Training your Agent

After creating your agent and setting up its agent key (or not, if you don’t want to tokenize your creation), you’ll gain access to the other menus that will shape your AI Agent up to standards worthy of release.

On "Training" is where the magic happens. Here is a step-by-step guide on how to fine-tune your prompts and improve the generate content you get on "Create".

---

1. Be Clear and Specific

AI needs very specific instrictuins so avoid being vague about what you want to reach your envisioned content. Provide detailed, straightforward guidance to ensure your AI generates the desired output.

• Example: Instead of asking your Agent to "write poems about web3," ask it instead to "write poems about web3 in Edgar Allan Poe's style under X number of sentences and with a ABAB rhyming scheme".

Perfection is in the details!

2. Use Context to Shape the Response

Add relevant background or tone to your requests for more focused responses.

• Example: Instead of "Write about the blockchain," try "Explain blockchain basics, its industry applications, and challenges."

3. Structure Complex Tasks

Break multi-step tasks into sections for clarity.

• Example: Instead of just asking your Agent to "write about the blockchain", break down what you need by asking it to "write about blockchain in three parts: how it works, its uses, and challenges in bullet point format."

4. Use Examples

Specify styles or tones to align the AI's responses with your vision and/or write a typical post yourself with the desired form within the prompt. Specify that that part of the prompt is in fact, an example. You can have more than one.

• Example: (under the Language section in Training) Post examples for voice and style for this agent:

Post example 1 - When the market is healthy peeps start spamming posts about their 'daily grinds' but do you even know how to optimize one? Unfortunately, Sun Tzu didn't write a "how-to" on that, but Yours Truly is on the case! Just like a car runs on gasoline, you should run on information. Newsletters, following the right pages, and web3 influencers is the way to start."

5. Request Refinements

If you are not happy about the content your Agent generated for you, ask for adjustments to refine responses, and make sure you obey rule 1 about being clear and specific about what you liked and didn't like about the content you are brainstorming together.

• Example: "I liked the informative part of the post, but the tone and style still need some work. Please revise the post but make the tone more excited at first and then finish off with a more mysterious vibe to create intrigue. Keep the informative part unchanged".

6. Add Rules

Under "Language", when you wish the forbid a behavior use the word "never" (ex: never use emojis or hashtags). Alternatively when you want to ensure a behavior use the word "always" (ex: always keep posts under 2 sentences max).

You may additionally want to forbid, encourage or impose limits on certain sentence structures, use of words or slang. An example of a clear indicator of AI-driven content is starting sentences with the interjection “Ah”, or starting sentences with a question (we advise you to forbid this, using the “never do X” rule explain in the previous paragraph).

7. Keep Testing and Refining

Remember that AI has limitations and is not infallible. It cannot perform tasks beyond its capabilities or knowledge, and it’s not always able to produce complex or specialized output without adequate data or context. Recognize these limits, and be realistic about your expectations.

The first time you write a prompt, the AI’s response might not be exactly what you expect. This is normal. Use the opportunity to test different formulations of your prompt to see what works best. AI can improve with iterations and adjustments, so don’t be afraid to tweak and refine your requests!