English isn’t an LLMs first language
Choose a Format Structure
When writing your prompt, you can structure your text in different ways to help the LLM interpret it. We strongly recommend choosing to use either Markdown or XML structure. Markdown is a lightweight markup language that you can use to add formatting elements to plaintext text documents. You can learn how to use Markdown here. XML is a different, more complex markup language designed to store and transport data in a structured format, commonly used by engineers. You learn how to use XML for prompting here. We recommend Markdown for simplicity.(1) Assign a Identity
Clearly define the AI’s role and the target channel identity before prompting, providing context and mentioning tone- Introduction (with Markdown)
- Introduction (with XML)
- Introduction (Plain Text)
(2) Add Context
Including information that is constant throughout the show will help the agent decipher good vs. bad content. Here you can add information about the hosts, topics that will always do well, or clips to stay aware from.- Context (with Markdown)
- Context (with XML)
- Context (Plain Text)
(3) Clearly Define the Goal
Explicitly declare the agent’s goal and what success looks like- Goal (with Markdown)
- Goal (with XML)
- Goal (Plain Text)
(4) Provide Rules
Constrain the agent’s behavior by providing comprehensive rules. Focus on high-level behavior which will elicit proper behavior downstream. For the purpose of finding clips, provide rules that clearly outline what makes a good clip vs. a bad clip.- Rules (with Markdown)
- Rules (with XML)
- Rules (Plain Text)
(5) Finish with a Reminder
Like humans, LLMs tend to have a slight preference towards what comes at the beginning and end of their prompt. So, just as you start with the goal, you should end with it too.- Conclusion (with Markdown)
- Conclusion (with XML)
- Conclusion (Plain Text)
Putting it all Together
Combine all of the above sections for a singular prompt- Full Prompt (with Markdown)
- Full Prompt (with XML)
- Full Prompt (Plain Text)
Frequently asked questions
Why should I use Markdown or XML?
Why should I use Markdown or XML?
Many of the popular LLMs were trained using structured markup languages such as Markdown, XML, or in a JSON structure. Using this same format helps the AI decipher meaning and focus on particular sections.
Do I include editing instructions in the prompt?
Do I include editing instructions in the prompt?
No. The prompt in the Find Clips node is purely for finding the best moments. You should only include information about how to clip the content. You’ll assign your own editing style via the Workflows
Does including example clips help?
Does including example clips help?
Yes!! The more example clips, the better. If you are going to include examples, however, be sure to include as much information about it as possible. This may be the title, the description, the transcript, etc.