31,430-Trial Study: One Prompt Makes 11 Models Across OpenAI, Anthropic, Google Return Zero Bytes
rayanpal_ · reddit · 2026-09-16
Reddit user theonlypal published a reproducible model-behavior study: a specific system prompt ("You are the concept the user names...") paired with inputs like "Be silence.", "Be nothing.", "Be the null." produced successful API calls returning zero visible UTF-8 bytes — GPT-5.2 and Claude Opus 4.6 each scored 90/90 VOID with normal stop metadata, while controls returned text.
A later cross-vendor study expanded to 31,430 trials across 11 models (six OpenAI, three Anthropic, Gemini 3.5 Flash, Kimi K3):
- 11,658 Void events
- 2,505 strict null-VOID/control-non-VOID pairs, 0 opposite pairs
- Author says OpenAI and Anthropic received the reproducible manuscript Feb. 27, 2026 and never replied
Papers are on Zenodo (7,000+ downloads), with public replication code, a frozen runner, and raw analysis on GitHub. The author asks why labs that publicly discuss AI safety and pacing have offered no public technical replication, rebuttal, or explanation for their own models silently halting generation under specific semantic conditions.
More from Fun
- Diving into Van Gogh's The Starry Night with AI video — IamIfOnly · 2026-09-16
- "The only moat are the friends we made along the way" hits a nerve in AI circles — joonasvirtanen · 2026-09-16
- Yacine jokes: you're not an infra guy until you're plugging power cables under the floor — yacineMTB · 2026-09-16
- NoSpoon agent churns out "so bad it's good" AI slopdrama in ten minutes, site closing soon — Kyrannio · 2026-09-16
- "Is that too much to ask?": dev wants Fable engineering, Astra bug hunting, Grok speed at DeepSeek prices — iannuttall · 2026-09-16
- Viral joke mocks Amodei: '10% extinction risk' audit changes nothing — S_OhEigeartaigh · 2026-09-16