At the Google I/O that was held last month, they handed out a guide to the audience telling them how to use Google Gemini in the right way, and how to write an ideal prompt. Those tips are not just for Google’s own LLM though and they can be applied to any LLM out there. Although they were designed for Gemini, almost all of it is still relevant to Google’s number one competitor; ChatGPT. For those wondering if these Gemini tips will apply to OpenAI: Yes, they will if you want to increase your efficiency with ChatGPT and the new GPT-4o model. The way to enhance ChatGPT experience with Gemini
Mistakes to be avoided with Gemini and ChatGPT
The biggest mistake you can make with Gemini and, therefore, with ChatGPT in particular is to overdo it with the length of prompts. Of course, being specific about what is necessary is one thing, but creating an overly elaborate call that takes multiple paragraphs to explain can lead to hallucinations. LLMs do not strictly adhere to rules and are inventive in their answers; any restriction of such inventiveness results in weird or erroneous out comes.
In that manner, you should also avoid the use of general statements. When asking for something, be specific about the same while also avoiding complex sentence structures that can be given different meanings.
The correct way of giving prompts to the ChatGPT
You also should not presuppose; if you’re posing a question for which some background information is needed, be sure to provide it to the language model in case it does not know it. For instance, if you wanted to make a comparison of two processors, you wouldn’t use ‘the latest AMD processor’ but rather ‘The AMD Ryzen 9 9950X’.
Last but not least, ensure that you can distinguish between tasks. Do not confuse several activities at once. Divide it into steps 1, step 2, step 3 if it is possible, do not try to mix some of those steps. Your instructions must not be the kind that would be difficult to explain to a real person if you want them to be easily understood by an LLM.
This information also applies to ChatGPT very much and some principles can be followed by you.
- Persona: State what position the LLM should take.
- Task: Express clearly and specifically what you want the LLM to achieve.
- Context: Elaborate on the task and the intended result.
- Format: There should be a clear output format like a paragraph or a set of bullet points.
How to make use of LLM?
In this situation, you should specify what kind of role you expect the LLM to play. This helps it identify the “category” in its database within which the operation is going to take place. For example, let’s attempt to develop a plan for organizing a LAN tournament based on Counter-Strike 2. This is a clear indication that we have to concern ourselves with the network conditions, where this is taking place there has to be a provision for this and there has to be specific answers on how to implement this plan.
This is good advice for anybody using an LLM and applies to ChatGPT as well. Software is a tool, and it becomes your duty you manage these tools in such a way that they make your work easier, not harder. Google has provided LLM advice in Gemini for the Workspace Handbook of the company and you should read it.