Google's AI Agents Beat Minecraft's Ender Dragon in 27 Hours

Key Takeaways

• <ul><li>Four Gemini 4 Argon AI agents beat Minecraft's Ender Dragon in 27 hours and 14 minutes without human help</li><li>The agents wrote 2,783 JavaScript programs and exchanged 229 messages through in-game chat to coordinate</li><li>AI course enrolments in South Africa rose 75% year-on-year, with critical thinking courses growing 132%</li><li>Human judgment is becoming more valuable in investing as AI cannot share the judgment clients need</li><li>A new position paper warns AI risks can build up slowly like social media harms through daily use</li><li>Figurobot launched FR-NIA, an 11-inch bionic AI companion robot starting at $599</li><li>Raagverse debuted as a human-led music label using AI, with content available on YouTube Music, Apple Music and Spotify</li><li>App State held an AI ethics panel discussing misinformation, inequitable benefits, and environmental harm from data centers</li><li>Researchers at Mass General Brigham created FAHR-Face, an AI model trained on over 40 million face images</li><li>Broadcom says version currency is vital for defending against frontier AI exploits</li></ul>

Four Gemini 4 AI agents beat Minecraft in 27 hours

Google DeepMind used four Gemini 4 Argon AI agents to beat Minecraft's Ender Dragon in 27 hours and 14 minutes. The agents wrote 2,783 JavaScript programs on their own and died 605 times on Easy mode. They used the open-source Mineflayer library and talked in game to split tasks. The run showed AI agents can handle long, complex teamwork tasks.

Google AI agents defeat Minecraft Ender Dragon alone

Four Google Gemini 4 Argon AI agents defeated the Ender Dragon in Minecraft without any human help. They used in-game chat to coordinate, exchanged 229 messages, and ran 2,783 programs over 27 hours. The agents died 605 times and used 1.35 billion input tokens. Researchers say the goal was to study how AI teams work together on hard tasks.

Four AI agents slay Minecraft's Ender Dragon

Four AI agents completed Minecraft from start to finish on their own, likely the first time this has been done. They wrote 2,783 JavaScript programs using only the standard Mineflayer library. The agents were general-purpose coding agents, not built just for the game. They showed they can split work, share results, and learn from mistakes like a human team.

Human judgment matters more as AI spreads in investing

AI is making investment information easier to find and summarize, but human judgment is becoming more valuable. Maryann Selfe of Calabar Capital says AI can share information but cannot share the judgment clients need. She says all AI-made material must be checked by a person before it reaches investors. The growing question is how firms will use time saved by AI to give better advice.

AI course enrolments surge in South Africa

AI course enrolments in South Africa rose 75% year-on-year, while critical thinking courses grew 132%. The Coursera Global Skills Report 2026 also found big jumps in AI security, agentic workflows, and generative AI learning. Micro-credential enrolments in AI topics rose 74%. Coursera has about 1.6 million users in South Africa studying skills tied to AI and work.

AI risks can build up slowly like social media harms

A new position paper says AI can cause harm that builds up over time through daily use. It compares this to how social media narrows perspectives and weakens judgment. The paper calls for studying repeated use patterns and adding safeguards. It warns that focusing only on dramatic disasters may ignore these quieter but growing risks.

Human-AI music label Raagverse debuts with Janaa Hai Toh Jao

Amita Madhvani launched Raagverse, a music label that uses AI while keeping humans in charge. The debut album includes three songs with lyrics by Mamta Javkar. The lead track Janaa Hai Toh Jao will be available on YouTube Music, Apple Music and Spotify. The label aims to create original Indian content for a global audience.

AI tool estimates biological age and predicts cancer survival from photos

Researchers at Mass General Brigham created FAHR-Face, an AI model trained on over 40 million face images. The tool can estimate a person's biological age and predict cancer survival from a simple photo. This offers a non-invasive and low-cost way to assess health risks. The researchers believe it could help improve healthcare outcomes.

Why keeping software versions current matters against frontier AI attacks

A Broadcom article explains that version currency is vital for defending against frontier AI exploits. Version currency means tracking and updating software versions to keep systems secure. The article says attackers can act in seconds, making up-to-date systems essential. It calls for combining people, process and technology to maintain strong cybersecurity.

New Bedford says AI needs safety tests before speed tests

A New Bedford Light article urges employers to test AI for safety before focusing on speed. It suggests starting with one narrow process and tracking error rates, completion times and corrections. Workers should be paid time to test improvements and be free to reject inaccurate results. The goal is to keep training and mentoring while using AI to make safer decisions.

Figurobot launches FR-NIA, a 11-inch bionic AI companion robot

Figurobot launched FR-NIA, an 11-inch desktop bionic AI companion robot that uses multimodal AI and facial expressions to interact naturally. The robot features a biomimetic silicone face with 10 expressive degrees of freedom and precision motion control. It is available in Lite, Pro, and Pro Max models starting at $599. Users can customize its appearance, personality, and memory for personalized companionship. The robot builds on 18 years of miniature torque joint development by founder Jim Lu.

App State panel explores AI ethics and sustainability concerns

App State held an AI ethics and sustainability panel on Oct. 14 featuring professors Michael Vaughn, Christopher Bartel, Richard Rheingans, and Jacqui Ignatova. The experts discussed AI risks including misinformation, inequitable benefits, and environmental harm from data centers. Vaughn compared AI regulation to driver training, while Bartel argued AI carries more inherent ethical problems than other tools. Ignatova expressed concern that AI use undermines student critical thinking and writing skills. After the panel, professors met with students to discuss the value of human friction in daily life.

Sources

• Google DeepMind says four AI agents beat Minecraft in 27 hours by writing 2,783 programs of their own — AI team died 605 times on Easy difficulty and ran up an estimated $520 in tokens
• It Took 4 Google AI Agents Over 27 Hours to Defeat the Ender Dragon in Minecraft Without Help From Humans
• Four AI Agents Have Slain Minecraft's Ender Dragon
• Human Judgment Gains Value As AI Expands Access To Investment Information
• Spike in AI course enrolments in SA
• What Social Media Teaches Us About Accumulative AI Risks
• Human-AI Music Label Raagverse Launches With ‘Janaa Hai Toh Jao’
• Advanced AI tool estimates biological age and predicts cancer survival from face photos
• Version Currency in the Age of Frontier AI
• New Bedford’s AI future needs a safety test before a speed test
• Figurobot Launches FR-NIA: A 11-inch Bionic AI Companion Robot Bringing Intelligent Interaction into Everyday Life
• Artificial Intelligence ethics and sustainability panel gathers experts on emerging tech