Thesis: Using AI with financial systems is a ticking time bomb

LLM's are revolutionizing a lot of things about our lives. There are dozen of companies offering AI agents that will interpret your financial data for you. Given you hive level summaries, forecast, charts whatever you want.

But it's all a lie.

The underlying data that powers the worlds financial systems is modeled using double-entry ledgering, which is fundamentally non-deterministic. When you layer the non-determinism of an LLM on top of that you magnify error margins to the point where the result can be completely disconnected from the truth. While all LLMs deal with hallucinations the nature of the underlying system makes it impossible provide sources or traceability follow LLM decisions back to their core proof.

Participants will learn:
- The basics of double entry ledger
- Why double entry is non-deterministic
- How layering non-determinism magnifies errors
- Why hallucinations are even more problematic in this paradigm
- How can we make it better?