Fetching from the wire…
Research2026-08-28 · source-backed
The first defect state-aware multi-round review benchmark: 2,269 real tasks across five languages, each annotated with defect description, type and severity plus cross-round state labels tracking a defect's full trajectory (arXiv 2608.27442). Mainstream models degrade significantly as rounds increase, disproportionately missing semantically complex or low-salience defects. Error analysis names cross-round temporal misalignment and inadequate long-range memory. Direct instruction: re-anchor the defect list explicitly each round instead of trusting the conversation to carry it.
Each link below shares sources, entities, or timing with this story.
Simon Willison released LLM / Shared entity: Bench / Shared topic / Earlier coverage
Linked by a graph relationship (Simon Willison released LLM); both cover Bench; overlapping topics (benchmark, code).
Simon Willison released LLM / Shared entity: LLM / Earlier coverage / Tension
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-08-16.
LLM uses OpenAI / Shared entity: LLM / Earlier coverage / Tension
Linked by a graph relationship (LLM uses OpenAI); both cover LLM; earlier LLM coverage from 2026-07-27.
Simon Willison released LLM / Shared entity: LLM / Earlier coverage / Tension
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-06-19.
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-06-18.
LLM uses OpenAI / Shared entity: Bench / Earlier coverage / Tension
Linked by a graph relationship (LLM uses OpenAI); both cover Bench; earlier Bench coverage from 2026-04-25.
LLM uses OpenAI / Same source domain / Shared topic
Linked by a graph relationship (LLM uses OpenAI); reported by the same outlet (arxiv.org); overlapping topics (code, each).
Simon Willison released LLM / Same source domain / Shared topic
Linked by a graph relationship (Simon Willison released LLM); reported by the same outlet (arxiv.org); overlapping topics (code, review).