Welcome to Issue #04 of The AI Trailblaizers. This week it is difficult to know where to start. AGI has allegedly arrived — depending on who you ask and how they define it. Anthropic’s next model may have already leaked. And there is a robot sweating in a lab somewhere, wondering why it has sweat glands. Three concepts every Trailblaizer needs to understand this week.
THIS WEEK’S TOP 3 MUST-KNOW CONCEPTS
Jensen Huang Says AGI Is Here — But Read the Fine Print
AI RACE
NVIDIA CEO Jensen Huang publicly declared that AGI has been achieved — that AI can now conceive, build, and operate billion-dollar businesses without human involvement. This is significant coming from the man who supplies the hardware that runs the entire AI industry. But his definition of AGI is narrower than the one most researchers use. Traditional AGI means general human-level reasoning across any domain; Huang means something closer to ‘AI can run a business autonomously.’ Many leading AI scientists disagree. For SME leaders, the practical question isn’t whether the label is correct — it’s that the world’s most influential chip CEO has publicly stated the threshold has been crossed. Expect that framing to accelerate investment, hiring decisions, and competitive pressure across every sector.
Anthropic’s Mythos LLM Leaks — And People Are Worried
AI MODELS
Details about a new Anthropic model called Mythos have leaked, and the early descriptions are raising eyebrows. It is being described as a significant capability jump — and unlike the usual pre-launch excitement, the chatter this time includes serious researchers flagging concerns about potential misuse. This is worth paying attention to, not because of the hype, but because Anthropic has a track record of being more cautious than its competitors before shipping. When even Anthropic’s own community is nervous about something, that is data. Separately, OpenAI is reportedly preparing to ship GPT-5.5 in April, suggesting both labs are in an active sprint. The next 60 days may look very different from the last 60.
Claude Code Leaked 512,000 Lines of Its Own Source Code
PRODUCT
Anthropic’s Claude Code product accidentally published its entire codebase — all 512,000 lines — publicly before it was taken down. What was discovered inside is genuinely interesting: persistent memory being built directly into Claude Code; a proactive background agent called Kairos that works autonomously to improve your projects without being asked; a terminal companion called BUDDY that behaves like a Tamagotchi and reacts to what you are building; and a hidden counter that tracks every time a user swears at Claude to monitor frustration levels. Also visible: references to a new model called Capybara, described internally as a major capability leap forward. Whether intentional or not, this leak told the world more about Anthropic’s product roadmap than any press release.
OTHER MAJOR STORIES THIS WEEK
Google Launches Gemini 3.1 Flash Live
TOOLS
Google has unveiled Gemini 3.1 Flash Live, calling it its best audio model to date. The model handles conversations twice as long as its predecessor, applies real-time background noise filtering, and is now powering Search Live across 200 countries. For SME leaders who have been watching voice interfaces develop from the sidelines, this is a meaningful step — conversational AI that actually works in a noisy office or warehouse is a genuinely different product to one that only functions in ideal conditions.
Xiaomi Gives Its Robots Sweat Glands
ROBOTICS
Xiaomi is integrating sweat gland-like structures into its robot hands, enabling perspiration-style cooling. Rather than heat sinks or fans, the mechanism mimics the way human skin regulates temperature under load. It is an unusual convergence of biology and engineering — and a signal that advanced humanoid robotics is increasingly looking at the human body as the engineering blueprint, not just the inspiration.
Claude Can Now Use Your Computer
PRODUCT
Anthropic has launched full computer use capability for Claude — meaning Claude can now open applications, navigate files, read your screen, and execute tasks on your actual machine, not just inside a chat window. For SME leaders, this crosses a meaningful threshold: AI that previously gave you instructions you then had to act on can now simply do the work. The practical question is no longer whether this technology exists. It is which tasks you are comfortable delegating, and which require you in the loop.
QUICK DIGEST
Tesla Optimus 3: Nearly Human
ROBOTICS
Tesla says Optimus 3 will closely match human body proportions and movement. Unlike earlier iterations that were clearly robotic in scale and gait, Optimus 3 is designed to fit standard human workspaces and use standard human tools without modifying the environment. As the hardware becomes more human, the bar for what counts as useful in the workplace drops significantly. Worth watching for any business that relies on physical labour.
DoorDash Pays People to Film Their Chores
AI TRAINING
DoorDash launched an app paying people to film themselves doing household tasks — not for content, but for robot training data. The people being filmed are, in effect, teaching the systems that may eventually replace them. The company appears comfortable with the irony. It is a useful reminder that AI training requires enormous volumes of real-world human behaviour, and that people will voluntarily supply it for surprisingly modest compensation.
Meta and Google Found Liable for Addictive Design
AI ETHICS
A US jury found Meta and Google liable for harming a young woman’s mental health through intentionally addictive platform design, ordering $6 million in damages. Small figure, large precedent. This is the first time a US court has held social media companies accountable for how their systems were engineered, not just what they distributed. For any business building digital products — or evaluating AI-powered engagement features — design choices that prioritise retention over user wellbeing are now carrying measurable legal risk. The direction of legal travel is clear.