LLMs vs Decision Models
A high level visualisation of the differences between how a large-language-model and a decision model produce output.
Input (Prose)
Today's scam patterns: safe account calls; private car sales...
Flagged payees: QuickCoin Ltd, J. Marku.
Rank each transaction as safe, low or high risk.
Reply ONLY with JSON: {"txn_1": {"choice": ...}, ...}
1. £14,900 to "Quick Coin Limited", new payee, 23:40...
2. £42 to "Tesco", regular payee, "weekly shop"
3. £880 refund from "GymFlex", known payee
Input (JSON/Image)
{ "state": {
"scam_patterns": ["safe account calls", "private car sales..."],
"flagged_payees": ["QuickCoin Ltd", "J. Marku"] },
"questions": { "type": "choice", "criteria": ["safe","low","high"],
(not needed: output is typed)
"txn_1": { "instructions": "£14,900 to 'Quick Coin Limited'..." },
"txn_2": { "instructions": "£42 to 'Tesco', regular payee..." },
"txn_3": { "instructions": "£880 refund, 'GymFlex', known payee" }
} }
LLM (e.g. ChatGPT)
Decision Model (e.g. Jev)
Output
...
Output
...
passes: 0 · output tokens: 0
passes: 0 · output tokens: 0