Zenan

[Zenan @ZenanLii]

@krisgaudel Thanks! We considered Verus early on, but felt Lean was a better fit for the nonlinear and transcendental reasoning we need, e.g. for softmax.

[Zenan @ZenanLii]

Can an AI agent help optimize GPU kernels—and prove the changes correct?

Introducing VeriTile, our work on verifying Triton-style GPU kernels in Lean, with agent-assisted proofs of correctness and equivalence.

https://t.co/sguZwQQk38

[Kaiyu Yang @KaiyuYang4]

There’s a lot of excitement about AI + formal verification right now.

I’ve worked on this since 2018. I share the excitement��but the progress has also made me rethink what formalization can solve.

What Happens When Formalization Becomes Cheap?

https://t.co/Bainzj5VRs

[Zenan @ZenanLii]

Joint work with @KaiyuYang4, @__zrrr__, @_Zhaoyu_Li_, @1SHL10, and Aarti Gupta.

Code:
https://t.co/0JlkPRQrVd

Proof coverage:
https://t.co/sJWnHriTDC

[Zenan @ZenanLii]

VeriTile also models selected aspects of data movement and floating-point behavior, including memory loads/stores and abstract rounding with bf16 casts.

Agents build proofs using reusable lemmas and Lean's feedback, with explicit specifications and assumptions.