Using AI for trip planning

[JohnFoxeSheets]

BTW all this planning got overridden by a higher power, my wife told me that she wants to stop twice to take breaks.

She who must be obeyed. Good call.

Sponsored

 

[JohnFoxeSheets]

On AI accuracy: Not too long after ChatGPT first took the world by storm, I tried it for directions to a place that I've been many, many times: From San Francisco to Gualala. It's about 130 miles and there are half a dozens likely routes, each with pros and cons. So I figured I get AI to weigh in on the matter.

The first answer was straightforward: 101 north to 128 to the coast, with a bit of commentary about the drive. So far, so good (well, other than it is a looong way to get there). But then it said something along the lines of, "When you get to Highway 1, turn right and head to Gualala." Only problem: Gualala is south of where Highway 128 meets Highway 1 and we're talking about the West Coast, so it would be a looooooooooong (and impossible) drive.

So I corrected it, then asked again. Second answer was much the same, only this time it was improved with "When you get to Highway 1, turn right and head north to Gualala."

So I corrected it again, then asked again. On the third try, it finally got it right. (I didn't ask a fourth time.)

I then wrote, "What about Skaggs Springs Road?" I included no other description of what I meant, but it figured out that I was asking about using Skaggs Springs Road as an alternative route to get to the coast, and gave me a truly wonderful and insightful discussion of the pros and cons of this very rural, beautiful, and somewhat dangerous narrow road (especially in the winter). And then it said, "When you get to Highway 1, turn left and head to Gualala."

You guessed it: Skaggs Springs road joins Highway 1 south of Gualala.

I dunno about trust, but most definitely verify! (Yeah, I know this was a couple years ago, so maybe things have gotten better. Maybe.)

 

[ChrisO]

Just to play with it a bit more I asked CoPilot to do it with two stops, and no surprises, it picked Kettleman City and Santa Clarita, which are the stops have always used.

But then I noticed that the end SOC charge was wrong, with just using its own prediction on how much kWhs would be used, like @Tampamike noticed in the other predictions. So, I decided to dig into it asking it why it said that.

And I think it is really interesting. Was it a math error? Was it just making it up?
Nope.

Instead of giving the predicted SOC for the ending it was "giving the estimate for the whole trip". In other words, it was putting the "accumulated error possibilities".

Now that really doesn't make sense in this case because as you charge at each charge station, you shouldn't carry forward is error margin, but it shows that this "logic" didn't come out of nowhere.

 

[music_cities]

I love using AI for trip planning! I go back and forth between ChatGPT and ABRP and I paste the table of charge stops (as a screenshot usually, since it's easier than downloading it) into ChatGPT. Then I have chatGPT build me a PDF of each day's plan. For some reason, ChatGPT can't really hold a 10 day trip plan in it's memory, but after each day is downloaded I combine it into one big PDF and then I can upload that the ChatGPT and it can tweak it (font, format, etc). So, I end up with a nice PDF plan for the whole vacation/trip.

So, it's wonderful. But:

1) ABRP is better for the charging stops (fortunately it's easy to feed ABRP results to ChatGPT), and

2) notice in the screenshot my tables of arrival time could be formatted better — it flows over the right margin to the edge of the page!

I love it.

 

[Mach1E]

Part of me is glad AI is still dumb in certain instances.

I just got a spam voicemail that said-

“Please call us back at 8 billion, 667 million 936 thousand……….”

Pretty sure that wasn’t a human leaving that voicemail. I better call the back to be sure.

 

[celestial_knight]

I love it. Cross referencing with multiple AIs gives a more complete picture, and they research attractions and build itineraries to spec - e.g. minimal walking, parking, theoretical best times, optimized routes, specific types of attractions etc

I'm sure it's curated and sanitized, but time is of the essence

Sponsored