Andi Gu (@AndiGu00) profile picture
Cambridge, MA
Joined June 2022
Photo by Andi Gu (@AndiGu00)Photo by Andi Gu (@AndiGu00)Photo by Andi Gu (@AndiGu00)Photo by Andi Gu (@AndiGu00)Photo by Andi Gu (@AndiGu00)Photo by Andi Gu (@AndiGu00)Photo by Andi Gu (@AndiGu00)Photo by Andi Gu (@AndiGu00)Photo by Andi Gu (@AndiGu00)Photo by Andi Gu (@AndiGu00)
Andi Gu retweeted
Excited to share our new paper with @AndiGu00, the best mentor ever, and Professor Susanne Yelin! Rapids is a streaming, accurate, and scalable neural decoder for real-time quantum error correction on FPGAs. Thank you Andi and Professor Yelin for such a wonderful collaboration!
Quantum computers make errors constantly. A classical chip has to diagnose them from error signals, a million times a second, without ever falling behind. We built a neural network that does this on one FPGA, 240× more accurately than today's real-time decoders.
1
6
406
Quantum computers make errors constantly. A classical chip has to diagnose them from error signals, a million times a second, without ever falling behind. We built a neural network that does this on one FPGA, 240× more accurately than today's real-time decoders.
12
29
5
333
10,973
This matters at scale. Useful fault-tolerant algorithms need thousands of logical qubits, each with its own decoder running nonstop. At one FPGA per logical qubit, that's hundreds of kilowatts, tens of server racks. The same circuit maps to custom silicon. A 7 nm ASIC layout estimate of the d=19 decoder takes 7.4 mm² and about 10 W, against ~250 W for the FPGA. Paired with high-rate codes, decoding thousands of logical qubits could fit in under a kilowatt.
1
7
372
Paper: arxiv.org/abs/2611.26447 Joint work with @ChenyangJinEric and Susanne Yelin. Eric is applying to PhD programs this cycle. Working with him was one of the best parts of this project. If you're recruiting this cycle, he should be at the top of your list!
1
10
389
LLMs have gotten very good at making classical code do more with less, from faster CUDA kernels to leaner compilers. This summer, as part of the Anthropic STEM Fellows program, we asked whether Claude could do the same for quantum computers. We found it can. Across a wide range of tasks, Claude's circuits need fewer qubits and less running time than the best published designs. The video shows the most striking one: modular inversion, the costliest step in the quantum algorithm for elliptic-curve discrete log. Claude built it on a classical method few thought could work on quantum computers.
6
7
3
50
4,068
Circuits are only one part of building a useful quantum computer. Error correction, decoders and hardware control share what made this work: a design can be written as code, checked and scored. Next is designing those pieces together instead of one at a time.
1
1
3
143
Huge thanks to my coauthors @alekhkemeny, Susanne Yelin and Mikhail Lukin, and to @J_Pablo_Bonilla, Arjun Kar, @nazli_koyluoglu, Rohan Mehta, @varun_menon0 and @vinayramasesh for feedback. Paper: arxiv.org/abs/2610.22367 Site: andigu.github.io/projects/un… xitter.cf/AnthropicAI/status/211…
New from our STEM Fellows Program: Designing more efficient quantum circuits with Claude. Given plain-language instructions, Claude Mythos 5 designed quantum circuits that improve on the best published ones, sometimes by an order of magnitude or more. More efficient circuits reduce the resources needed by quantum algorithms, which could bring quantum computing applications in chemistry and materials science within reach sooner. Read more: anthropic.com/research/desig…
3
8
256
Andi Gu retweeted
New from our STEM Fellows Program: Designing more efficient quantum circuits with Claude. Given plain-language instructions, Claude Mythos 5 designed quantum circuits that improve on the best published ones, sometimes by an order of magnitude or more. More efficient circuits reduce the resources needed by quantum algorithms, which could bring quantum computing applications in chemistry and materials science within reach sooner. Read more: anthropic.com/research/desig…
481
617
355
5,489
948,544
With @AndiGu00, we introduce Cascade; a state-of-the-art quantum error correction decoder. We show that utility-scale logical error rates are achievable with modest code sizes at current physical error rates, with real-time compatible latency. 📄 scirate.com/arxiv/2604.08358
1
3
25
1,296
New preprint with @J_Pablo_Bonilla! We built a neural network that decodes quantum errors — and discovered that fault-tolerant quantum computing might need far fewer qubits than we thought. 📄 Paper: arxiv.org/abs/2604.08358 Website: andigu.github.io/projects/ca…
3
23
1
144
7,084
On the [[144,12,12]] Gross code, Cascade reaches ~10⁻¹⁰ logical error rates at physical error rates current hardware already achieves. That's 17x more accurate than the best practical decoders, and 1000-100,000x faster.
1
1
394