Tag Archives: chatgpt

I’m getting mighty bored sitting around waiting for my airplane to get fixed. I thought it might be fun to go see some fast jets, so I asked some AI tools to help me plan an outing.

Here’s the prompt I used: make me a list, sorted by distance from Huntsville, Alabama, of the remaining airshows this calendar year where either the US Navy Blue Angels or the USAF Thunderbirds are performing.

ChatGPT

ChatGPT was first up. It searched for the airshows and gave me a nice table of locations sorted by distance. Here’s its summary.

HUGE miss in that its #2 choice, Meridian, isn’t on the official Blue Angels schedule. It then compounded its error with this gem of a closing comment: “If you’re willing to make a 2–4 hour drive, I’d personally put Jacksonville #1, Pensacola #2, and Peachtree City/Meridian/Owensboro as the convenient options.” There is no universe in which you can drive from Huntsville to Pensacola, much less Jacksonville, in 2-4 hours. It doesn’t matter how long it takes to get to Meridian, since no one is performing there. Not a great result.

Grok

Grok is usually pretty dumb at reasoning, and its writing quality is poor, but I find that it often has better web search results than its peers. Here’s what it came up with. Note that it threw Terre Haute in as an option that ChatGPT missed, so that’s worth a few brownie points.

Overall, not a bad result, with 5 reasonably solid choices, all of which appear to be accurate.

Claude

Claude did a decent job too. It pointed out that the distances shown were all straight-line distances, which is useful to know. It’s not clear whether the other tools were quoting straight-line distances, driving distances, or some made-up hybrid. Given that issue, it’s hard to know how much credence to give the results… is Branson closer than Terre Haute? I’m genuinely not sure.

Copilot

Copilot gave me the helpful hint that it had ranked the shows by approximate driving distance, but instead of giving me a neat table it gave me what I would describe as a listicle. Not super useful. It also told me the list was ranked in order of distance, but it wasn’t; its top 5 were Peachtree City, Pensacola, Lake Charles, Cleveland, and Owensboro. Confusingly, it also gave me a top-5 summary list that was in the correct order. As a nice side feature, it offered to make me other lists (like “best for photography”), put the events on my calendar, or fetch me Google Maps driving directions. Mixed bag here; the extra utility was nice but the fundamental inability to build a simple table was a big miss and it’s bizarre that it was simultaneously right and wrong about the sort-by-distance metric.

Gemini

I don’t like Google, so I don’t use their services much. I almost skipped Gemini for this trial, and I’m glad I didn’t… the results gave me a good laugh. Its “sorted by distance” list offered me Pensacola, Chicago, Virginia Beach, Houston, Huntington Beach, San Francisco, and Nellis AFB/Las Vegas. Because its summary included links to the official schedules, I can only conclude that Gemini is evil and wanted me to spend unnecessary time driving to San Francisco instead of going to Atlanta.

Summary

If I didn’t know anything about US geography, the Copilot and Gemini lists would have been disastrously bad. If I didn’t know anything about how AI assistants can lie, the Gemini results might have fooled me. ChatGPT included a “nearby” show that doesn’t exist, so in this case it wasn’t trustworthy. Grok and Claude did an acceptable job, certainly good enough to narrow down my choices a bit.

Which show am I going to? Of the remaining venues for both demo teams this year, I’ve already been to Cleveland, San Francisco Fleet Week, and Pensacola, and of those I think Pensacola would be my first choice, but it’s hard to beat the ease of getting to Peachtree City or Owensboro by car. Stay tuned.