A guest thread by @big_tech_sux, lead developer of @vyperlang 
For example, `f(x) = x/2` can never return an output which is larger than the input. You can see this, and formal verification can prove this, by reasoning algebraically about the function.
And formal verification can catch 1 in 1 quadrillion edge cases which would never come up during testing – and that edge case is the case that can lead to catastrophic failure. 
Now that LLMs are rapidly approaching AGI (Claude Fable casually proved the Jacobian conjecture false after being an open problem in mathematics for 80 years), a lot of this work becomes (much!) easier.
This translation step in particular can either be shallow or deep, depending on how thorough the translation is, and formalizing may reflect more or less of what you want to know about the system (in the literature, this is called “minimizing the verification gap”).
Besides being more feasible than before, it’s actually crucial to cybersecurity – advances in AI make it easier to produce exploits. For instance, an unreleased model at OpenAI recently discovered 0-day exploits to break through multiple layers of security.
Formal verification shifts the asymmetry towards defenders. While attackers only need to find one sequence of inputs which lets them break (into) a system, defenders can use formal verification to prove robustness against *all* inputs.
For mission-critical software like smart contracts where billions of user funds are at stake, formal verification is becoming a prerequisite, not an option.
Math and physics tell you the rules of what is possible; formal verification lets you build things that adhere to those rules.
For example you could specify “3d print me a structure which satisfies these constraints.” The AI would build it, and FV would ensure that it adheres to spec.
By having cheap, automated verification, you can try new things faster (“doing it right the first time”), ultimately accelerating human progress.
Vyper is using formal verification to verify language and compiler correctness down to bytecode, and Sequent is building infrastructure for AI and humans to harness formal verification.
I strongly think that formal verification is crucial for a future where AI and humanity coexist to make a better place for us all.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
