You could do that indeed, but the performance would be abysmal. For this kind of use-case, it would be a LOT better to use a small pre-trained model and either fine-tune it on your materials, or use some kind of RAG workflow (possibly both).
Well that's also what I used to believe, kinda makes sense. But drawing from my own anecdotal experience I had to conclude that's not the case. Maybe I'm wrong, and it's just VERY bad at it, or it doesn't work where I live.
If the developers are around, I would like to ask a simple question : Does this app only use realtime data for calculating ETA and routes, or does it learn some statistical regularities in the traffic patterns to make predictions ?
I'm asking because this is the main weakness of all similar apps that I know of, they always give you an estimation using only the current state of traffic on the whole journey, so if you are going towards a busy city at peak traffic time it may add one or two hours to the ETA even if you are still 5 hours away and everything will be gone by the time you get there. Obviously the reverse situation is even more annoying, I live in the Paris area, and when I leave home around 4pm and ask google maps an itinerary, it may tell me I'm only 30 minutes away, but 15 minutes later shit hits the fan and I end up an hour late on my schedule.
Obviously after a while you start learning the traffic patterns and plan accordingly, which is okay I guess, but we're in 2023, how can google not be up to the task of correctly predicting a traffic spike when it's regular on a daily or weekly basis ? Is that just too much data / compute for them ? If anybody has a clue, I'm curious.