Anonymous
6/12/2025, 5:36:58 PM
No.105572340
>>105572182
>and requires intensive and very careful ahead-of-time architectural planning
Does it? Can't you just specify properties for each function and then prove that the code for that function meets them?
That's why I don't like OOP or even just complicated data structures in functional programming. because then the proof depends on the implementation of each object that the function interacts with. if you have a simpler data structure like, say, a list of lists of strings or something simple like that, then the properties of that data structure are obvious unless you add additional restrictions on what that list can contain.
And that's where I see AI coming in. you specify in natural language what the function does and the AI writes the formal spec for you, then tries to find a proof, possibly leveraging human intuition if the proof is hard but also helping with language features so the user doesn't have to be an expert on that specific proof language to help design the proofs. Nobody expects AI to verify a whole project all by itself. The Isabelle proof assistant already uses machine learning models to try to automatically prove lemmas.
>and requires intensive and very careful ahead-of-time architectural planning
Does it? Can't you just specify properties for each function and then prove that the code for that function meets them?
That's why I don't like OOP or even just complicated data structures in functional programming. because then the proof depends on the implementation of each object that the function interacts with. if you have a simpler data structure like, say, a list of lists of strings or something simple like that, then the properties of that data structure are obvious unless you add additional restrictions on what that list can contain.
And that's where I see AI coming in. you specify in natural language what the function does and the AI writes the formal spec for you, then tries to find a proof, possibly leveraging human intuition if the proof is hard but also helping with language features so the user doesn't have to be an expert on that specific proof language to help design the proofs. Nobody expects AI to verify a whole project all by itself. The Isabelle proof assistant already uses machine learning models to try to automatically prove lemmas.