LLMs (Large Language Models) like Claude 3 and ChatGPT 4 thrive off of specificity. If you want a very helpful answer to whatever it is you are looking for, you’ll want to include as many relevant details as possible.
Doing so might feel as though you are oversharing to a stranger. But think of it this way: imagine you were paying an expert to help you with whatever issue you were having, and they ask you to tell them as much as possible about the situation. I’m guessing you probably not would respond the following way:
“Family of four, Mexico trip, ideas.”
That would make for a pretty solid Google query; but LLMs are, in a sense, the opposite of Google. For them to truly shine, you will need to turn off your “Google brain” and start treating them like the human expert who asks you to tell them about your specific situation and then sits back and listens as you give all the deets. For instance, assuming they are a travel agent:
“We are a family of four. It’s my 8-year-old son and 12-year-old daughter who both like doing outdoor activities. My son, especially, likes wandering around town and exploring all the nooks and crannies (with us by his side). My daughter likes souvenir hunting and a really good meal (I think we spoiled her early!) She loves taking cooking lessons. In general, neither are surprisingly that big of beach people. My wife and I are the fun adventurous types who like immersing ourselves in the local culture. We are big foodies and really extroverted. We are thinking of spending a few weeks in Mexico this summer, possibly renting an Airbnb for an entire month. We hope the entire family can take Spanish lessons to get the most of the experience. Could you suggest several options for us!”
Real ChatGPT 4 outputs
Here are the two ChatGPT 4 outputs based on the queries/prompts above: just to show you how the different the two results can be:
Prompt 1: Family of four, Mexico trip, ideas
Now, you might be thinking, those options look great! But remember, we are looking at a specific family with specific tastes in what they are looking for. For our hypothetical family (granted, it’s loosely based on my own), this list isn’t too helpful. For one, we know the kids don’t care too much for beaches, yet the list starts off with beaches, and they are sprinkled throughout.
Additionally, there isn’t anything specific to this family; this list could be for all the families interested in traveling to Mexico. What makes ChatGPT and LLMs unique is the output can be tailored exactly to your situation — the output is just for you!
So, what does that look like? Let’s take a look at the next output.
Prompt 2:
“We are a family of four. It’s my 8-year old son and 12-year old daughter who both like doing outdoor activities. My son, especially, likes wandering around town and exploring all the nooks and crannies (with us by his side). My daughter likes souvenir hunting and a really good meal (I think we spoiled her early!) She loves taking cooking lessons. In general, neither are surprisingly that big of beach people. My wife and I are the fun adventurous types who like immersing ourselves in the local culture. We are big foodies and really extroverted. We are thinking of spending a few weeks in Mexico this summer, possibly renting an Airbnb for an entire month. We hope the entire family can take Spanish lessons to get the most of the experience. Could you suggest several options for us!”
Notice how every answer is tailored to what I included in the prompt. Almost every recommendation includes something about my children. For instance, take a look at Mexico City, and you get, “Your son can enjoy exploring Chapultepec Park”, “…perfect for your daughter’s souvenir hunting.”
And while there are some similar contenders (Mexico City and Oaxaca both make the list), the beach-centric cities are absent. What we do have is a re-framing of the recommendation: each specific ‘Activities’, ‘Food Experiences’, and ‘Spanish Lesson’, three important components I mentioned in the prompt specific to my family.
More--and more targeted--recommendations
There’s also two additional cities not mentioned in the previous list, opening up new places to explore, places right up our cobblestoned-alley (see San Miguel).
By thinking about ChatGPT as an expert being and not a Google search box, you, too, can quickly find your perfect vacation.
FWIW, when I entered in “family of four, Mexico trip, ideas” into Google, I got offers for resorts or lots of beach lists. (Not surprisingly, when I entered the long prompt into google, I got posts about how to raise happy kids.)
Essentially, ChatGPT can do something Google simply can’t do: offer you a bespoke travel itinerary, the way a professional travel agent could. That doesn’t mean you shouldn’t google some of the places above, especially if you want to have an idea what they look like (what exactly does ‘San Miguel de Allende’ look like, for instance?)
San Miguel de Allende, courtesy of Google.
Other Use Cases
This, of course, is just one use case, though one many can relate to. But think of something that is super specific to you and the world you inhabit. Provide all the details as though you are discussing this with an expert human, and watch the magic happen.
Some interesting use cases I’ve seen with those I’ve worked with:
A father helps improve his son’s batting ability by creating a prompt describing exactly where his son struggles with his form. What recommendations have and haven’t worked. The output pinpointed possible areas his son could focus on, as well as creative drills to help solidify the new movements, ideas not even his son’s batting coach had come up with.
A friend who owns half a rental unit in California was hoping to sell it, but the neighboring unit continually parked an old large truck on the main driveway, leaking oil everywhere. My friend wondered what rights she had as a joint owner of the condo, so that at the very least the truck wasn’t parked there while she was showing the unit. The output laid out her basic rights, suggesting what she could do and the road to legal recourse, should that be necessary.
In both cases, specific and detailed prompting was used, getting them on the right path. And the prompting was nothing more than them writing out what they would like say were they talking to an experienced batting coach or someone with a deep knowledge of the ins and outs of property law in the state of California
Takeaway
To get the most out of ChatGPT and the other LLMs, you will need to turn off your “Google brain” and turn on your “prompting brain.” Whether it’s your next vacation or a very specific situation you happen to have, doing so can make all the difference!
Comments