AI Road Trip 2024

In late 2024, I embarked on a cross-country journey from San Francisco to Hilton Head Island and back, with an unusual twist: my primary companions were AI assistants. Armed with a Tesla Model 3 on Full Self-Driving, ChatGPT, Claude, and Meta’s AI-powered Ray-Ban smart glasses, I set out to experience America through a uniquely technological lens.

The AI Companions

My journey wasn’t just about reaching a destination; it was, if anything, an experiment in human-AI interaction. I found myself developing distinct relationships with each AI system:

ChatGPT and Claude: Know-it-all friends

Throughout the drive, out of curiosity, I was peppering my AI companions with questions about everything from local history to flora and fauna. These conversations were interspersed with periods of quiet reflection, (often just voice notes to myself) creating a natural rhythm to the journey.

An interesting discovery came through ChatGPT’s voice modes. When I hit the limit on advanced voice mode and switched to standard, I found the latter’s more detailed responses preferable. The tap-to-interrupt interaction, rather than feeling limiting, became a comfortable way to manage the conversation. Over time, I developed what I’d call a “human-adjacent” relationship with the AI – like having a companion with infinite patience.

Tesla FSD + Snowy Colorado: Making out-of-distribution visceral

The self-driving experience was a tale of two extremes. In California, where the system was clearly most extensively trained, FSD has virtually no issues, suggesting confidence that made autonomous driving feel tantalizingly close to widespread adoption. However, venturing into snowy Colorado provided a stark reality check. Where lane markings disappeared under snow and surfaces became treacherously slippery, the limitations of current autonomous technology became glaringly apparent. The system, so capable on sunny California roads, became dangerous enough that I had to take full manual control. This dramatic contrast highlighted how much work remains before self-driving technology can truly be considered ready for nationwide deployment.

Meta Ray-Bans: An observing note taker and hands-free camera

Also, sunglasses, a little heavy though.

The Meta AI Ray-Ban glasses, while an extremely great form factor, proved less compelling. Their responses to visual inputs were shallow and unengaging, lacking the depth of insight I’d grown accustomed to with ChatGPT and Claude. By the midpoint of the trip, they had transformed into simple sunglasses after breaking down.

Over time, my favorite use for the Meta glasses has actually become taking notes to myself. I also appreciate keeping the visual context of what I was looking at.

When I got home, I downloaded and processed most of them through Whisper and only go back to the visual if needed.

It was also a great way to take photos hands-free. To me, that’s the killer feature.

Life on the Road

National-park centric

My route was largely structured around national parks. I’d aim to reach cities next to parks by nightfall, and make it into a park before sunrise to enjoy them.

Sleeping Arrangements

Apart from one midway break at a hotel in Hilton Head, I turned a Model 3 into a mobile bedroom. AI helped me identify safe spots near gas stations and hotels, with Buc-ee’s in the south emerging as a favorite for its well-lit premises, available superchargers, and clean facilities. The experience taught me to read the patterns of American highway towns – each exit offering its predictable constellation of McDonald’s, Holiday Inns, and gas stations.

AI + Maps is an experience I’m looking forward to improving though.

Showers and stretching

ClassPass was a godsend, dropping into yoga studios along the route for stretching, showers, and recovery. While sleeping in the car took some adjustment – my hip muscles needed a few days of gentle gym work afterward to fully recover – the body proved remarkably adaptable.

It also helped that an EV meant forced breaks for charging every 2-3 hours. It slowed down the driving, but I appreciated it in the long run.

Culinary Journey

My diet became a mix of practicality and local exploration. McDonald’s fish fillets were a staple. But I made sure to explore local cuisine in food-centric cities like Asheville (where Chai Pani was a highlight) and Memphis.

The Human Side of AI Travel

One of the most fascinating developments was how my interaction with AI evolved. I found myself naturally referring to the journey as “we” in my recordings, despite traveling solo. My voice communication became more fluid and confident, understanding that the AI’s transformer-based architecture allowed for self-correction and refinement of thoughts.

The reliance on voice interaction had an unexpected benefit: it made me a clearer thinker. It didn’t make me more dense, which is valuable in Human-Human interaction or AI -> Human interaction, but not as much in Human -> AI interaction. I liked that I could meander, but it did make me more articulate and comprehensive.

Rather than trying to make AI more human-like, I found value in embracing its bot-like nature. This acceptance led to a more authentic and useful interaction pattern, where the AI’s role as a tool and companion found its natural balance.

Reflections

This journey demonstrated both the potential and limitations of current AI technology. While AI can be an knowledgeable companion and useful tool, I missed having a human friend and their spontaneity. Next time, maybe two humans and some AIs. But I was able to go a lot deeper with the AI than I would have otherwise.

I did want to try an AI companion app at some point, but didn’t get around to it. In light of this, it was hard to imagine how it could do better.