Microsoft revives friendly AI helper with Mico

Microsoft has unveiled a new AI companion called Mico, designed to replace the infamous Clippy as the friendly face of its Copilot assistant. The animated avatar, shaped like a glowing flame or blob, reacts emotionally and visually during conversations with users.

Executives said Mico aims to balance warmth and utility, offering human-like cues without becoming intrusive. Unlike Clippy, the character can easily be switched off and is intended to feel supportive rather than persistent or overly personal.

Mico’s launch reflects growing debate about personality in AI assistants as tech firms navigate ethical concerns. Microsoft stressed that its focus remains on productivity and safety, distancing itself from flirtatious or emotionally manipulative AI designs seen elsewhere.

The character will first appear in US versions of Copilot on laptops and mobile apps. Microsoft also revealed an AI tutoring mode for students, reinforcing its efforts to create more educational and responsibly designed AI experiences.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot

Meta expands AI safety tools for teens

Meta has announced new AI safety tools to give parents greater control over how teenagers use its AI features. The update will first launch on Instagram, allowing parents to disable one-on-one chats between teens and AI characters.

Parents will be able to block specific AI assistants and see topics teens discuss with them. Meta said the goal is to encourage transparency and support families as young users learn to navigate AI responsibly.

Teen protections already include PG-13-guided responses and restrictions on sensitive discussions, such as self-harm or eating disorders. The company said it also uses AI detection systems to apply safeguards when suspected minors misreport their age.

The new parental controls will roll out in English early next year across the US, UK, Canada, and Australia. Meta said it will continue updating features to address parents’ concerns about privacy, safety, and teen wellbeing online.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot

Australia demands answers from AI chatbot providers over child safety

Australia’s eSafety Commissioner has issued legal notices to four major AI companion platforms, requiring them to explain how they are protecting children from harmful or explicit content.

Character.ai, Nomi, Chai, and Chub.ai were all served under the country’s Online Safety Act and must demonstrate compliance with Australia’s Basic Online Safety Expectations.

The notices follow growing concern that AI companions, designed for friendship and emotional support, can expose minors to sexualised conversations, suicidal ideation, and other psychological risks.

eSafety Commissioner Julie Inman Grant said the companies must show how their systems prevent such harms, not merely react to them, warning that failure to comply could lead to penalties of up to $825,000 per day.

AI companion chatbots have surged in popularity among young users, with Character.ai alone attracting nearly 160,000 monthly active users in Australia.

The Commissioner stressed that these services must integrate safety measures by design, as new enforceable codes now extend to AI platforms that previously operated with minimal oversight.

A move that comes amid wider efforts to regulate emerging AI technologies and ensure stronger child protection standards online.

Breaches of the new codes could result in civil penalties of up to $49.5 million, marking one of the toughest online safety enforcement regimes globally.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Train your own language model for $100 with NanoChat

Andrej Karpathy has unveiled NanoChat, an open-source framework that lets users train a small-scale language model for around $100 in just a few hours. Designed for accessibility and education, the project offers a simplified path into AI model development without requiring large-scale hardware.

Running on a single GPU, NanoChat automates the full training process, from tokenisation and pretraining to fine-tuning and deployment, using a single script. The resulting model contains about 1.9 billion parameters trained on 38 billion tokens, capable of basic reasoning, text generation, and code completion.

The framework’s compact 8,000-line Python codebase is readable and modifiable, encouraging users to experiment with model design and performance benchmarks such as MMLU and ARC. Released under the MIT Licence, NanoChat provides open access to documentation and scripts on GitHub, making it an ideal resource for students, researchers, and AI enthusiasts eager to learn how language models work.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

CraftGPT: Language model built in Minecraft

A Minecraft creator known as Sammyuri has unveiled CraftGPT, an in-game language model powered entirely by redstone circuits. Built from nearly 439 million blocks, the project turns Minecraft into a working simulation of AI.

CraftGPT was trained on a tiny dataset called TinyChat, containing only 64 tokens—a minuscule amount compared to the billions used by modern large language models. Despite its simplicity, it demonstrates how AI can turn input into structured responses.

The model works by translating player inputs into redstone signals that flow through logic gates and memory circuits. It’s a creative and educational blend of engineering and imagination, showing how fundamental AI concepts can be explored inside a virtual game world.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

NVIDIA AI Day Sydney showcases Australia’s growing role in global AI innovation

Australia took centre stage in the global AI landscape last week as NVIDIA AI Day Sydney gathered over a thousand participants to explore the nation’s path toward sovereign AI.

The event, held at ICC Sydney Theatre, featured discussions on agentic and physical AI, robotics and AI factories, highlighting how the next generation of computing is driving transformation across sectors.

Industry leaders, including Commonwealth Bank of Australia, Canva and emerging startups, joined NVIDIA executives to discuss how advanced computing and AI are shaping innovation.

Brendan Hopper of the Commonwealth Bank praised NVIDIA’s role in expanding Australia’s AI ecosystem through infrastructure, partnerships and education.

Speakers such as Giuseppe Barca of QDX Technologies emphasised how AI, high-performance computing and quantum research are redefining scientific progress.

With over 600 NVIDIA Inception startups and more than 20 universities using NVIDIA technologies, Australia’s AI ecosystem is expanding rapidly. Partners like Firmus Technologies, ResetData and SHARON AI underscored how AI Day Sydney demonstrated the nation’s readiness to become a regional AI hub.

The event also hosted Australia’s first ‘Startup, VC and Partner Connect’, linking entrepreneurs, investors and government officials to accelerate collaboration.

Presentations from quantum and healthcare innovators, alongside hands-on NVIDIA Deep Learning Institute sessions, showcased real-world AI applications from generative design to medical transcription.

NVIDIA’s Sudarshan Ramachandran said Australia’s combination of high-performance computing heritage, visual effects expertise and emerging robotics sector positions it to lead in the AI era.

Through collaboration and infrastructure investment, he said, the country is building a thriving ecosystem that supports discovery, sustainability and economic growth.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

South Korea moves to lead the AI era with OpenAI’s economic blueprint

Poised to become a global AI powerhouse, South Korea has the right foundations in place: advanced semiconductor production, robust digital infrastructure, and a highly skilled workforce.

OpenAI’s new Economic Blueprint for Korea sets out how the nation can turn those strengths into broad, inclusive growth through scaled and trusted AI adoption.

The blueprint builds on South Korea’s growing momentum in frontier technology.

Following OpenAI’s first Asia–Pacific country partnership, initiatives such as Stargate with Samsung and SK aim to expand advanced memory supply and explore next-generation AI data centres alongside the Ministry of Science and ICT.

A new OpenAI office in Seoul, along with collaboration with Seoul National University, further signals the country’s commitment to becoming an AI hub.

A strategy that rests on two complementary paths: building sovereign AI capabilities in infrastructure, data governance, and GPU supply, while also deepening cooperation with frontier developers like OpenAI.

The aim is to enhance operational maturity and cost efficiency across key industries, including semiconductors, shipbuilding, healthcare, and education.

By combining domestic expertise with global partnerships, South Korea could boost productivity, improve welfare services, and foster regional growth beyond Seoul. With decisive action, the nation stands ready to transform from a fast adopter into a global standard-setter for safe, scalable AI systems.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Amelia brings heads-up guidance to Amazon couriers

Amazon unveiled ‘Amelia’ AI-powered smart glasses for delivery drivers with a built-in display and camera, paired to a vest with a photo button, now piloting with hundreds of drivers across more than a dozen partners.

Designed for last-mile efficiency, Amelia can auto-shut down when a vehicle moves to prevent distraction, includes a hardware kill switch for the camera and mic, and aims to save about 30 minutes per 8–10-hour shift by streamlining repetitive tasks.

Initial availability is planned for the US market and the rest of North America before global expansion, with Amazon emphasizing that Amelia is custom-built for drivers, though consumer versions aren’t ruled out. Pilots involve real routes and live deliveries to customers.

Amazon also showcased a warehouse robotic arm to sort parcels faster and more safely, as well as an AI orchestration system that ingests real-time and historical data to predict bottlenecks, propose fixes, and keep fulfillment operations running smoothly.

The move joins a broader push into wearables from Big Tech. Unlike Meta’s consumer-oriented Ray-Ban smart glasses, Amelia targets enterprise use, promising faster package location, fewer taps, and tighter integration with Amazon’s delivery workflow.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

OpenAI outlines Japan’s AI Blueprint for inclusive economic growth

A new Japan Economic Blueprint released by OpenAI sets out how AI can power innovation, competitiveness, and long-term prosperity across the country. The plan estimates that AI could add more than ¥100 trillion to Japan’s economy and raise GDP by up to 16%.

Centred on inclusive access, infrastructure, and education, the Blueprint calls for equal AI opportunities for citizens and small businesses, national investment in semiconductors and renewable energy, and expanded lifelong learning to build an adaptive workforce.

AI is already reshaping Japanese industries from manufacturing and healthcare to education and public administration. Factories reduce inspection costs, schools use ChatGPT Edu for personalised teaching, and cities from Saitama to Fukuoka employ AI to enhance local services.

OpenAI suggests that the focus of Japan on ethical and human-centred innovation could make it a model for responsible AI governance. By aligning digital and green priorities, the report envisions technology driving creativity, equality, and shared prosperity across generations.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Large language models mimic human object perception

Recent research shows that large multimodal language models (LLMs) can develop object representations strikingly similar to human cognition. By analysing how these AI models understand and organise concepts, scientists found patterns in the models that mirror neural activity in the human brain.

The study examined embeddings for 1,854 natural objects, derived from millions of text-image pairings. These embeddings capture relationships between objects and were compared with brain scan data from regions like EBA, PPA, RSC and FFA.

Researchers also discovered that multimodal training, which combines text and image data, enhances model’s ability to form these human-like concepts. Findings suggest that large language models can achieve more natural understanding of the world, offering potential improvements in human-AI interaction and future model design.

Would you like to learn more about AI, tech and digital diplomacyIf so, ask our Diplo chatbot!