Thoughtful search strategies and approaches

- We talked about the importance of developing good prompt engineering skills. Here, I'd like to expand on that plus offer some additional tips for thoughtful reasoning engine search strategies. First, be specific. The more specific the prompt, the more customized and nuanced the generated results will be. Provide context. Reasoning engines do well when you include context within your query which often includes providing an example of the type of answer that you're looking for. - Rather than having to infuse it with knowledge, it already knows what it knows. Instead, we need to give it examples of what right looks like and how we can answer the question that we've asked in the very specific way that we want it answered. And your ability to provide that instruction happens in the prompt as well. - Break things down. If you've got a complex query, try breaking it down into smaller parts. This can make it easier to get a thorough non-confusing answer. Use clear language. Reasoning engines are designed to synthesize human speech so it tends to do a fine job understanding colloquial language, jargon or slang. But if you want the best results, write your prompt as clearly as possible with proper grammar. One thing though, if you're creating a prompt for a specific industry or use case, using very clear examples of industry language is the way to go. - One of the most interesting things about this role is it's very domain specific. Someone who's building the prompt has to know the problem they're trying to solve. It's less about the technology and more about the business problem or the challenge you have. I want to be an animator, so I'm like, I can build animation. But in order for me to get like a GPT model, for example to do an an animation, I actually have to know all the lingo and you know all the jargon of animators. In order to do that work, I need to be an animator to use that tool. - Experiment. Try out all sorts of ways to write your prompts. Often slightly changing your query leads to different responses and different levels of creativity. And the neat thing is this type of experimentation isn't just making you learn it's also making the reasoning engine learn. - It's building over time and learning from the inputs. And it's more of a sort of a dance between human and machine because it's always, you're iterating the prompt. You're trying to push it into these new directions to be able to be more creative and in alignment with what you envision. And I think over time, as it continues to learn from this it might actually get more creative itself. - Every time we use the model, the outcome changes the ground truth shifts even just a little bit. So we're going to have to make sure that we're continuously training and monitoring the output of that model over time. - Along those lines, here's a list of some creative ideas for how you can expand your prompt engineering skills. Role play scenarios. Frame your query as a role play scenario where the reasoning engine takes on the persona of someone relevant to your query. For example, imagine you're the product manager for a brand new smartphone company. What are 10 potential innovative features that could be added within the next five years? Analogies. In your prompt, explain concepts using an analogy to better understand complex ideas. For example explain quantum mechanics using a sports analogy. Debate style questions. Ask for arguments for and against a particular topic to get a more comprehensive understanding of different perspectives. For example, present arguments for and against the implementation of universal basic income. Creative exercises. Ask the model to brainstorm, ideate or write fictional scenarios related to your query. For example, I'm writing a story about traveling to a parallel universe. Help me brainstorm some unique laws of nature that this parallel universe might have. And finally, during all of the excitement of finding new ways to generate content, don't be afraid to say that you've used the tool. - I think there's also a level of transparency another responsible AI principle that should encourage us as organizations, as users, to not be shy about saying when we've used GPT, many times I find people are almost like I used GPT for that. Almost like it's a secret. And I think part of it is normalizing the use of it so that you're transparent about what's being generated and what isn't. - There's been a lot of work especially in the academic setting right now about appropriate guardrails for students in using this. And I think that that also needs to be applied in corporate settings. If you are going to be using these tools to aid in your creative process or in writing or creating imagery, it must be transparent that you did use them. - Now go forth and engage in those thoughtful search strategies. The possibilities are limitless.