I’m using GPT 4, I basically uploaded the pdf, it then did a basic analysis of the pdf but had trouble finding the points section so I pointed it to the right page and then asked it to give me a strategy with the amount of points scored for a low tier, mid tier, and high tier team.
The film’s soundtrack was so forward thinking it effectively is the first fully electronic film soundtrack and broke a bunch of concepts and ideas at the time.
More fun:
Marshall (hey, that’s me) got the soundtrack as a Christmas gift from a professor in college for helping them break copyright laws to show a German film in class with English subtitles.
Ive used ChatGPT for many FIRST related things and I will say it is no match for people who know the facts. It’s better than nothing, but you and your own team will do a much better job.
For example even using a much smaller fake game, it still was missing critical components of the game and mixing up terminology.
This is about as good of an analysis as I would expect from someone glancing the rules but not studying them in depth. Essentially that’s what the LLM is doing really with your uploaded PDFs. If it’s incredibly long it will not be able to recover each and every detail and will have to go back and be pointed to specific sections. Robot Construction rules in the official manual is 38 pages. There’s just too much data there for it to know what’s actually important.
What it is better at is helping you draft up discussion points, turning rough half sentences into full coherent thoughts, etc. you need to give it the crucial strategy and specifics if you want anything close to a coherent answer back. If you copy pasted the key chunks about game elements, score, periods of time and etc it could give you a rough strategy but based on what? It’s making educated guesses based on what it was trained on and more generally what most of the world is interested in. FRCFIRST Robotics Competition is such a small community of interest that FRCFIRST Robotics Competition game manuals arent an intentional part of the training process. Custom GPTs might be your best bet, but those are still limited to ChatGPT Plus users and they still need lots of fine tuning in the setup mode to get what you want.
It takes more time to fine tune the system and you have to know the game well enough to train it what’s good and what’s bad. At that point you know the strategy and game possibilities. If you know all of that why do you need the AI at that point?
