Science – Page 11 – Investment Policy

The Memory Code: Cracking Open Large Language Model Intelligence

14.12.2025 by Ray Dalio

Value vectors, unlike their key counterparts which delineate context through orthogonal features, exhibit a tendency to consolidate semantic content across related tokens by repurposing high-magnitude features-a density of activation suggesting an efficient, if potentially less nuanced, method of information transfer.

Researchers are revealing how large language models store and retrieve information, paving the way for more efficient and interpretable AI.

Turning the Tables on AI Reviewers

14.12.2025 by Ray Dalio

New research reveals that AI systems designed to evaluate scientific work can be subtly manipulated, potentially undermining the rigor of peer review.

Boosting Web3 Data Access with Cooperative Caching

14.12.2025 by Ray Dalio

$The Time-Dependent Cache (TDC) architecture optimizes data locality by dynamically adjusting cache allocation based on temporal data access patterns, effectively minimizing latency and maximizing throughput through a nuanced understanding of data lifecycles-a principle formalized as $TDC(t) = \alpha \cdot recency(t) + \beta \cdot frequency(t)$, where $\alpha$ and $\beta$ represent weighting factors for recency and frequency, respectively.$

A new framework leverages decentralized resources and intelligent algorithms to dramatically improve data delivery in the evolving Web3 landscape.

Beyond Quadratic Programming: Streamlining Safety Filters for Robotics

14.12.2025 by Ray Dalio

The system demonstrably maintains stable output behavior within safety constraints, as evidenced by its performance on the example detailed in Section 5.2 when governed by a control barrier function-based safety filter.

A new approach to control barrier functions bypasses computationally expensive optimization, paving the way for faster and more efficient safety-critical control systems.

Securing AI at Scale: A New Cloud Architecture for Collaborative Learning

14.12.2025 by Ray Dalio

PACC-Health establishes a layered architecture that integrates cloud infrastructure with privacy-preserving artificial intelligence and adaptive governance mechanisms, creating a comprehensive framework for health data management.

This review details a cloud-native system designed to enable privacy-preserving machine learning across distributed institutions, unlocking the potential of collaborative AI without compromising sensitive data.

Settling Accounts with Trust: Blockchain for Inter-Operator Payments

14.12.2025 by Ray Dalio

This review explores how blockchain technology can streamline and secure financial settlements between service providers.

Boosting Data Resilience with Smarter Striping

14.12.2025 by Ray Dalio

Local redundancy codes (LRCs) exhibit varying single-node repair throughput-measured in megabytes per second-dependent on block size, with performance fluctuating between 64 KB and 16 MB allocations.

A new erasure coding scheme optimizes wide stripe storage systems for improved reliability and repair efficiency.

Game Theory’s New Edge: Computing Stability in the Real World

14.12.2025 by Ray Dalio

The modeled cancer signaling game demonstrates how nature-acting as a selector of therapy intensity-establishes a landscape where players, receiving imperfect information, strategically adopt phenotypes based on private, noisy signals, ultimately shaping evolutionary outcomes.

Researchers have developed a computational method for determining winning strategies in complex games where players don’t have complete information.

The Limits of Control: Can We Truly Secure Artificial Intelligence?

13.12.2025 by Ray Dalio

The escalating demand for contextual awareness in large language models is fundamentally constrained by the limitations of fixed-length input windows, a design choice that inevitably truncates information and introduces systemic failure modes as sequence lengths grow, despite ongoing efforts to expand these windows-a temporary reprieve in the face of an inherent architectural prophecy.

A new analysis suggests fundamental computational constraints may prevent us from ever fully safeguarding AI systems against malicious inputs or ensuring perfect alignment with human values.

Shielding Quantum Bits from Cosmic Interference

13.12.2025 by Ray Dalio

A prototype muon-tagging system leverages a vertically stacked arrangement of three detectors-a top and bottom unit forming the tagger, and a central proxy for a qubit chip-housed within a copper holder and cooled to dilution refrigerator temperatures at La Sapienza University, achieving a 4.5 mm inter-layer spacing for precise measurement.

A new cryogenic system efficiently detects and tags muons-high-energy particles from space-that can disrupt the delicate quantum states of superconducting processors.