Travels with my AI
What happens when your AI tags along on your vacation?
This week we’re on vacation. And, this year, three new travel companions have snuck along. Yes, Claude, Gemini and ChatGPT have come with us. Even better it doesn’t cost anything to bring them. No need for a flight. Or passports. They aren’t fussy eaters. Ideal.
Your personal tour-guide
On our first day we found ourselves at the top of the CN tower with a view out over Toronto. Beautiful and impressive… but I quickly realised I didn’t know what I was looking at. What were the interesting buildings? What was the history of the impressive, old hotel next to the railway station?
Never fear, o3 to the rescue. Take a picture and then ask o3 about the buildings: what’s the newest, most significant, the one with the most intriguing story?
o3 introduced me to the Canada Life building and its weather-beacon mast, which flashes colour-coded forecasts four times a day (green means clear skies, solid red for a cloudy day, flashing red means rain and flashing white means snow). It told me about the old hotel - it was the Fairmont Royal York which opened in 1929 and was the tallest building in the British Empire at the time. It was like having an enthusiastic, all knowing, tour guide with us.
And then we got more adventurous. You can see Toronto City airport from the top of the CN tower.
My son asked how long the runway was. I guessed at 600m. In the past that would have been good enough. But now he wanted a second opinion. So we asked o3…
And who was correct? Yes. o3. The runway is 1215m long.
Oh well. Fortunately I’m not employed to estimate the lengths of runways…
From the tower you can also see a wind turbine. So we took a picture of that and asked o3 how far away it was. And off went o3 to get the co-ordinates of the tower and the windmill and work out how far apart they were. 3.3km, apparently. A quick check on Google Maps suggested that was right.
But then o3 made an interesting claim. Apparently the wind turbine was a kilometre below us in elevation.
Toronto is fairly flat. We were 350m up. Clearly the turbine was not 650m below ground. As ever AI can be simultaneously incredibly impressive and make silly errors. You need to have your wits about you.
Having fun
We continued to have fun. In the aquarium we took pictures of sharks and stingrays and asked o3 “We’re in the ocean and we’ve seen this, should we worry?”
But o3 wasn’t having any of our nonsense.
In Little Canada (an extensive miniature model of large parts of Canada) we took another picture and asked if we’d turned into giants…
Silly fun, perhaps. But interesting to discover we couldn’t fool the models.
Canada Day fireworks
And the uses keep coming. Some more practical than others…
A maze where we got to the middle and discovered a patch of dead grass. So we added an impressive statue to, err, improve our completion photo.
Working out how much riding a log flume would cost in GBP (it cost 6 tickets at $1.80 per ticket per rider for four riders).
Working out whether we’d get a good view from our hotel room of the Canada Day fireworks (o3’s view: probably but the atmosphere would be better outside). Again this involved taking a photo and o3 carefully examining it to decide what we might see.
Why was there a protest outside the CN tower? (It was members of Unifor Local 4271, the union that represents about 250 front- and back-of-house employees at the CN Tower protesting over working conditions).
Working out how much water is in Lake Ontario. And how long it could provide drinking water for every human on Earth assuming no water flowed in or out (answer: two to three centuries).
Playing guess-where-we-are using real-time video. This is a new game where we show the model a cryptic scene and ask it to work out where we are. It’s amazing how little information the models need to figure it out… Gemini, in particular, is especially good at noticing minor details and asking for close-ups to see them better.
Real-time voice and C++
And then there’s real-time voice. On holiday I often get up early and go for a walk or a run. It’s a good chance to explore bits of the city I might not otherwise see - and a chance to listen to podcasts and do some learning.
The problem, though, is that if I have a question I don’t have anyone to discuss it with. I’ve got to try and figure it out myself. And that can be hard. And slow.
But this year it’s all change. I’m using ChatGPT and Gemini real-time voice to update my C++ knowledge. And, wow, they are a game changer. Not only can they teach me. But I can keep asking questions until I properly understand. I can repeat back my understanding and get the models to spot the gaps - the flaws - in my understanding. Learning is faster, and, I suspect, going to stick better.
And then when we’re queuing, I can get Claude to create an interactive artefact to illustrate a concept. Want to understand co-routines? Or explain red-black trees to your kids? What members of a truss bridge are in tension? Claude will build interactive artefacts while you wait.
A lot has been written about how AI is devaluing education. But it’s a double-edged sword. Sure, AI is great at enabling cheating. But if want to learn - really want to learn - then AI is like having a personal tutor in your pocket.
The downsides
But it’s not without problems.
o3 is slow. It took two and a half minutes to answer my first question about interesting buildings. Our progress around the CN tower was, err, rather slow.
My wife and I independently asked o3 for opinions about firework watching locations. And o3 gave us two opposing answers. Which prompted an interesting debate both about where to watch the fireworks and whether we were starting to rely on AI too much.
ChatGPT and Claude are pretty hopeless for restaurant recommendations. Both regularly found great sounding, convenient options. Except they didn’t exist.
I still find talking to an AI when there are a lot of humans around a bit, well, weird. But I’m getting over this. And, in reality, I suspect many folk are listening to music or already talking on their phones - and don’t care about me.
And so?
This is the first vacation we’ve had where AI has played a major role. Claude and ChatGPT helped plan the trip. They found museums and sights we’d have otherwise missed. Even some of the restaurants. And, despite the frustrations and jagged edges, AI is net positive.
And it has continued to play a significant role while we’re here. Claude, Gemini and ChatGPT are fundamentally changing our experience.
Waiting for o3 in the CN tower, I watched other people taking photos. Documenting their experience rather than being present in the moment. And yet, with o3, I found myself more present. I was learning, understanding what I was seeing in a way I’ve never done before. Noticing buildings I’d have otherwise missed. The Fairmont Royal York became a monument to imperial ambition. It was a new way of experiencing a tall tower.
And I wondered. In years to come might we start to switch from mindlessly capturing photos that will never be viewed? Might we start to actively inquire? To better understand the history, the people, the decisions that resulted in the sights in front of us today?
Today o3 is slow. But that will change. When there’s no cost to curiosity, will we become more curious? I’d like to think so.
More disruption
AI is going to disrupt the travel industry. Why pay for a generic tour guide when you can have a personalized expert. One that can answer nearly any question, no matter how niche. Even questions about how much water is in Lake Ontario…
AI tools will plan our trips. Book our trips. Guide us while we’re away. Travel will become richer, more personalised. Until this week I hadn’t properly realised the change that’s coming. But it’s increasingly obvious the changes AI will bring.
But there are two bigger realisations. Firstly hallucinations remain, although they are getting more subtle. How many people might have accepted the claim the CN tower was 1 km above the wind turbine? These subtle hallucinations are more dangerous because they are harder to spot.
The second landed harder. My son’s instinct to seek a second opinion lingers. Am I starting to witness a fundamental shift in authority? In the past I was the authority. But now? Could I be about to be replaced? Or at least have my answers regularly checked. For now I have a role to play in spotting hallucinations - but how long does that last for?
And the fireworks? Well, despite all the calculations, the fireworks were barely visible from our hotel room. Just as well we trusted our intuition and went down to the shoreline after all.