Technology & AI
CompactionRL lifts AI coding agents 7 points on SWE-bench
Tsinghua's reinforcement learning method teaches long-horizon AI agents to compress their own context, lifting open models on coding benchmarks and entering GLM
Technology & AI
Daily AI and technology news decoded in plain English — models, chips, agents and research, and what each development actually means for you and your business.
418 stories
Technology & AI
Nvidia RTX Spark: 1-petaflop power for personal AI
A new superchip brings data centre power to laptops, challenging Apple and Intel.
Technology & AI
kNNGuard: 10x faster AI safety without retraining
A new method uses hidden activations to block bad prompts instantly, requiring no model retraining.
Technology & AI
XGBoost Scored 0.98 Against Hackers. Its Explanations Still Collapsed.
New arXiv research reveals a dangerous blind spot in cybersecurity AI: models can resist attacks on paper while their decision logic falls apart.
Technology & AI
Masked image modelling beats contrastive learning in messy distributed AI training, Sydney-led team shows
New ICLR 2026 research finds masked image modelling is inherently more robust than contrastive learning when decentralised unlabelled data differs across nodes.
Technology & AI
Training-free AI guardrail runs 10x faster using 50 prompts
An arXiv paper claims kNNGuard reads hidden LLM activations to block unsafe prompts without fine-tuning, cutting latency and setup time to seconds.
Technology & AI
Mobile coverage map standard 2026: ACCC warns on misleading claims
ACMA's new 4G and 5G rules require a standard method across Australia, giving the ACCC a benchmark to judge network claims it previously had to drop.
Technology & AI
New EvoPolicyGym benchmark forces AI agents to improve policies under a tight feedback budget
Researchers measure whether agents can iteratively edit executable policies, with trajectory diagnostics replacing single final scores.
Technology & AI
Trust boundary semantic gap: Why passing every security test is no longer enough
Researchers analysed 75 breaches and found a hidden flaw—syntactic validation clears, but meaning fails when data crosses trust boundaries.
Technology & AI
VeriChat: Multi-agent AI claims 87.73% faithfulness in chip-security verification
An unreviewed arXiv paper shows how three specialised AI agents catch chip-security flaws that general chatbots miss.
Technology & AI
WorldSample: New AI Method Cuts Real-Robot Training by 59% With Synthetic Data
An unreviewed arXiv paper shows how grounding world models on real rollouts can slash training time and boost manipulation success.
Technology & AI
ACCC welcomes mobile coverage map standard and warns telcos on misleading claims
Telcos must now publish comparable 4G and 5G coverage maps, but broad geographical claims still sit outside the rules.