[ad_1]
OpenAI just released an official prompt engineering guide to help its 180 million users get better results from the platform. The guide shares strategies and tactics that can be combined for greater effect. OpenAI encourages experimentation with its large language model ChatGPT, so you can save hours each week and supercharge your business like never before.
Following ChatGPT’s guide to perfect prompting, including the 9 components of an effective prompt and avoiding the 4 most common pitfalls people make with ChatGPT, you’ll be well on the way to excellent results for your business. Open up a chat window, test some prompts, and reprompt until you have what you need. Prompt like a pro and never look back.
ChatGPT prompts made better: the official guide from OpenAI
Write clear instructions
“These models can’t read your mind,” said OpenAI, even if it seems like they can. “If outputs are too long, ask for brief replies. If outputs are too simple, ask for expert-level writing. If you dislike the format, demonstrate the format you’d like to see.” Keep it super simple and instruct as if you were explaining a task to a junior member of the team. Assume nothing. “The less the model has to guess at what you want, the more likely you’ll get it.”
Examples of how to do this include asking the model to adopt a persona, for example “act as a business coach”, using line breaks or extra formatting to clearly indicate distinct parts, specifying the steps required to complete a text, providing examples, and specifying the desired length of the output. Tell the model exactly what you want in as much detail as possible to set it up for success when prompting.
Provide reference text
If you’ve used ChatGPT for any length of time you will know that, “language models can confidently invent fake answers, especially when asked about esoteric topics or for citations and URLs,” as OpenAI admitted. ChatGPT doesn’t want to disappoint, so will just fabricate a response without a sufficient answer, much like a student might blag an exam. OpenAI explained, “In the same way that a sheet of notes can help a student do better on a test, providing reference text to these models can help in answering with fewer fabrications.”
In practice this means instructing the model to answer using a reference text, making absolutely sure it uses the text’s content in the response. A step further, instruct the model to answer with citations from the reference text. Keep the responses on track by controlling the inputs yourself, and stand over ChatGPT like a doting mentor, making sure it doesn’t stray off course.
Split complex tasks
“Just as it is good practice in software engineering to decompose a complex system into a set of modular components, the same is true of tasks submitted to a language model,” said OpenAI’s team within the guide. “Complex tasks tend to have higher error rates than simpler tasks.” It makes sense. When doing any task, from writing a business plan to compiling your website homepage or figuring out how to sell your business, you don’t execute everything at once.
Breaking tasks down into simpler subtasks is the goal here. “Furthermore,” continued OpenAI, “complex tasks can often be re-defined as a workflow of simpler tasks in which the outputs of earlier tasks are used to construct the inputs to later tasks.” Ask something, then use a key part of the response to ask the next thing. Don’t lose context in complicated prompts, don’t leave ChatGPT to decide which parts of your request are important. Here’s where you use intent classification, which basically means specifying the priority of requests, or summarize previous dialogue to remind it of context, if you’ve been chatting for a while.
Give the model time
If I asked you a tricky question and demanded you answer immediately, you wouldn’t be best pleased. ChatGPT is the same. It needs time to “think”. “If asked to multiply 17 by 28, you might not know it instantly, but can still work it out with time. Similarly, models make more reasoning errors when trying to answer right away, rather than taking time to work out an answer.” So how do you work out answers in collaboration with ChatGPT? “Asking for a ‘chain of thought’ before an answer can help the model reason its way toward correct answers more reliably,” explained OpenAI. Ask it to show its working.
You can, “instruct the model to work out its own solution before rushing to a conclusion,” allowing it to take its time. You can also ask if it missed anything on previous passes, to ensure it’s not leaving out vital details. ChatGPT is not perfect, neither does it claim to be. Figure answers out collaboratively and keep your questioning skills sharp with every output generated.
Use external tools
“Compensate for the weaknesses of the model by feeding it the outputs of other tools,” said OpenAI. But what does this mean for entrepreneurs? Ignoring the technical recommendation of a text retrieval system, you can “tell the model about relevant documents,” said OpenAI. And you definitely have those. “A code execution engine like OpenAI’s Code Interpreter can help the model do math and run code,” which means you can upload spreadsheets or documents and ask for information based on the stats and numbers.
This behaviour is encouraged. A large language model is simply predicting the next word in a sentence, it’s not a genius business consultant, analyst or therapist. “If a task can be done more reliably or efficiently by a tool rather than by a language model, offload it to get the best of both,” OpenAI advised.
Test changes systematically
As with everything in business, “improving performance is easier if you can measure it.” First measured, then managed. “In some cases a modification to a prompt will achieve better performance on a few isolated examples but lead to worse overall performance on a more representative set of examples,” they added. So you can’t just change everything at once, because you likely won’t know what worked.
To be sure that a change is net positive to performance it may be necessary to define a “comprehensive test suite,” said OpenAI, also known as an “eval”, which simply means making incremental changes to ensure the results mean what you think they do. If you’re making a website section more persuasive and more funny, specify one change first and see how it performs, as well as trying both at once. If you can combine this with actual results, for example A/B tests and actual conversion data, you will understand which prompts and output leads to better results.
Get better results from ChatGPT by following OpenAI’s strategies
If you’re going to use ChatGPT as your entrepreneurs’ assistant, you might as well do it right. Sending rubbish prompts leads to rubbish results that your audience knows were generated by ChatGPT. It leads to content that an intern could have written better, that doesn’t match the quality of your brand. And no one wants that. Use this official guide from OpenAI for perfect prompts every time.
Follow me on Twitter or LinkedIn. Check out my website or some of my other work here.
[ad_2]
Source link