tcworld conference 2024 – Generative AI and prompting

Chatbots, generative AI, and large language models were still everywhere at this year’s tcworld conference. Alex Jitianu from our partner Syncro Soft talked about GenAI and the evolving domain of prompting.

He pointed out that while answers might be linguistically perfect, they can still be factually wrong. Anyone playing around with ChatGPT or Copilot has experienced this phenomenon which is known as hallucination.

Influencing the way answers are generated can mitigate some of the problems. But how do you influence a probabilistic text generation algorithm? The answer is prompt engineering. Prompt engineering adds context to a prompt. The prompt is no longer just a question but contains instructions for the AI. The context can be a persona, a concise objective, an expected style, and information about response length and formatting. All of these additions improve the answer. Breaking up prompts into smaller tasks can also be beneficial. You’ll end up with more prompts but there are frameworks like SWARM that help orchestrate multiple GenAI agents and chain prompts.

Building better prompts: frameworks, tools, and techniques for GenAI success

If you invest time in developing your prompts, building up a prompt library, you’ll probably need a test plan to compare prompts and results. Some companies provide test suites. Anthropic has their Evaluation tool as part of their Anthropic Console. OpenAI also has an evaluation tool. The more prompts you have the more effort it takes to organize and maintain them. Did you know that you can modularize your prompts and reuse them? You can even use variables in your prompts. Sound familiar? If you’re a tech writer you might already have experience modularizing and reusing chunks of content.

However, even with the most refined prompts, the answers will never be 100 % correct all the time. That's just the nature of this technology. And with that final thought Alex Jitianu pointed out that GenAI is no silver bullet. In-depth knowledge of the project’s domain is essential. As a tech writer you’re the content expert. You’re the one best suited to drive GenAI.

If you want to learn more about prompt engineering, get in touch.