RealFeedApp beta
Top science & technology news, July 29, 2026
- GPT-5.6 vs Claude Fable 5: Which Physical AI Performs Best?
Physical AI's success hinges on whether the modeled physics is correct. A model of an aircraft, a separation column or a charged particle can compile and run cleanly while the physics it encodes is impossible. The issue with agentic AI is that agents steer by feedback from tests, which are often written by the same agent.
- SenNet Collection
Nature Aging publishes a study mapping senescent cells across tissues, species, and lifespan.
- Choose DuckDB rather than SQLite
DuckDB is 100x faster than SQLite on a $16 server, providing better performance and saving space.
- Bioengineering Strategies to Interrogate and Modulate the Aging Immune System
The authors from Nature Aging highlight immunoengineering approaches including advances in extracellular matrix and biomaterials, microphysiological and organoid systems, and tissue-targeted vaccine and drug delivery methods. These strategies aim to improve understanding of immune aging and inform translational strategies.
- Engy – Verified LLM Inference
Verified inference means cryptographic proof that the exact open model you requested produced your output, not a cheaper or quantized stand-in. Run frontier open models like GLM-5.2, billed by the token.
- Plasmonic metamaterial time crystal driven at terahertz frequencies
The material achieves strong, ultrafast temporal modulation and transitions to the photonic time crystal regime with reduced plasmonic losses.
- Robot for Telescope Lift Tumbles Out of Control
NASA's robot to lift a space telescope has lost control and is tumbling uncontrollably. The mission, which was the first private company-assisted lift by NASA, needs to be re-pushed to higher orbit.
- Flycheck 38: Might & Magic
Today I'm happy to announce Flycheck 38 - the biggest and most ambitious release since taking over the stewardship of the project. It grows a proper story for language servers, gets modern editor-style inline diagnostics, and learns to apply fixes without giving up the checker model that made it popular in the first place.
- The Scientific Literature Is Poisonous to LLMs
Scientific literature is toxic for training useful LLMs, often belonging in multiple quadrants and masquerading as work in other quadrants.
- Cyera Acquires Oasis Security for $1B to Protect Growing AI Agents
Data security company Cyera has acquired Oasis Security for $1 billion. Oasis focuses on protecting the growing number of AI agents, which are becoming increasingly prevalent in cybersecurity.
- A digitally controlled silicon quantum processing unit
Researchers have created a silicon quantum processor that executes high-fidelity multiqubit circuits with all time-varying control signals generated by a digitally programmed cryogenic complementary metal–oxide–semiconductor controller.
- Can Vision-Language-Action Models Learn from Real-World Data Continually without Forgetting?
In this work, we construct a real-world continual learning benchmark comprising ten diverse sequential manipulation tasks across both single-arm and bimanual settings. We explore the ability of vision-language-action (VLA) models to learn continuously from real-world data without forgetting prior skills.
- Open-Source Magnetofluorescence Imaging Platform for Plate-Scale Screening of Magnetic Field Effects in Live Bacteria
The authors from bioRxiv present an open-source magnetofluorescence imaging platform designed for bacterial plate-scale screening of magnetic field effects in live colonies. The instrument integrates a programmable three-axis vector electromagnetism system.
- SPAK as a candidate mediator of TRPV4-induced NKCC1 activation and a possible pharmacological entry point in posthemorrhagic hydrocephalus
We demonstrate that the STE20 proline alanine rich kinase (SPAK) may be a candidate mediator of TRPV4-induced NKCC1 activation, and we propose that this pathway could represent a potential pharmacological entry point for treatment of posthemorrhagic hydrocephalus.
- Longevity AI and Clalit collaborate to improve heart disease and diabetes risk predictions
Longevity AI and researchers from Meir Medical Center of the Clalit Health Group announced a joint research collaboration to evaluate and recalibrate clinical risk prediction models for cardiovascular disease and Type 2 diabetes using longitudinal real-world data from Clalit’s integrated healthcare system.
- GLIDE: Guided Layerwise Hybrid Attention for Efficient LLM Inference
The research proposed GLIDE, a guided layerwise hybrid attention system that integrates sliding-window softmax attention with linear recurrent aggregation to reduce I/O and computational overhead during Large Language Model inference.
- LLM Scheming Inversely Scales with Pretraining Language Coverage
With the growing capabilities of frontier models, AI alignment becomes increasingly critical in high-risk deployment settings. While recent work has empirically demonstrated in-context scheming -- the covert pursuit of misaligned objectives while feigning alignment -- in frontier language models, most work has been performed exclusively in English, leaving a major gap in multilingual safety.
- ProMIS Neurosciences Reports Positive Early Safety and Biomarker Results for Alzheimer's Trial
ProMIS Neurosciences reported positive early safety and biomarker results from the PRECISE-AD trial of PMN310 in patients with mild cognitive impairment due to Alzheimer’s disease or mild AD. In a blinded analysis of 136 patients, the company reports no cases of amyloid-related imaging abnormalities-edema, 4.4% total ARIA (all mild ARIA-H), and no adverse events.
- Optimal Quantum Eigenvalue Transformation via Linear Combinations of Hermitian Matrices
A new formula for optimal quantum eigenvalue transformation using linear combinations of Hermitian matrices has been discovered. This could be significant in biohacking and AI fields.
- Quantum Transformer BSDE Solver via Multi-Layer Fully-Connected Variational Quantum Circuits
Proposes a quantum transformer-based solver for solving high-dimensional parabolic partial differential equations using variational quantum circuits. The method reformulates semilinear PDEs as backward stochastic differential equations and leverages model-based reinforcement learning.
