Prompt Engineering 101

Modern AI models evolve rapidly, rendering old prompting techniques obsolete. Recent academic research suggests avoiding tag questions like ‘right?’ and the ‘step-by-step’ instruction, as they can bias AI responses or be less effective than direct questions. Instead, focus on clear goals and limit the number of rules in your prompts, as AI struggles to adhere to more than three simultaneously. Examples can also be detrimental, with specifying goals proving more effective.

• Avoid ending prompts with tag questions like ‘right?’; use neutral phrasing or ‘maybe?’ instead.
• The ‘step-by-step’ approach is less effective than direct questioning, especially when combined with a specified role.
• AI can hallucinate, and even expert reviewers miss incorrect citations, so always verify AI-generated information.
• AI models reliably follow a maximum of three rules; more rules significantly decrease accuracy.
• Clearly stating goals is more effective for modern AI models than providing examples.
• A prompt template incorporating these principles is recommended for consistent results.
  https://bender.layer3.press/articles/3566c417-13d3-450f-bfd9-ab098b3124fe