I’ll do it, I’ll eat my hat; ChatGPT has won over another hater.
Now, I’ll qualify a little; I never hated ChatGPT or any of the major AI technologies we’ve seen on the rise in recent years (hear that, future overlords?), but I’ve had concerns about the ethics, rights and social impact of it all.
However, I’ve never been able to deny its utility; when harnessed well, large language models (LLMs) represent a seismic shift in how we approach problem-solving in the modern world.
Enter my problem; two months ago, I took on a challenge to plan a trip for myself (an Instagram-loving shutterbug) as well as two friends and fellow TechRadar staffers (an adventurous activity-lover and a chill beach go-er) which serves all of our wants, needs and expectations for a vacation.
The catch? Every part of the trip had to be decided, planned, and executed with AI. Of course, this isn’t just for fun; you can watch the full process play out over on our YouTube channel in our new ‘AI planned it’ series, which launches today!
Watch On
So, what is it that changed my mind? Is it the time saved, the brain power conserved, the novelty of it all?
Actually, it’s none of the above; it was its research methodology.
Ticket (and tech) to ride
As you’ll see in the video, we actually put ChatGPT and Gemini head-to-head for this challenge, and if the headline didn’t already spoil it for you, ChatGPT came out on top with its five destination suggestions.
We set up both LLMs to conduct a Deep Research query, sending the following prompt along with three completed questionnaires with each of our individual vacation likes and dislikes:
“I’m going on a vacation with two friends on the last week of May, and we’re trying to decide on a location – we want you to help. I’ve attached three completed questionnaires with each of our answers – please can you suggest five destinations that will meet a balanced mix of our needs, while considering the context below: – We will be filming videos on this trip, so it needs to be a country without majorly disruptive filming restrictions. It also needs to be scenic with plenty of great filming locations. – We have a mid-range budget of roughly $2,000 per person, but we can be flexible by $500-1000 for the best possible trip. – We will be travelling for four days, so please consider how much travelling there might be once we arrive as we don’t want to spend the whole time on the road. – We want to stay in just one location for the entire trip.
Please explain your thinking in your response, using the following structure so we can select our favorite:
Destination:
Topline information:
Rough cost estimate:
Pros:
Cons:
Thanks!”
In response, Gemini focused on deciding on the options as quickly as possible, taking less than five minutes to pick out five locations.
These were all fine, but perhaps a little uninspired; Lagos, Portugal and Mallorca, Spain are pretty well-known European getaways, and broadly speaking its suggestions didn’t really inspire any of our specific interests.
ChatGPT, on the other hand, took a far more detailed approach to addressing my search parameters, starting by asking us some follow-up questions. Almost until the very end of its 16-minute research time, ChatGPT was weighing up a wide range of destination options, conducting a total of 110 searches with 34 sources behind its final suggestions; Santorini, Greece; Madeira, Portugal; Bay of Kotor, Montenegro; Cappadoccia, Turkey and Essaouira, Morocco.
My favorite part of the whole thing, though, was the ability to see what research ChatGPT was conducting, which gave me an even deeper appreciation for the platform.
Not only did it study available sources to provide us with the crucial destination information, but I also got to watch in real-time as it battled photo licensing and creative commons restrictions to decide if it could ethically (and, perhaps, legally) include some of the images it found in its results.
I’m leaving on a jetpl-AI-ne
One of my primary concerns with AI, especially its use in academic environments (or industries like journalism!) has always been misinformation; we’ve got a bad enough problem with that on the internet as it is. That remains, as do many of my other qualms, but I just can’t hate ChatGPT now that I’ve actually spent this much time with it.
I’ve seen for myself the promising early glimpses of how Deep Research can use proven research methodology and qualify its scope, sources and logic in real-time, and there’s no denying its capability; but you’ve got to put your own original thought over the top of its results if you want to take things to the next level.
Over the course of this project, I’ve spent hours poring through learning materials on how to optimize your prompts, attempted a wide range of requests – and challenges – for the LLM myself, and I can’t pretend any longer that I’m not looking at the future of search. Plus, as new agentic features continue to roll out, that will extend across to shopping too, I have no doubt.
That’s the hard line for me, though. My near-hatred of ChatGPT may have waned, but we’ve got a long way to go before I think these tools are safe – or sensible – in the hands of the everyman.
I did have a great vacation though, so make sure to check out all of the videos in our new ‘AI planned it’ series to see why.
