Prompt engineering has emerged as a controversial but very important space within the thriving subject of AIGC. In his presentation on State of GPT, Andrej Karpathy, Co-founder of OpenAI, briefly touched on the rules of immediate engineering. The slide deck for his presentation might be discovered at https://karpathy.ai/stateofgpt.pdf.
Within his presentation, two major targets have been highlighted:
To obtain the highest attainable efficiency, significantly in one-shot immediate debugging,I’ve additional divided it into two particular features, Correctness and Accuracy. Therefore, to make sure the reliability of immediate engineering, it’s crucial to uphold a set of guiding rules. These rules embrace:
1. Correctness: This entails verifying if the generated completions align with our expectations. For occasion, if we count on the output in JSON format, we should be certain that fashions like LLM can comprehend and produce outputs within the anticipated JSON format.
2. Accuracy: After reaching elevated creativity in generated completions, often by utilizing a temperature better than zero, it’s important to confirm if the completions proceed to stick to the immediate directions. Every completion ought to precisely produce the supposed outcome.
3. Cost: While sustaining correctness and accuracy, it’s equally necessary to manage the token rely. Prompt engineering ought to try to cut back token consumption successfully, bettering effectivity and cost-effectiveness.
Prompter can effectively help you in bettering the standard of your prompts from these three features. Feel free to offer it a attempt.
