Guided vs. Unguided Refinement Prompts
A distinction can be made between two types of prompts for refining LLM outputs. Unguided prompts, such as the simple command "Please refine it!", depend entirely on the model's intrinsic instruction-following capabilities without specific direction. In contrast, guided prompts provide targeted feedback, like "Please correct all the grammatical errors," which directs the model to focus on particular aspects, leading to a more effective refinement process.
0
1
Tags
Ch.3 Prompting - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Related
Example Prompt for Refining a Chinese-to-English Translation
Guided vs. Unguided Refinement Prompts
Highlighting Modifications in Iterative Translation
A team is building a machine translation system. They find that a single translation request sometimes produces awkward or slightly inaccurate phrasing. They are considering implementing a multi-step process where the model is first asked to translate a text, and then, in a subsequent step, is asked to refine its own translation. What is the most likely trade-off the team will face by adopting this multi-step refinement approach?
Improving E-commerce Translation Quality
A developer is using a language model to translate a document from Spanish to English. To improve the quality, they decide to use a two-step refinement process. Arrange the following actions into the correct chronological sequence.
Learn After
Example of an Unguided Refinement Prompt
Example of a Guided Refinement Prompt for Grammatical Correction
A user is working with a language model to improve a short story it generated. The user wants to make the main character more compelling. Consider two possible follow-up instructions they could give the model:
- "Make it more interesting."
- "Revise the story to give the main character a clearer motivation for their actions and add more descriptive details about their internal thoughts."
Which instruction is more likely to lead to a successful revision, and why?
Improving an Idiomatic Translation
Transforming an Unguided Prompt