Pickaxe not following prompt

Bugs / Site Issues
1

Hi there,

My pickaxes have recently stopped following the prompts as they used to.

I’ve not changed how I go about creating the prompts, but a couple have recently been behaving erratically.

here’s an example of what I mean:

2

Have you tried adding your copyrighting brief to a PDF and uploading it as a knowledge base as opposed to a google docs link.

3

Try troubleshooting your main prompt with ChatGPT 4o. Checkout this example. You can always take screenshots and attach them to the convo with ChatGPT for added context.

image
image875×555 36.9 KB

If that doesn’t work, try adding a Code Interpreter action, and use ChatGPT to help you generate a trigger prompt. The job of the Code Interpreter action would be to figure out the IF-Statement computation in the background.

How to video:

Example of an IF-statement in a Core Prompt:

Here’s how it might look conceptually in a core prompt:

IF the user question includes "how to," provide a step-by-step guide.
IF the user asks about pricing, explain the tiers clearly and suggest the most relevant option.
IF no specific query is provided, ask the user clarifying questions to better understand their needs.

Notes & considerations:

  1. Using ChatGPT 4o OR 4o-mini for the LLM model. Either should be enough.

  2. You mentioned you tried pushing the Token Lengths to the limit. Although I don’t believe that’s the culprit, here is some useful info:
    those should be maintained within controlled ranges to control the input and outputs, the intended duration of each chat session, and the overall LLM API cost on your business.

Let me know!

~Ned

4

The issue is that it’s not following the chat in the order it should

5

Have you tried telling it specifically to answer in order ?

6

I went and tested your Pickaxe a couple times but did not run into this issue. All your settings look good to me. If you’re still experiencing this issue, I would just revisit the prompt to see if you can clarify anything / trim some fat. For example, listing the conversations steps twice might be helpful (once at high level, and once with details for each step). That will make it more important in the LLM’s mind.

7

OK thank you, I’ll try that. I’ve noticed that sometimes they get a bit ‘stuck’ when I make changes (don’t work as expected), but are fine the following day. Is there anything to do to make things stick more quickly? I’ve tried clearing my cache, which sometimes seems to help.

8

Was your issue resolved?

I just came back to Pickaxe to build my first real bot and my bot also fails to follow instruction.

Works perfect when posted straight to the AI model but fails inside the bot.