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
Distributed AI training study finds Masked Image Modeling more robust than contrastive learning on heterogeneous data, reshaping how teams train models across m
A new benchmark called IG-Bench tests whether LLMs can trace how scientific ideas inherit and mutate from prior work — the best system managed just 27.3% exact
A new arXiv preprint introduces reasoning consistency scanning — a method that flags when a model's stated reasoning doesn't match its answer, using transcripts
A new arXiv benchmark from University of Michigan evaluates whether data-science AI agents can distinguish causation from correlation — and know when to abstain
A July 2026 preprint argues echo chambers and misinformation miss the real threat: strategic manipulation in mixed human-LLM communicative networks.
QANTIS shows IBM Heron quantum hardware updates beliefs across 32 planning steps without changing the action a classical planner would take — on one benchmark o
Agentic AI beats single-LLM systems in small commercial underwriting, especially when data is missing or rules need multiple steps, finds a new arXiv paper.
A new arXiv survey maps clinical needs to AI reasoning capabilities across 18 models, revealing specialist LLMs win diagnosis while general models lead decision
Infinity-Parser2 Pro scores 87.6% on olmOCR-Bench with a 5-million-sample open dataset, reshaping document parsing for developers and small businesses.
New arXiv paper shows code-owned enforcement stops recommendation and trace leaks in enterprise LLM agents at 120/120 utility, where bolt-on guardrails drop to
New arXiv paper proposes EvoSOP, where AI agents turn repetitive steps into reusable procedures, cutting interaction rounds and boosting task success rates.
A Stanford preprint testing 85 LLMs finds bigger models better predict opinions and behaviour, but fail to capture cognitive biases like risk aversion.