Want to push the boundaries of prompt engineering?
Keep the fundamentals and forget the absolute "rules".
Follow this and you will be amazed at what your brilliant brain will come up with. Run with your ideas, don't walk slowly.
---
Always follow proper ethics when interacting with these systems, but try something "new". Ask a ridiculous question. You might be surprised with the output. Experimentation is the entire learning process, in my 4 years of work with prompt engineering.
---
One example: During the Lean Solutions Granite Hackathon, uncertainty on whether our concept fit well with the established judging criteria and boundary of the rules, creating a temporary bottleneck.
To speed this bottleneck up, I used this prompt process:
Prompt Solution:
-
Upload: contest rules, judging metrics, and project outline to Granite.
-
Prompt Granite similarly to: "You are an experienced IBM judge who has participated in many IBM hackathons. Analyze the uploaded {rules.pdf}, {judging.pdf) and compare alignment with {project_concept.pdf}. Score {project_concept} on a scale from 1-100. If {project_concept.pdf} scores less than 99/100, revise until {project_concept.pdf} scores 100/100 on alignment with {rules.pdf} and {judging.pdf}."
This prompt is odd, yes. Did it work when I needed it to? Yes. A dash of "what if?" Is all that's needed.
---
Failure isn't Failure. It's learning and optimization. Reframe poor outputs as learning opportunities and document what works for you and what doesn't.
Most importantly: If you are passionate about prompt engineering, never give up. The only Failure us giving up. The rest is learning.
#LLM------------------------------
Julian Gonzalez
Generative AI Engineer
CreativeAct Technologies
Orlando FL
------------------------------