Chat GPT for route planning

[jwestdog]

Has anyone else tried this. I've asked it for a few plans and I'm not sure they make sense?

 

[avanti]

Has anyone else tried this. I've asked it for a few plans and I'm not sure they make sense?

Meh.

It will give you a route that will probably get you to where you want to go, but likely not a very good one. I tried a route from Pittsburgh to Santa Fe. Route was generally ok, but it has me travel 1.25 hours in the wrong direction to get on the PA Turnpike somewhere East of New Stanton. It also suggested that I take 4 days for the drive, driving almost 10 hours to St Louis on the first day, but less than four hours on the third day, from Oklahoma City to Amarillo. Pretty random.

It is slightly more interesting if you ask for specific points of interest or other constraints. I did the same trip, but specified that I wanted to camp in State Parks. It mostly did OK, but had me spend the first night camping at Cuyahoga Valley National Park, which does not have camping. You just cannot yet rely on these programs to get basic facts correct.

Bottom line: Sort of fun, but not ready for prime time.

 

[VTPete]

Hi... I'm a bit of an expert on both ChatGPT / Generative AI and navigation - basically this has been my career since the 1980s.
In a nutshell, you absolutely do not want to use ChatGPT for this sort of thing. And, to confirm my beliefs, I tried it and ChatGPT even warns you against using it.
The instructions will be wrong in the same way that if you listened to people talking about traveling from point A to point B and tried to make sense of it, you'd get the wrong answer. It might sound (and be) reasonable, but in the end it won't be precise.
Proper navigation requires a specialized planar graph, road impedances & speeds, address to coordinate matching, and most importantly, up to date information about the roads.
Will AI be able to do this in the future? Absolutely, but don't use AI for this yet - you'll end up extremely frustrated.

 

[sysimms]

Ouch - glad I read the last post. I thought what a good idea when I read the post title

 

[WhR555]

Hi... I'm a bit of an expert on both ChatGPT / Generative AI and navigation - basically this has been my career since the 1980s.
In a nutshell, you absolutely do not want to use ChatGPT for this sort of thing. And, to confirm my beliefs, I tried it and ChatGPT even warns you against using it.
The instructions will be wrong in the same way that if you listened to people talking about traveling from point A to point B and tried to make sense of it, you'd get the wrong answer. It might sound (and be) reasonable, but in the end it won't be precise.
Proper navigation requires a specialized planar graph, road impedances & speeds, address to coordinate matching, and most importantly, up to date information about the roads.
Will AI be able to do this in the future? Absolutely, but don't use AI for this yet - you'll end up extremely frustrated.

I think the important distinction here is between using an LLM as the navigation engine and using AI as an interface to an actual navigation engine.

The point about road graphs, geocoding, routing costs, current road restrictions, traffic, etc. is still valid. A language model can produce a perfectly plausible route that is simply wrong, and that's a pretty bad failure mode when you're actually driving.

Where things have changed since the older posts in this thread is that AI can now be connected to proper mapping/routing data rather than trying to invent the route from its own knowledge. Google, for example, now exposes traffic-aware routing specifically for AI applications.

So I'd still use a dedicated navigation app for turn-by-turn directions, but I'd be much more comfortable asking an AI to explain or compare routes when the underlying route comes from a real mapping service. That's a very different proposition from asking ChatGPT to navigate you from A to B by itself.