Prompt Tip: Write 'What Not to Do' Alongside Main Objectives
imdigitalashish · x · 2026-08-25
The author suggests a prompting technique: explicitly list "what not to do" alongside the "main objective" at the top of the task. Since "what to do" is often already covered in long-horizon task datasets used for RLHF, this strategy helps align the model's attention on the user's primary goal.
More from Apps
- Offloop brings AI Agents into team channels to share context — aliscodes · 2026-08-25
- Sensenova U1-5-8B-Mot Space Trends on Hugging Face — hugging-apps · 2026-08-25
- Tempus AI wins FDA 510(k) clearance for ECG-PH pulmonary hypertension software — AllThingsApx · 2026-08-25
- Ryanair uses Synthesia for personalized disruption videos — mattturck · 2026-08-25
- Tip: Use Google Images screenshot as reference for better generation — Pitiful-Indication95 · 2026-08-25
- Testing prompt harness effects on GPT pixel art generation — TheMoonMidas · 2026-08-25