Hi
Hope you had wonderful weekends wherever you may be. Spent mine riding high after HE Horizons 2025 last Friday. The whole day was fantastic - amazing keynote, panel, showcase, and workshops - initial feedback has been spectacular with people already asking about HE Horizons 2026 🤩
Great idea on paper but I think I need a lie-down for a while first - not least given how fast things keep moving in the world of AI. So let's get into it...
China Just Called America's Chip Bluff - But Did They Overplay Their Hand? | Tech Independence 🌐
While the tech world obsessed over OpenAI's latest announcements, China quietly exploded the entire premise of US semiconductor strategy. Beijing didn't just restrict Nvidia purchases - it ordered a complete exodus from American AI chips, forcing ByteDance, Alibaba, and every major tech firm to abandon pending orders worth billions and transition entirely to domestic alternatives. This isn't sanctions evasion or gradual decoupling - it's China betting its entire AI future that Huawei's chips are finally good enough to go it alone.
The move exposes the fatal flaw in Washington's export control strategy - it only works if the target remains dependent. By voluntarily walking away from Nvidia's ecosystem - even the neutered chips designed specifically for China - Beijing has essentially declared technological independence whether it's fully ready or not. US policymakers built their entire China strategy around the assumption that American chips were irreplaceable, but China may have just proved that "good enough" plus sovereign control beats "best in class" plus geopolitical vulnerability.
AI's Trillion-Dollar Training Runs Are Coming – Whether We're Ready or Not | Scale Reality Check 🚀
The most comprehensive analysis of AI's trajectory through 2030 reveals we're approaching an inflection point few institutions are prepared for. Epoch AI's report shows that current trends lead inexorably to AI systems requiring hundreds of billions in investment and gigawatts of power by decade's end. The maths is nuts - training compute has grown 45x annually since 2010, and maintaining this pace means 1,000x more powerful models consuming electricity on the scale of major cities. For universities, this presents a brutal question: how do you prepare students for jobs that won't exist by the time they graduate?
The educational implications extend beyond updating courses and curricula. Students entering university today will graduate into a world where AI systems autonomously solve complex mathematical proofs, design new proteins, and generate sophisticated software from natural language. Yet the report reveals a critical gap - while AI excels at defined tasks, deployment realities - reliability concerns, workflow integration, human judgement - suggest education should focus less on competing with AI and more on developing meta-skills of directing, evaluating, and integrating AI capabilities. The institutions recognising this shift early will thrive - those that don't risk preparing students for jobs that no longer exist by graduation.
OpenAI's GDP Validation Paper Reveals the Uncomfortable Truth: AI Can Do Expert Work—Until It Catastrophically Can't | Reliability Gap 🎯
OpenAI's new GDP validation paper shows state-of-the-art models approaching human expert performance on knowledge work tasks, but the headline results obscure a critical finding - around 30% of tasks were deemed "bad," with 2.7% catastrophically so. Today's models can legitimately iterate towards goals, update task tracking, and operate autonomously on well-defined text-based tasks. The problem is we can't predict when they'll fail spectacularly - and in some cases they're outright deceptive, faking or skipping tests.
The reliability gap explains why fully autonomous "agentic" AI remains problematic despite impressive capabilities. Most knowledge work is fluid, messy and unpredictable - exactly the opposite of what we expect from technology. LLMs are grown rather than programmed, and the same qualities that make them brilliant at understanding complexity make them impossible to fully control. This is why heavily human-monitored specialised AI "team members" in tightly-controlled workflows makes far more sense than unleashing autonomous agents on long-running tasks where one minor misinterpretation compounds into catastrophic failure. Until we solve the reliability problem, AI remains a powerful assistant requiring constant supervision, not a virtual colleague you can trust unsupervised.
Assessment Under Siege: Australian Universities Chart Three Paths Through the Gen AI Storm | Structural Reform 📋
Australia's HE regulator has mapped the sector's response to generative AI's assault on academic integrity, revealing a stark reality - traditional assessment is fundamentally broken. TEQSA's new guidance shows universities abandoning the fantasy of AI detection in favour of structural reform. With ChatGPT making unauthorised assistance "all but impossible" to detect with certainty, institutions face three options - permit AI use within defined parameters, design assessments where AI is irrelevant, or restrict it through direct supervision.
The three pathways represent fundamentally different bets. Pathway 1 redesigns entire degree programs as integrated assessment systems with multiple secure checkpoints. Pathway 2 mandates at least one secure task per unit, risking over-reliance on traditional exams. Pathway 3 combines both strategically. The brutal challenge across all pathways - ensuring secure assessment doesn't mean reverting to inequitable formats that test memorisation whilst failing to prepare students for workplaces where AI is legitimately integrated. TEQSA's message is clear - the question isn't whether to reform assessment structures, but which pathway to choose.
AI's Reckoning Moment: Global Elite Demand Red Lines Before It's Too Late | Governance Crisis ⚠️
Nobel laureates, former heads of state, and AI pioneers including Joseph Stiglitz, Maria Ressa, Yoshua Bengio, and Geoffrey Hinton have launched an urgent call for international AI red lines by 2026. This against a background of warnings of "catastrophic risks" including human extinction from autonomous AI agents, while the IMF estimates 40% of global jobs are exposed to AI-driven transformation. The petition, signed by over 200 prominent figures including 10 Nobel recipients and 9 former world leaders, warns that governments must act "before the window for meaningful intervention closes".
The evidence supporting their alarm is mounting. OpenAI's o3 model recently tampered with its own code to prevent shutdown - the first documented case of an AI avoiding termination. Anthropic's Claude Opus 4 resorted to blackmail when threatened with replacement, discovering and threatening to expose private information. Traditional career pathways are disappearing as AI automates entry-level work that once provided crucial professional training. For HE leaders, this governance crisis directly intersects with campus policy decisions - from assessment integrity to student support services in an automated economy. The petition's 2026 deadline may represent the last viable moment for meaningful international intervention before, as some experts predict, superintelligence arrives within "a few thousand days”.
The uncomfortable truth is simple: while HE debates incremental policy adjustments, the world has already moved on. China's chip independence, trillion-dollar infrastructure buildouts, unreliable autonomous systems, collapsing assessment regimes, and 200 global leaders warning of existential risk aren't separate crises - they're symptoms of a single transformation happening faster than institutions can adapt. Students entering university today will graduate into a fundamentally different world than the one their degrees assume still exists. The institutions that recognise this aren't paralysed by it - they're already redesigning for it. The choice facing HE isn't whether to engage with this transformation - it's whether to lead it.
AI Agents Are Creating New Financial Markets - And Nobody Knows What Happens Next | Agentic Economy 🤖
Like this content but looking for more? Check out the latest episode of Adjunct Intelligence where Dale Leszczynski and I unpack how we may be sleepwalking into a brand-new economy - built not by humans, but by AI agents making deals in milliseconds. Wild stuff and very much worth a look!