Aws Albarghouthi

Member of the load-bearing staff

[Aws Albarghouthi @awsTO]

The trajectory is clear

[Anthropic @AnthropicAI]

New on our Frontier Red Team blog: using Claude to formally verify software at large scale.

Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).

With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.

Read our research in full here: https://t.co/zUyx8CjBEk

[Aws Albarghouthi @awsTO]

We should rebrand Formal Verification to Proof Against the Machine

[Aws Albarghouthi @awsTO]

I've discovered that all popular quantum computing SDKs / compilers have serious bugs when it comes to handling angles -- floating point strikes again!

IBM's Qiskit, Google's Cirq, Rigetti's quilc, IQM's Qrisp, NVIDIA's CUDA-Q, Xanadu's PennyLane, qBraid's SDK, ...

🧵1/n

[Lucy Li @lucy3_li]

Models are sycophantic, and when they do challenge faulty or incorrect inputs, they're ambiguous and ineffective.

In a new paper w/ multilingual co-authors, we find that people may view LM responses that push back, yet still end up believing false content is true.

[Aws Albarghouthi @awsTO]

@ShriramKMurthi https://t.co/xP3lLjgsN3

[Gary Shiu @gary_shiu]

More cake to celebrate Francis Halzen’s well-deserved Nobel Prize @NobelPrize ! 🎂 Proud to be part of UW–Madison Physics @UWMadPhysics, where extraordinary science continues to happen. Here’s to more discoveries—and more milestones worth celebrating—in the years ahead!

[Aws Albarghouthi @awsTO]

@xwangsd Security is solved

[Aws Albarghouthi @awsTO]

I used to Claude to build some security tool. It built the thing. Then at some point it realized it’s debugging a security tool and the guardrails kicked in 😂

[Aws Albarghouthi @awsTO]

@exocognosis I didn’t say that supremacy experiments are affected. I have just lost confidence due to the poor quality of the tool chain.

[Aws Albarghouthi @awsTO]

I've discovered that all popular quantum computing SDKs / compilers have serious bugs when it comes to handling angles -- floating point strikes again!

IBM's Qiskit, Google's Cirq, Rigetti's quilc, IQM's Qrisp, NVIDIA's CUDA-Q, Xanadu's PennyLane, qBraid's SDK, ...

🧵1/n

[Aws Albarghouthi @awsTO]

@zengola exactly :) not surprising in retrospect, but one would have higher expectations from widely used packages

[Aws Albarghouthi @awsTO]

What should quantum software builders do?

(1) for now, get rid of floating point throughout (except at the boundaries). Replace it with symbolic, arbitrary-precision arithmetic. This is what I did with tzap.

(2) formally verify the entire stack. There are no excuses now.

6/6

[Aws Albarghouthi @awsTO]

IMO, any future experimental claims should come with the entire software stack. Otherwise, take them with a grain of salt and a shot of Guadalajara's heaviest offerings.

5/n

[Aws Albarghouthi @awsTO]

@TaliaRinger nice — exactly what I’m looking for.

[Aws Albarghouthi @awsTO]

Is anyone teaching a course on the foundations of interactive theorem provers?

[Aws Albarghouthi @awsTO]

@doenges_cs Thanks, Ryan!