Adding "Do not guess" to prompts cuts made-up LLM fields from 71% to 20%
FKJ · hn · 2026-09-28
The author benchmarked LLM data extraction and found that adding a single line — "Do not guess" — to prompts reduced fabricated fields from 71% to 20%.
- Setup: tasks with leading/misleading fields, measuring how often the model invents non-existent ones
- Finding: LLMs follow hints in the prompt so closely that one line of instruction sharply suppresses confabulation
- Takeaway: explicitly telling models not to guess is a very cheap anti-hallucination measure in production prompts
Full benchmark details in the original post.
More from Research
- Astra saw through perturbed Zork observations, calling it a 'variant' — tw_killian · 2026-09-28
- FuseReg: Layer-Fusion Regularization Cuts gFID up to 29% in Representation Autoencoders — USC-PSI-Lab · 2026-09-28
- Kaggle Game Arena: Evaluating LLMs via Head-to-Head Chess, Poker, and Werewolf — kaggle · 2026-09-28
- InternW0-Δ: A World Action Model Trained on 20K+ Hours of Open Robot Data, Fully Open-Sourced — Xingyu Miao · 2026-09-28
- Berkeley's Morphometric Imitation Hits 89.3% Zero-Shot Real-World Success Across 3 Robot Hands — Berkeley · 2026-09-28
- Fermi estimate: brain may pack 500k-5M molecular switching units per 'parameter' — JosephJacks_ · 2026-09-28