LLMs are obsessed with non-negative balances?
Something interesting about asking LLMs to explain what formal verification is to me: They always use the example of "a bank" or "a store" where either one customer tries to make a transaction twice -- or two customers attempt to deplete a finite resource (a balance) simultaneously.
I think this is:
a) There is some strong stuff in the training data.[1]
b) This is a nascent field, really.[2]
Linked from
No incoming links yet.
