Why LLM predictions are dangerous for Finance?

https://dev.to/pvgomes/why-llm-predictions-are-dangerous-for-finance-j2i

Thats right, prediction are not the same as calculations, and LLMs don't calculate, they predict the next word. Stanford's HELM report shows GPT-4 hits 90%+ accuracy on math tasks. But 90% accuracy in financial calculations means 1 in 10...

#clojure #clj #cljs !clojure@lemmy.ml @clojure

0 points · 5 comments · view on lemmy.world

5 Comments

TootSweet@lemmy.world · 1 pts · 1y

LLMs are not appropriate for any use case.
What does this have to do with Clojure?

Alexstarfire@lemmy.world · 1 pts · 1y

I'm impressed by 90% accuracy for math. I thought it would be much lower. Still, useless if it's not 100%.

N0x0n@lemmy.ml · 1 pts · 1y (1 reply)

GPT-4 hits 90%+ accuracy on math tasks

How many Bs in blueberry? 🤷‍♂️

greenCoder@functional.cafe · 2 pts · 1y

@N0x0n
@planet 10%

okwhateverdude@lemmy.world · 1 pts · 1y

This is why tool use exists? They are adept at using the tools correctly to do math. Literally developing automations at $work with agents and tools to do credit analysis from financial reports.