Bit2Watt: 1,000 GPUs can destabilise data centre power grids
Bit2Watt preprint shows how manipulating 1,000 GPUs in a renewable-heavy data centre can push power distortion to 46.8%, threatening both compute and grid stabi
A newsletter breaking down AI research, technology, and Australian property in plain English
AI and technology, explained through what they actually mean. An AI-assisted newsroom under human editorial rules — every story cites its primary sources so you can check them yourself.
127 stories
Bit2Watt preprint shows how manipulating 1,000 GPUs in a renewable-heavy data centre can push power distortion to 46.8%, threatening both compute and grid stabi
Researchers show a single email can silently poison an AI agent's long-term memory, succeeding 87.5% of the time on benchmark tests — affecting all persistent-a
An arXiv preprint defines world models as internal simulators and proposes a staged roadmap — but AI researchers across subfields still disagree on what the ter
A new arXiv preprint quantifies why trained networks outperform their kernel limit on compositional functions by up to six orders of magnitude in test error.
Cortex bridges AI planning and robot execution with 32 skill primitives, enabling zero-shot multi-stage chemistry experiments without task-specific training.
Tsinghua's reinforcement learning method teaches long-horizon AI agents to compress their own context, lifting open models on coding benchmarks and entering GLM
The median Sydney dwelling has slipped to $1.222m as rate hikes and budget uncertainty accelerate falls. Here's who's most exposed.
National home values slipped 0.4% in June as Sydney led a quarterly decline of 3.2%. HSBC says the correction is 'just the beginning'.
A new superchip brings data centre power to laptops, challenging Apple and Intel.
A new method uses hidden activations to block bad prompts instantly, requiring no model retraining.
New arXiv research reveals a dangerous blind spot in cybersecurity AI: models can resist attacks on paper while their decision logic falls apart.
New ICLR 2026 research finds masked image modelling is inherently more robust than contrastive learning when decentralised unlabelled data differs across nodes.