PLUS: NVIDIA slashes AI token costs by 90%, OpenAI launches medical data vault & PayPal enables chatbot purchases

Morning All, A leading UK AI researcher warns humanity may only have five years left to effectively protect ourselves as autonomous systems rapidly improve. AI capabilities are even outpacing Moore's Law by nearly threefold, meaning the day artificial intelligence surpasses human intelligence in critical economic roles will be here quicker than we thought.

With AI projected to automate full research cycles by late 2026, are current safety protocols evolving quickly enough to maintain meaningful human oversight? It's an urgent countdown which demands unprecedented collaboration between policymakers and technologists. Are they ready to answer the call? because humanity depends on it.

Today's dots:

• Emergency timeline for AI safety governance
• NVIDIA cuts AI costs 90% with Rubin platform
• OpenAI's Health Vault speeds medical prep
• PayPal enables chatbot purchases in ChatGPT
• Twin dangers in AI's trust crisis

AI Safety Countdown: 5-Year Window to Control Advanced Systems

Here's the thing: A top UK AI expert warns we may have less time to implement safety measures than previously thought as autonomous systems accelerate towards outperforming humans across critical economic tasks by 2028. Recent Guardian interview reveals startling timelines.

Let's unpack that:

• David Dalrymple from UK government-backed ARIA predicts most economically valuable human tasks will be machine-dominated within 5 years, requiring urgent governance frameworks
• AI capabilities are now doubling every 8 months according to UK Security Institute data - nearly 3x faster than Moore's Law
• New systems autonomously complete 60% of expert-level tasks that previously required human oversight
• By late 2026, AI could automate full R&D cycles - creating a feedback loop accelerating its own development
• Current safety research focuses on safeguarding energy grids and healthcare infrastructure as first-line defences

If you remember nothing else: The window for meaningful human oversight is closing faster than safety protocols can develop. This demands accelerated technical safeguards alongside policy frameworks that prioritise risk mitigation today.

NVIDIA Slashes AI Costs 90% With Rubin Platform While Advancing "Thinking" Autonomous Vehicles

Here's the thing: NVIDIA just unveiled game-changing AI infrastructure that could make large-scale deployments 10x cheaper, alongside autonomous vehicle tech that reasons like humans. Rubin platform announcement

Let's unpack that:

• Their new Rubin AI platform slashes token generation costs by 90% – making enterprise AI deployments dramatically more affordable
• The architecture combines liquid-cooled GPUs with custom Vera CPUs designed specifically for developing agentic AI systems at scale
• Mercedes-Benz is already deploying this in production vehicles, with their new CLA range using Alpamayo-powered reasoning VLA models for US roads this year [Details]
• Siemens showcased how they're integrating Rubin to turn entire factories into "giant robots" through industrial AI operating systems

If you remember nothing else: NVIDIA just reshaped the economics of large-scale AI while solving autonomous driving's toughest challenge – unpredictable scenarios. This combo could accelerate physical AI deployments from factories to streets faster than most predictions.

OpenAI's health vault could speed up your doctor visits

Here's the thing: OpenAI just launched ChatGPT Health, a medical data vault that securely connects your health records and fitness apps to help you prepare 28% faster for medical visits based on early trials.

Let's unpack that:

• The system pulls data from Apple Health, MyFitnessPal and other sources into one HIPAA-compliant space, keeping your medical chats walled off from regular ChatGPT conversations
• Over 260 doctors from 60 countries helped build safety guardrails, testing 600,000+ AI responses using medical validation tools like the HealthBench framework
• It’s explicitly designed to support - not replace - doctors, focusing on helping you understand your test results and options rather than making diagnoses
• Synthetic data generation helps the AI understand rare conditions without exposing real patient information
• It's currently US-only for medical records integration, with UK/European users excluded from early access due to stricter privacy regulations

If you remember nothing else: This could save millions of people hours spent digging through scattered health portals. The real win? AI that helps demystify medical jargon before you even reach the clinic.

PayPal chatbots now take payments in ChatGPT

Here's the thing: PayPal just unleashed AI-powered purchases within ChatGPT, letting 400M users buy products mid-conversation – powered by their new open-source Agentic Commerce Protocol.

Let's unpack that:

• Your weekly chat with ChatGPT just became a checkout counter: Ask for product recommendations, then pay instantly without leaving the conversation
• The protocol gives Shopify's 1M+ merchants a frictionless sales channel – imagine customers buying beauty products while discussing skincare routines
• Early tests show 1.2-second transaction speeds (faster than most contactless payments).
• PayPal's engineering teams are already using OpenAI's Codex internally – hinting at future workforce AI upskilling opportunities
• Stripe handles the payment plumbing, ensuring compatibility with existing systems through their new AI economic infrastructure

If you remember nothing else: This transforms AI assistants from search tools into commercial platforms. GEO, AEO, SEO whatever you want to call it, it just became top priority for any quality brand in tomorrow's economy. The winners? Brands ready to meet customers wherever they chat.

AI's Trust Crisis Hits Breaking Point

Here's the thing: This week exposed AI's twin dangers - malicious propaganda like a fake viral Venezuela video (5M+ views) and systemic failures as Google’s AI Overviews gave life-threatening health advice. The incidents show why hallucinations aren't just quirks - they're becoming societal hazards.

Let's unpack that:

• Propaganda factories weaponised AI to create fake celebratory footage of Venezuelans thanking Trump - complete with distorted faces and robotic voices that still convinced millions
• Google’s health summaries told pancreatic cancer patients to avoid high-fat foods - the exact opposite of medical guidelines - while hallucinating psychosis treatments
• 79% of people now use AI for health queries per UPenn research, creating a perfect storm when models confidently spout incorrect facts
• 57 national AI safety regulations are now in development globally, suggesting policymakers see this as a critical inflection point

If you remember nothing else: These cases reveal how quickly AI risks escalate from theoretical to life-altering. While innovation continues, verifying AI outputs through human expertise remains non-negotiable - especially for high-stakes decisions.

The Shortlist

Higher education faces transformative changes, with AI integration expanding to curriculum and faculty development - Rand research suggests institutions must balance efficiency with human-centered skills as economic pressures mount.

MIT scientists discovered AI systems disproportionately impact marginalised communities through cultural assumptions in training data, creating systemic vulnerabilities in critical infrastructure like credit scoring and fraud detection.

Slingwave launched an AI-native measurement platform combining attribution and experimentation to optimise eCommerce growth, reporting 20-50% campaign lift for DTC brands through real-time scenario modeling.