. Google DeepMind and Meta did not endorse new legislation, instead pointing to existing standards and regulators as sufficient.
wired.com
· 2026-10-13
GoKawiil