Hi there! I’m working on a recommendation module for dining establishments, aiming to provide advice on cutting expenses and boosting revenue based on various financial reports (primarily aggregated) and performance indicators, along with some preliminary data.
How suitable could the GPT API retrieve assistant be for crafting this solution?
My current worry is that GPT might struggle with processing extensive numerical data and the sheer amount of information.
Has anyone embarked on a similar development? I’d love to hear your insights and experiences.
I would wager it would be terrible at that, it’s going to see your huge chunks of numbers and ignore them entirely, intead hungrily focusing on whatever accompanying words you provided. Then hallucinating some garbage based on your text, which at first glance, might appear valuable and correct, but it won’t be at all.
Instead what you can do is program an algorithm to deal with all that hard data and just use GPT to create fluff for your presentations, to extrapolate on your findings and create sales copy to emotionally highlight the things your algorithm has come up with and identified. Still incredibly useful for your entrepreneurial efforts, but not so much for digging into numbers and analyzing them.
By the way, if I can offer some words of discouragement in regards to this:
I’m working on a recommendation module for dining establishments, aiming to provide advice on cutting expenses and boosting revenue
Thank you for sharing your thoughts and experience.
PLEASE NO! AHHH!! ANY NICHE BUT FOOD SERVICE
Although, I don’t understand how the restaurant industry fundamentally differs from others when it comes to aggregated data and numbers, which are present in any field.
Tbh I’m not quite sure what you’re actually trying to do here. Retrieval works great (in theory) for conceptual things and patterns. For retrieving numerical stuff? I don’t know. You might find that, for instance, 14 might be closer to 4 than to 13.
So it depends on what you’re actually trying to retrieve.
Are you just trying to retrieve similar financial situations? It could work out of the box, but (experience with rag) would say that you’d be better off turning those cases into a structured sort of prose first.
OpenAI has gobbled up almost every last one of microsoft’s GPUs. If you find them struggling, you’re gonna have a hard time finding alternatives
