PAPER PLAINE

Fresh research, simply explained. Updates twice daily.

Privacy is Fungibility: Why Endogenous Tokens Are Not Money

Why most cryptocurrencies don't work like real money

Most cryptocurrencies fail a fundamental test of money: they don't protect users' privacy the way cash does. The researchers show that blockchain ledgers expose transaction details in ways that create harmful power imbalances between parties, even when encryption is added on top. This means cryptocurrencies and stablecoins built on these systems are missing something essential that makes money actually work.

If cryptocurrencies aren't functioning as real money, they can't fulfill the role their backers envision—whether as payment systems, stores of value, or alternatives to government currency. This affects how regulators should treat these assets and what users should realistically expect from them. It also matters for anyone considering stablecoins or blockchain-based central bank digital currencies, since the underlying ledger design creates privacy vulnerabilities no amount of encryption can fully solve.

Watts vs. Bytes: Turning Data Centers into Grid Assets via Storage Compute Co-Optimization

Making data centers help power grids by timing computing and batteries together

Data centers can slash operating costs and help stabilize power grids by coordinating when they run computer tasks with when their backup batteries charge and discharge. A new framework shows that when grids get tight and can't accept more peak power, this coordination doubles the value of the battery system while still completing computing work on time.

As data centers consume more electricity, grids face real capacity limits. This approach lets data centers become grid helpers instead of problems—they can absorb power at off-peak times and reduce demand during crunch hours. For grid operators, that means deferring expensive infrastructure upgrades. For data center operators, it means lower bills and new revenue from selling grid services. Under tight grid conditions, the value compounds: the same battery system becomes twice as valuable simply because computation and storage work as a team.

Confirming Correct, Missing the Rest: LLM Tutoring Agents Struggle Where Feedback Matters Most

Why AI tutors spot perfect answers but miss the learning opportunities

Large language models used as tutoring agents excel at recognizing correct student solutions but systematically fail at distinguishing between wrong answers and right answers that use flawed reasoning—exactly the feedback that helps students improve. Across seven different AI models tested on 10,836 logic problems, the models over-accepted incorrect reasoning and over-rejected valid but inefficient approaches, suggesting these failures stem from how the models are built rather than from missing information.

As schools and tutoring platforms increasingly deploy AI as learning tools, this gap could undermine their effectiveness. Students might receive approval for sloppy reasoning or harsh rejection for approaches that actually work, neither of which promotes real understanding. The research suggests that AI tutors work best not as standalone replacements for human judgment, but as part of a hybrid system where traditional logic-based systems diagnose student reasoning while AI handles open-ended conversation and encouragement.

Performance Gains in Quantum SAT Solvers Using ESOP Encoding

A smarter way to write problems for quantum computers to solve

When quantum computers try to solve logic puzzles using Grover's algorithm, the way you write down the puzzle matters enormously for how many quantum resources you need. Researchers found that switching from the standard way of writing these puzzles (CNF) to a different format called ESOP cuts the number of quantum bits needed, reduces complex quantum gates, and shrinks the overall circuit — sometimes substantially — while solving the same problems.

Quantum computers are still extremely resource-constrained; every qubit and gate matters for whether a quantum machine can actually run a useful calculation. This encoding trick could let quantum computers tackle larger satisfiability problems with the limited hardware we have today, moving these machines closer to practical applications in optimization, scheduling, and constraint solving — areas where SAT solving is already central to industry.

ATLAS: Agentic or Latent Visual Reasoning? One Word is Enough for Both

One special word that lets AI think visually without slowing down

Researchers created ATLAS, a system where a single special word acts as both a visual reasoning step and an executable operation, eliminating the computational waste of generating intermediate images. The approach outperforms existing methods on visual reasoning benchmarks while remaining compatible with standard AI training techniques.

Current AI systems that reason about images either generate entire intermediate pictures (expensive and slow) or use hidden calculations that don't generalize well. ATLAS cuts through this tradeoff by embedding visual reasoning into a single token that's processed like normal text, making visual reasoning faster and more practical to deploy. This could meaningfully reduce the computational cost of AI systems that need to understand images and work through complex visual problems step-by-step.

RoSHAP: A Distributional Framework and Robust Metric for Stable Feature Attribution

Making machine learning explanations reliable across different data splits

Machine learning models often rank features differently depending on random choices in training, making it hard to trust which factors actually matter. This paper introduces RoSHAP, a new method that accounts for this natural variation by treating feature importance as a distribution rather than a single number, and shows it identifies truly influential features more reliably than standard approaches.

When doctors, banks, or regulators rely on machine learning to make decisions, they need to know which factors the model actually used—not just a ranking that changes every time the model is retrained. RoSHAP makes those explanations stable and trustworthy. The method also lets companies use fewer data inputs while keeping the same prediction accuracy, reducing complexity without sacrificing performance.

The Value of Information: A Puzzle

Why stock traders earn far less from secrets than they pay to find them

Researchers measured how much money informed traders actually make from their information advantage in US stock markets and found it's about $3.5 million per stock annually—surprisingly small. The real puzzle: investors collectively spend roughly 17 times more in fees chasing superior returns than the actual gains those advantages deliver.

This finding suggests that most of the money flowing into active fund management, algorithmic trading, and research-driven strategies may be wasted effort. If the genuine payoff from having better information is genuinely this thin, it raises hard questions about whether the enormous resources devoted to beating the market could be better spent elsewhere—and whether individual investors chasing high-fee funds are effectively paying for a mirage.

RefDecoder: Enhancing Visual Generation with Conditional Video Decoding

Making AI video generators keep fine details from reference images

Video generation models typically use heavily conditioned networks to create new frames but leave their final decoder step unconditional, losing fine details and consistency with the input image. Researchers introduced RefDecoder, which feeds the reference image directly into the decoder at every step, improving visual quality by up to 2.1 decibels and maintaining consistency across subjects and backgrounds. The upgrade works with existing video generators without retraining and extends to tasks like style transfer and video editing.

Video generation powers content creation tools, special effects, and AI video platforms. This improvement means generated videos now better match what users provide as reference material—sharper, more consistent, and closer to the original—making the technology more practical for real production work. Because RefDecoder retrofits into existing systems, it can improve countless deployed video tools immediately.

Approximate Macroscopic Dynamics of Spiking Neural Networks Based on Solutions to the Transport Equation

How neurons' starting electrical states shape their collective firing patterns

When a population of neurons receives changing inputs, their firing rates fluctuate in ways that depend on where each neuron started electrically before stimulation began. Researchers derived a mathematical model that predicts these fluctuations by tracking how the distribution of neural voltages evolves over time, rather than assuming neurons behave in a steady synchronized state.

Brain activity emerges from billions of neurons firing in complex patterns, and understanding what drives these patterns is central to neuroscience. This work explains why the same stimulus can produce different collective firing patterns depending on recent neural history—a finding that could improve how researchers interpret experimental recordings and build more realistic computational models of brain circuits.

EntityBench: Towards Entity-Consistent Long-Range Multi-Shot Video Generation

Testing AI's ability to keep characters consistent across long video sequences

Researchers built EntityBench, a standardized test for video-generation AI that measures whether systems can keep the same characters, objects, and locations consistent across long sequences of shots. The test, based on real TV episodes, reveals that existing systems struggle dramatically when characters reappear after long gaps, and a new memory-based approach (EntityMem) achieved significantly better character consistency than existing methods.

Generating coherent multi-scene videos is a step toward AI that can create longer, more complex visual stories — from TV-like narratives to advertisements and filmmaking. Right now, when a character disappears from frame for several minutes then reappears, AI systems often render them looking completely different, breaking the viewer's experience. EntityBench gives researchers a concrete way to measure and improve this problem, accelerating progress toward AI that can maintain visual continuity over extended sequences.

Multi-regime Markov-switching models with time-varying transition probabilities: An application to U.S. Treasury yields

When financial markets switch moods, can we predict how long they'll stay that way?

Bond market behavior shifts between different regimes—periods of stability, volatility, or trend changes—but researchers have struggled to model when those shifts occur. This study develops better statistical tools to capture these regime switches and shows that while these models can predict bond yields reasonably well, getting the timing of regime changes right is much harder than previously thought, revealing a fundamental limitation in how economists identify these transition mechanisms.

Treasury bonds underpin the U.S. financial system, influencing everything from mortgage rates to pension valuations. Better models of when bond market behavior fundamentally shifts would help investors, central banks, and policymakers anticipate dangerous transitions—like shifts toward persistent volatility—rather than getting caught off guard. However, this paper's finding that transition timing mechanisms are nearly impossible to pin down statistically suggests that even sophisticated models may give false confidence in predicting exactly when the next regime change will strike.

Deep Mixture of Experts Network for Resource Optimization in Aerial-Terrestrial CF-mMIMO Systems under URLLC

Smart networks that juggle speed, power, and reliability for flying and ground signals

A new wireless system combines drones and ground stations to deliver extremely fast, reliable communication while using less power and bandwidth. The system uses machine learning to predict signal quality and automatically adjust power levels based on what each user actually needs, rather than applying one-size-fits-all settings.

6G networks need to handle time-critical applications like autonomous vehicles and emergency response—situations where delays or dropped connections can cause harm. This approach reduces the power and spectrum waste that typically comes with ultra-reliable communication, making it practical to deploy these networks without enormous infrastructure costs or energy consumption.