When it comes to Generative AI, it is not what you ask; it’s how you do it. The importance of correct prompting regarding Chat GPT and other Generative AI models can’t be overlooked. There are dozens of resources on how to prompt AI, but only some are based on actual scientific research methods. In this blog post, we will explore the evolution of prompting and discuss the latest developments in AI prompting that you can use to achieve better results for your business.
Before we jump into the discussion about the latest research on the most effective prompt method, let’s quickly overview the different types of prompting from the least effective to the most effective:
Input – Output
It’s straightforward: give a problem to the AI and get a direct answer without following with additional questions or other prompts.
Chain of Thought
In this prompting method, instruct AI with steps and examples that will help AI solve your problem.
Self-Consistency with Chain of Thought
This prompting approach involves feeding AI the same prompt each time, creating a new chat, and then identifying the answer that comes up the most often.
Do you recognize your prompting method from the list above? If yes, you are not getting the most out of Chat GPT.
It is essential to note that the research states that this advanced prompting method can be less effective when it comes to the tasks that Chat GPT already does well on.
The Most Advanced Prompting Method: Tree of Thoughts
The most effective AI prompting strategy is based on the theory of human problem-solving thinking called dual process theory. According to this theory, when we face a problem, there are two ways in which we approach it:
- A quick and intuitive way;
- A slow and thoughtful way.
Step-By-Step Tree of Thoughts Prompt
- Give AI a problem with an example of how it can be successfully solved and prompt the AI to develop multiple solutions to your problem. It will ‘quickly’ generate various answers, some potentially faulty.
- Now, make Chat GPT think slowly and analyze. Create a separate chat and prompt AI to evaluate the answers and pick the most appropriate ones to meet your goal.
The blog post describes a new generative AI prompting method called Tree of Thoughts. Researchers from Princeton University and Google DeepMind developed it to improve Chat GPT responses. This approach uses a branching reasoning process, evaluating different steps like human decision-making, and is compared to three other prompting methods.
Check out the full blog post here, or explore the original scientific paper in detail by clicking here.
Our Point of View on ChatGPT Prompt Development
Prompt development approaches continue to evolve as AI’s ability to solve more complex problems grows. What we’re seeing is a natural progression.
“The prompt programming approaches continue to move down the path toward more and more complex interactions with AI. When ChatGPT 3.5 hit the market, everyone assumed that AI was some sort of god-like construct that could answer anything and everything completely, accurately, and succinctly. The reality is that, at this stage, we are not dealing with god-like intelligence.”
“My belief is that we need to treat AI today in a way similar to how you treat a gifted 5-year-old. It’s more than simply asking questions and demanding answers. It’s about providing context. It’s about guiding and developing your AI to look at the world more broadly and infer from it much like humans do,”
commented Michael Marchese, Founder and CEO of Tempesta Media.
