One of the most profitable corporate skills these days is creating effective prompts. It's about making a chatbot request and getting what you need, but that's not as easy as it looks.
A chatbot is like a child. They do just what you ask, if even more. To get what you want, you must be explicit and concrete.
AI startup humanity offers some help in this aspect. The company recently published a “Speed Engineering Overview” to help users get started.
That guide applies to almost any chatbot, but it's tailored to your own Claude.
According to humanity, the first order of business is to understand exactly what Claude is. “When interacting with Claude, think of it as a great but very new employee (amnesia) who needs explicit instructions,” the company says in its guide.
Second, you have a rough idea or draft of your questions and have a sense of what a successful outcome will look like. Humanity also provides the “prompt generator” for the first draft.
Then it's all about refinement of that first prompt. Here is the best tip for humanity.
Be specific about your prompt
“Claude has no context as to your norms, styles, guidelines, or preferred ways of working. The more accurate you describe what you want, the better Claude's response will be,” Anthropic says.
The company suggests telling the chatbot what the results will be used for and what that means. You also need to tell Claude, or the chatbot you are using, what the final goal of the task is.
The more you organize the direction, the better. Humanity even recommends laying out requests as bullet points or numbered lists.
Be generous with examples
“The example is a shortcut to a secret weapon to generate exactly what Claude needs,” says humanity. “By providing some well-made examples at the prompt, we can dramatically improve the accuracy, consistency and quality of Claude's output.”
This strategy is sometimes called a multi-shot prompt. Humanity says that by giving examples, it reduces misunderstanding and implements uniform structure and style.
Related Stories
Business Insider tells the innovative stories you want to know
Business Insider tells the innovative stories you want to know
Give us space for a chatbot to think
“Giving the space to think about Claude space can dramatically improve that performance,” Anthropic says. “This technique is known as the Chain of Thought (COT) prompt, which encourages Claude to break down the problem step by step, leading to more accurate and subtle output.”
This means that users can lay out a series of steps and make the most of their chatbots and think about each one before answering. “This thorough reasoning leads to more confidence and legitimate recommendations,” says humanity.
role play
Humanity says that one of the most effective strategies is to assign a chatbot to specific roles, such as “news editors” and “financial planners.”
“This technique, called a roll prompt, is the most powerful way to use system prompts in Claude,” the company says.
“In complex scenarios such as legal analysis and financial modeling, role prompts can significantly improve Claude's performance.”
Assigning roles will help you get exactly what you need. Maybe you want the brevity of a news writer, or maybe you want the tone of an academic.
Reduce hallucinations
Chatbots make up things, so you need to check everything they say. However, there are some simple ways to reduce those hallucinations.
Anthtropic says the best thing you can do is give the chatbot permission to say “I don't know.”
“We explicitly grant Claude permission to acknowledge uncertainty. This simple technique can significantly reduce misinformation,” says humanity.
You can also ask Claude and other chatbots to cite claims with sources. “We can also verify each claim by finding a support estimate after a response is generated for Claude. If we can't find a quote, we must withdraw the claim,” says humanity.
