Signals this week

Agents & tooling (1) — Parallel agents in Zed
Model eval & reliability (1) — OMIBench: Benchmarking Olympiad-Level Multi-Image Reasoning in Large Vision-Language Model

Top links (curated)

Parallel agents in Zed (Hacker News) — https://zed.dev/blog/parallel-agents
OMIBench: Benchmarking Olympiad-Level Multi-Image Reasoning in Large Vision-Language Model (arXiv cs.AI) — https://arxiv.org/abs/2604.20806v1
Website streamed live directly from a model (Hacker News) — https://flipbook.page/
Over-editing refers to a model modifying code beyond what is necessary (Hacker News) — https://nrehiew.github.io/blog/minimal_editing/
Convergent Evolution: How Different Language Models Learn Similar Number Representations (arXiv cs.AI) — https://arxiv.org/abs/2604.20817v1
ParetoSlider: Diffusion Models Post-Training for Continuous Reward Control (arXiv cs.LG) — https://arxiv.org/abs/2604.20816v1
Diagnosing CFG Interpretation in LLMs (arXiv cs.AI) — https://arxiv.org/abs/2604.20811v1
Automatic Ontology Construction Using LLMs as an External Layer of Memory, Verification, and Planning for Hybrid Intelligent Systems (arXiv cs.AI) — https://arxiv.org/abs/2604.20795v1
Can "AI" Be a Doctor? A Study of Empathy, Readability, and Alignment in Clinical LLMs (arXiv cs.AI) — https://arxiv.org/abs/2604.20791v1
Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model (Hacker News) — https://qwen.ai/blog?id=qwen3.6-27b