1/Everyone in finance is using LLMs.
But no one can explain what’s actually happening under the hood.
What allows this thing to generate coherent answers to questions it’s never seen before?
🧵
1. You believe LLMs are neutral like a Google search? Time to talk about "RLHF" 😏
Every response you get has been shaped by Reinforcement Learning from Human Feedback (RLHF). After pretraining, LLM models are shown multiple output variations.
6. Ask about inflation hedging strategies, and LLM will favor vanilla instruments (TIPS, gold, REITs) over more aggressive plays like shorting duration or going long commodities futures.
Even when market conditions scream for it.
1/Everyone in finance is using LLMs.
But no one can explain what’s actually happening under the hood.
What allows this thing to generate coherent answers to questions it’s never seen before?
🧵
6/ It also learns that certain sentence structures tend to follow certain forms.
So it guesses, one token at a time, what fits best based on learned patterns.
What it lacks:
- No concept of ground truth
- No economic theory
- No mental model of cause and effect
6/ So next time you check your bank balance, remember: Don't just look at the number.
Consider your future cash flows, and plan your spending accordingly 💰👍
#personalfinance#forecasting
1/ Why does payday feel like a celebration but month-end seem like a harsh reality check? It's because look at today's snapshot of their finances instead of looking at what's ahead.
5/ The problem with most banking or budgeting apps? They're stuck in the past. They show you what you spent last month - that's 'backward-looking'.
And budgeting based past data points is a recipe for disaster 💣