Claude auto-files a bug report about itself: TLDRs still too long after user asked for shorter
andrey_kurenkov · x · 2026-10-13
Developer andreykurenkov shared a screenshot of Claude (Opus 5.5 in CLI) autonomously drafting a feedback report about its own failure, titled "TLDR answer was too long after the user had already asked for shorter answers on the same topic."
The report classifies itself as a bug with failure mode "instruction following": the user had asked for brevity three times ("be a bit shorter...", "I want this to be like 3 sentences", "shorter pls"), yet the model still returned a 6-point list plus two closing paragraphs. Claude included repro steps, evidence, and auto-sent the transcript to Anthropic. The author quips that Opus 5.5 is "not quite AGI yet."
More from Fun
- Ten AI agents, one email: users mute agents after notification overload — annbordetsky · 2026-10-13
- AI bot turns all 14 Starship launches into an 11-minute recap, Elon Musk amplifies it — elonmusk · 2026-10-13
- 9-year-old yawns at another SpaceX launch: 'rockets are just routine now' — MatthewBerman · 2026-10-13
- Chollet: agriculture is 'self-replicating, recursively self-improving open-source nanotech' — fchollet · 2026-10-13
- 'Remember Stack Overflow?' — one line nails the LLM-era decline of dev Q&A — jbarbier · 2026-10-13
- Omarchy Linux running natively on a $50 digital photo frame: Allwinner A33, 1GB RAM, Hyprland boots — gregmushen · 2026-10-13