30 prompts fine-tune AI to near-optimal energy storage control
Reinforcement fine-tuning with 30 prompts cut an open-weight model's building emissions to 61.2 kg-CO2, near the 60.8 optimum, a preprint shows.
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.
306 stories
Reinforcement fine-tuning with 30 prompts cut an open-weight model's building emissions to 61.2 kg-CO2, near the 60.8 optimum, a preprint shows.
Researchers find LLM judge bias lives in a low-dimensional subspace inside model hidden states, and steering along it can switch bias on and off.
MCP security scanners flag 96.89% of servers as risky, but a study of 64,611 servers finds fewer than half of alerts are true positives.
New arXiv preprint Prezta replaces application gateways at critical infrastructure edges with zero-knowledge proofs generated on client devices.
Researchers propose Q-DIBA, a backdoor attack generating a unique trigger for each input to quantum neural networks, evading three tested defenses.
A new arXiv preprint introduces StoryTeller, a training-free system that keeps film audio descriptions coherent for blind and low-vision viewers.
New arXiv preprint shows Transformer training on inductive reasoning can be confined to a low-dimensional manifold for automatic circuit detection.
An arXiv preprint finds sampling temperature on retrieval-augmented LLMs controls how strongly ideological framing from source documents bleeds into
Brisbane's decades-long boom is cooling and the balance of power is shifting from sellers to buyers, as new home loans fall 6.2% nationally.
AdvancedMathBench tests AI on graduate and doctoral math proofs, finding frontier models struggle with both writing and verifying advanced mathematics.
Reasoning scaffold lifts GPT-4.1-mini by 0.21 but degrades GPT-5-mini by 0.63, arXiv study finds, exposing an architecture-dependent split.
A new arXiv survey of 60+ papers finds federated learning for vehicle intrusion detection relies on artificial data splits and weak attack tests.