Chengkai Zhu

Ph.D. student in Quantum Information @HKUSTguangzhou

[Mo Bavarian @mobav0]

This is a surreal moment. Few people could have predicted that the AI will advance to solving math problems at the highest level only a few years after GPT-2/3. The models then couldn't reliably solve grade school math problems. They barely were good enough to draft emails. They still very much felt like statistical parrots.

All things that looked like fundamental limitations slowly faded with some advances (e.g. high scale RL) in a span of a few years, which is really a short time. We should behold this moment both in awe and disbelief. What will a few more years of progress bring? How is it going to impact the world and society at large? Are we ready for the tsunami of intelligence at our fingertip?

More than any other moment, this feels to me like the eve of singularity. Glasswing & huggingface incident further increase the gravity.

A few years ago, deep down I felt working on alignment is premature. It's nice to do if it's your passion, but the shapes of things weren't clear enough for it to be critical in my opinion. The chances that you end up working on things that are useless for aligning the actual AGI was high. It's different now. Now, it feels like it's the most critical thing facing us.

https://t.co/SDdrvNRefr

[Lijie Chen @wjmzbmr1]

10 proofs from our next major model Astra on long-standing open problems in mathematics and theoretical computer science (also including new circuit lower bounds for computing the permanent!)

GPT-5.6 has already enabled so much exciting work in math and science. Can’t wait to see what comes next!

[Steve Flammia @S_Flammia]

On the future of papers. 🧵

I can envision several futures for papers. In Future 1, we raise the standard for what goes on the arxiv. Only papers solving hard problems, using formal verification, open source code, and clear contribution statements get put online. 1/8

[Dmitry Grinko @dmitry_grinko]

HAWK is withdrawn from the NIST competition https://t.co/jTI3DDhzD8

[Chengkai Zhu @chengkai_z]

@sujeetbhlr @wangxinfelix Thank you! It is a harness system we are still developing while it basically contains steps to do search, proof, and retry, etc.

[Chengkai Zhu @chengkai_z]

1/6 Frontier LLM models just helped us solve an open problem on quantum channel capacity (https://t.co/3VSAKX9F6V) I’d been stuck on since 2022, when @wangxinfelix first posed it to me. I honestly don’t think I would have completed this work without AI assistance.

[Fernando Brandao @fgslbrandao]

@letonyo @LamiLudovico @QuAntonioMele It's hard to extrapolate a few years when we are in an inflection point. But I think we should not be defeatist. These are amazing tools and we all ought to be trying to solve much harder problems than we could before.

[Kimi.ai @Kimi_Moonshot]

Releasing the model weights and technical report of Kimi K3.

Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.

New model architecture: 2.5x the intelligence per unit of compute, not just more params.

Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale.

Model weights: https://t.co/7m7eEg6Y0B
Tech report: https://t.co/yeu6cjpMCT
Tech blog: https://t.co/YTfiMSNM1f

[ali @waterloo_intern]

I spent 48 hours with the Kimi K3 modeling code.

It took:
- 650 mg of caffeine (mandatory)
- 40 cans of LaCroix (optional... world record (?))
- 8 papers
- 6 months off my lifespan

Finally grokked the entire lineage of Kimi K3 and how we got here... every single step, since 2019 GPT-2

[Chengkai Zhu @chengkai_z]

6/6 Standing at the edge of this new horizon, I feel we are witnessing the dawn of a new paradigm in discovery. Excited—and humbled—to keep exploring the uncharted frontiers where human insight meets machine intelligence.

[Chengkai Zhu @chengkai_z]

5/6 We checked and formalized AI proofs/ideas and developed the complete picture. To me, this experience offers a genuine glimpse into the future of AI-assisted scientific research—a prospect that fills me with both immense excitement and, quiet unease.😰

[John Preskill @preskill]

Thoughtful discussion @SimonsInstitute led by @RobertHuangHY on the topic researchers everywhere are talking about: How to respond to the automation of research?
https://t.co/us73LnXa5X

[Chengkai Zhu @chengkai_z]

We relate our problem to the quantum signal processing (QSP) and developed an algorithm to simultaneously realize a poly regardless which signal unitary (from two choices) is chosen for the QSP, to do the optimal ternary Hamiltonian recognition.

[Chengkai Zhu @chengkai_z]

Excited to share our new paper on Hamiltonian recognition! We explore the task of guessing which Hamiltonian is governing the unknown unitary evolution by multiple queries to the unitary. Joint work with Shuyu He, Yu-Ao Chen, Lei Zhang and @wangxinfelix
https://t.co/d2iTwDQ33w

[Chengkai Zhu @chengkai_z]

Essentially, it is a composite unitary discrimination problem that can bridge quantum hypothesis testing and metrology! We developed an optimal protocol for identifying H for an unknown qubit quantum evolution exp(-iHt) with unknown t, from two or three orthogonal Hamiltonians.

[Xin Wang @wangxinfelix]

New work on the limits to the scalability of circuit knitting in distributed q. computing. The sampling overhead is exponentially bounded by the exact entanglement cost of the target bipartite channel, even asymptotically. With @MingruiJing @chengkai_z
https://t.co/yWIFawkMfp

[Chengkai Zhu @chengkai_z]

This result reveals intrinsic similarities and distinctions between the quantum resource theory of magic states and entanglement in QSD. It highlights the operational advantage realized by the quantum resource of magic states. 5/5

[Chengkai Zhu @chengkai_z]

Excited to introduce our latest paper, “Limitations of Classically-Simulable Measurements for Quantum State Discrimination” https://t.co/3SKEerGqON which explores the utility and role of magic in extracting classical information from quantum states. 1/5

[Chengkai Zhu @chengkai_z]

Our research demonstrates that any pure magic state and its orthogonal complement of odd prime dimension cannot be unambiguously distinguished by non-magic POVMs, regardless of how many copies of the states are supplied. This is similar to entanglement theory. 4/5