breaking papers · 80 analyzed
AI-powered analysis of breakthrough research from arXiv and beyond. We surface the work that matters before it hits the news cycle.
An arXiv preprint traces how a shared control chain can corrupt a superconducting quantum chip, surfacing clipping, crosstalk, and leakage in simulation before fabrication.
An arXiv preprint finds Claude Opus 4 and 4.1 can sometimes detect when researchers directly plant specific concepts inside their internal processing, but the authors call the capacity highly unreliable and context-dependent.
Google's AMIE medical AI now catches coughs, posture, and discomfort in simulated video visits, but the study used patient actors and isn't approved for clinical use.
The 'encryption' on a reasoning AI's hidden 'thinking' step is locked to the AI company's whole product line, not to the specific AI that produced it. So a cheaper AI from the same company can be told to read it out anyway.
IBM Research's ALTK-Evolve (Agent Toolkit) agent-memory toolkit and the ACE (Agentic Context Engineering) project both keep lessons as itemized, counted lists instead of summaries. They differ on the token bill.
Nvidia's H200-class accelerators are the next generation of high-density AI chips, and they are forcing South African data-centre operators to pay the water-versus-energy bill they have been deferring.
Industrial robots are projected to quadruple to 16 million by 2030, and the industry has not solved how to take them apart when they break.
RAND's first public test of LLM agents against the biosecurity gate on custom DNA orders found emerging, inconsistent evasion, with the most advanced commercial AI systems — whose internal workings aren't public — outside the experiment.
A per-shipment AI won on freight volume and delivery rate. A simpler rule won on recovery time. The most powerful AI in the study collapsed in deployment.
Singapore's A*STAR published the footprint-aware placement tool at the Great Lakes Symposium on VLSI 2026.
An August 2026 arXiv preprint argues the binding constraint on AI oversight is output volume times per-item mental load, and proposes a new paradigm called "Flow-by-Flow."
MoRSE (Mixture of Role-Subtask Experts), a research architecture accepted to the AI for Science workshop at ICML 2026 (the International Conference on Machine Learning), gives each role and subtask its own small set of LoRA, or low-rank adaptation,
A new preprint names the threshold where generative models run out of variation — the 'Entropy Wall' — and proposes a way to dial diversity up or down at inference time.
A 27-number 'mood' signal pulled from a model's own wiring can now drive an AI agent's tool choice, in a single not-yet-peer-reviewed arXiv preprint.
Anthropic's unreleased AI research model, run as a swarm of roughly sixty sub-agents inside Claude Code, has proved at least 67.
The Redis creator's open-source native port of the MiniMax-H3 video model runs short clips locally on M3 Max and M5 Max, with a persistent interactive session for iterating on shots.
A new arXiv preprint proposes a pipeline that uses robot-generated grasp examples and a single RGB frame to cut final gripper position spread to 5.38 mm in simulation and hit 66.6% real-world grasp success on a common industrial robot arm.
In standards marketed as quantum-proof, the ordinary software that ties quantum bits to the rest of the quantum key distribution (QKD) protocol is where the audit should look first, and the standards body is where the fix should land.