#ai-research
6 articles exploring ai-research. Expert insights and analysis from our editorial team.
Articles
Executing Programs Inside Transformers: The Inference Breakthrough Nobody Expected
A new architecture from Percepta embeds a program interpreter directly into transformer weights, achieving logarithmic-time execution lookups that could reshape how AI agents handle deterministic computation—if the early claims survive scrutiny.
Swarm AI for Prediction Markets: Collective Intelligence Gets an Algorithm
MiroFish, an open-source swarm intelligence engine with 20k+ GitHub stars, deploys thousands of AI agents to simulate social dynamics and forecast outcomes. Early benchmarks suggest multi-agent collective reasoning can match human crowd accuracy, but the gap between simulation and validated prediction remains wide.
Why LLM Performance Gains Are Slowing—and What Comes Next
New research from METR reveals roughly half of AI-generated code PRs that pass automated tests would be rejected by human maintainers—exposing a fundamental gap between benchmark scores and real-world capability. Pre-training scaling is hitting structural limits, but three distinct scaling frontiers are emerging to replace it.
DjVu and Its Connection to Deep Learning: An Unexpected History
DjVu, the 1998 image compression format created by future Turing Award winners at AT&T Labs, pioneered techniques like layer separation and multi-resolution encoding that directly influenced modern neural image compression methods.
WiFi DensePose: Full-Body Tracking Through Walls Using Your Router
WiFi-based DensePose technology uses commodity mesh routers to perform dense human pose estimation through walls, raising critical privacy concerns as researchers demonstrate how standard WiFi signals can track body movements and positions without consent or line of sight.
AI Code Generation Benchmarks 2026: Which Model Actually Writes Better Code?
Claude 3.5 Sonnet, GPT-4o, Gemini 2.5 Pro, and open-source models like Qwen2.5-Coder and DeepSeek show competitive performance on benchmarks, but real-world coding tasks reveal significant gaps between benchmark scores and practical utility.