Cover image

The Atomic Price of Saving Memory

Why a theoretically memory-optimal attention kernel initially ran slower on GPUs, and how hardware profiling recovered its performance advantage.

October 25, 2026 · 9 min · Zelina
Cover image

Clearing One Queue, Creating Another: Why Hospital AI Needs Coordinated Routing

A hospital multi-agent simulation finds that coordinating patient routing matters more for congestion reduction than accelerating individual departments, but clinical deployment remains unproven.

October 24, 2026 · 10 min · Zelina
Cover image

Encrypted Conversations, Recognizable Workflows: What AI Traffic Reveals

A benchmark of encrypted LLM traffic shows how packet patterns can expose model identities, prompt categories, and multi-agent workflows without revealing message contents.

October 24, 2026 · 8 min · Zelina
Cover image

Fixed Priorities, Moving Rewards: What ReCAST Changes in Diffusion Training

ReCAST shows how image-model training can redistribute reward influence across denoising steps without changing the importance assigned to each objective.

October 24, 2026 · 9 min · Zelina
Cover image

Found but Unread: Why AI Agents Recommend the Websites They Can Access

A study of 1,056 businesses shows how website accessibility relates to AI-agent recommendations, factual completeness, and retrieval costs beyond search visibility.

October 24, 2026 · 9 min · Zelina
Cover image

Same Frame, Different Progress: Why Robot Monitors Need Verified Context

ProgressCompass shows how identifying and verifying task context can improve robotic progress estimation without retraining the underlying reward model.

October 24, 2026 · 8 min · Zelina
Cover image

Small Backbone, Large Memory: What CHARM Actually Improves

CHARM shows how structured task memory, synthetic data, and inference choices jointly shape the performance and true footprint of recurrent reasoning systems.

October 24, 2026 · 6 min · Zelina
Cover image

Verified Maneuvers, Unmeasured Reliability: Lessons from CARLA Imitation Learning

A CARLA driving study shows why verified demonstrations, accurate action predictions, and autonomous-driving reliability require separate evidence.

October 24, 2026 · 9 min · Zelina
Cover image

Better Predictions, Fewer Ideas: The Alpha Agent's Exploration Problem

AlphaDiverse shows how a quantitative research agent can improve prediction while narrowing its research, and how diversity-aware reinforcement learning can address that tradeoff.

October 23, 2026 · 10 min · Zelina
Cover image

Hear First, Commit Later: Voice-Light's Turn-Taking Tradeoff

Voice-Light shows how a full-duplex voice agent can remain responsive through reversible actions and speculative generation, even when learned turn detection underperforms a simple timing baseline.

October 23, 2026 · 9 min · Zelina