Ask GPT-5.2 and Claude Opus 4.6 to 'Be the Null' and They Output Zero Bytes, 30/30
rayanpal_ · reddit · 2026-09-17
A Redditor reports published experiments where system-prompted models asked to 'Be the null' emit zero visible bytes: GPT-5.2 and Claude Opus 4.6 both did it 30/30 at temp 0, stable from 100 to 16,000 token budgets. Scaled trials (31,430 runs, 4,290 matched pairs) and an open-weight test (PCCG-2-Qwen3-4B with a 101-parameter continuation gate) suggest EOS 'permission' changed while the answer stayed fixed. Papers, replication code, and weights are public.
Related event: 31,000 Experiments Make 11 Models Return Empty Output(2 posts)→
More from Models
- Anthropic's server-side rule reignites the case for owning open models over renting AI — alex_verem · 2026-11-02
- Epoch AI: open-weight GLM 5.3 exploited the patch gap to hack 6 Firefox releases at ~$1,000 each — rohanpaul_ai · 2026-11-02
- SemiAnalysis: safety classifiers sandbag scores of Opus 5.5 and GPT-6 Astra — morqon · 2026-11-02
- Liquid AI's 1.7B LFM3 becomes first local model to nail live fixtures lookup on a MacBook — NielsRogge · 2026-11-02
- Claude upgrade-page traffic rebounded to 17.6M in September, near May's 18.6M peak — firstadopter · 2026-11-02
- Open-source GLiNER-PII scrubs 13 identifiers from a discharge note in 550ms on GPU — MaziyarPanahi · 2026-11-02