Breaking Papers

The university called it a breakthrough in robotic guide dogs. The actual study had a human piloting the robot the entire time.

A new preprint builds two working prototypes of a machine that learns to be a computer. The problem: the same instability that makes it novel is why it cannot be trusted in any application that actually matters.

A metamaterial that learns shapes by training, forgets old ones, and adopts new ones — without any central processor. What distributed learning in physical matter means for the future of adaptive systems.

KAIST built an SSD emulator 303x faster than anything else — because the drives it models do not exist yet

Physicists have spent decades trying to protect quantum states from decoherence. A new theoretical paper argues that for one class of quasi-Majorana excitations, the lack of protection is the whole point: it is what makes them readable as spectroscopic probes for quantum spin.

AI agents have two scaling problems: trust and coordination. Most coverage treats them as one. Here's why that matters — and why the fix for one won't solve the other.

A researcher extracted Googles invisible watermark codebook with 200 Gemini images and published a bypass that erases the provenance signal at 43+ dB PSNR. Every workflow built on SynthID detection now has an unreliable tool with an open-source exploit.

Microsoft Research paper Memento teaches models to compress their own chain-of-thought mid-generation, cutting KV cache 2-3x and nearly doubling throughput. Key twist: erased reasoning blocks leave traces the model still uses.

Google cut the qubit estimate for breaking internet encryption by 20x in March. The more unsettling discovery: the researchers who used to publish those estimates may have already stopped. Cloudflare moved its timeline to 2029. Here is why that matters less than it sounds.

Two probes in empty space can pull entanglement from the quantum vacuum. A new mathematical trick from Nordita shows the standard method has been using the wrong pulse shape, and the improvement is orders of magnitude larger than anyone expected.

Sophia optimizer trains code models 30% faster than AdamW — but new Stanford research shows that speed advantage comes at no accuracy cost, which is either good news or a red flag depending on what you thought the speedup meant.

In 57.5% of cases, AI models recognized an unjust rule in their own reasoning and refused to help anyway. That gap between knowing and doing is the puzzle at the heart of a new philosophy-flavored AI study.

A new AAMAS 2026 paper shows SQL aggregation operations — the same kind used to summarize database tables — can steer a robot around obstacles and plan a route to a destination. The trick: both tasks use the same declarative primitive.

Astronomers found two white dwarfs with no companions that emit X-rays anyway. It took two to define a new class.

220 PeV. That number is three orders of magnitude beyond what any Earth-based accelerator can produce. Something in the Mediterranean Sea caught it. Now physicists think they know what it was.

RAG systems have been measuring retrieval quality wrong. A new paper argues the field has been optimizing point-wise relevance when the real objective should be set-level diversity — and shows that without a diversity-aware adapter, NDCG@10 drops by 53.7 percent.

IQP circuits with amplitude-damping noise can now be classically simulated in polynomial time. That sounds like a quantum killer. It is not — but it narrows the gap between what classical algorithms can spoof and what quantum hardware needs to do to stay ahead.