Qolab

Currently reading John Martinis, Robert Mc Dermott & al blueprint for a superconducting qubit utility-scale QPU (related to their Qolab company). This is the most amazingly comprehensive blueprint I have ever read on this matter. It adopts a real full-stack approach for quantum computing. It makes a honest inventory of all the challenges ahead and proposes either solutions or research avenues. Lots to unfold here on all scaling aspects from hardware to software, science and engineering, error correction, control, manufacturing, computing times and speed, etc. They are proposing the creation of very large superconducting qubit chips (14 cm x 14 cm) hooked to another wafer containing the wiring from 15 mK to 3.5K going to cryo-CMOS controls. Also, at the qubit level, readout is to be done with Josephson photomultipliers, removing the need for TWPAs and circulators. That is quite original. ... Finished reading the paper. The contribution from HPE is significant here. They provide lots of insights on HPC-QPU connectivity like using circuit knitting. Wondering how it scales. Then, resource estimates for QPE for benzene and FeMoCo, have computing times in years, even for the most optimistic qubits quality (for 76 orbitals FeMoCo). The authors view is to build an "horizontalized" QPU ecosystem with various specializations. Thus the participation of Applied Materials (cleanroom and fab), HPE (HPC-QPU integration), Synopsys (EDA), 1Qubit (resource estimation), etc.

Smash follow to find out how to build a quantum supercomputer computer !

This is a rant - understanding is fundamentally different than reasoning. LLMs are amazing at understanding the natural language, but there is nothing to indicate that it can reason well. Where I think where investors are confused is 1) underestimating the size of the market for applications that understand natural language. 2) overestimating the market size of Agentic systems because LLMs have a limited capability to reason. This is why I’m so bullish on quantum. With Convolutional Neural Networks, sudden computers can “understand” images and video, which created a huge unlock. With Transformers, suddenly computers can understand natural language which is a huge unlock. With Quantum neural networks, suddenly computers will understand materials and chemistry at a level never achieved which will be a huge unlock. My point is, we don’t have to solve the reasoning problem to move humanity forward.

Today the Nobel prize was awarded to Geoff Hinton and John Hopfield. Their impact on ML is undeniable, but what’s better ? Their full vision has not been realized yet ! Why - current architecture (feed toward neural networks) are a restricted form of Boltzmann machines / Hopfield networks. It’s also hypothesized that to achieve true AI with human level reasoning, we need to explore these architectures in more detail. (Read up on Yann LeCunn’s argument for energy based models). I hope this prize will reinvigorate the CS and Physics community to explore Geoff Hinton /John Hopfield’s original proposal in much more detail. https://www.nobelprize.org

Goldman Sachs in 2023 : Generative AI could raise global GDP by 7% Goldman Sachs in 2024 : Gen AI: too much spend, too little benefit? Both articles are on their website: https://lnkd.in/eMt8xCwi https://lnkd.in/enTk3VxV #GenAI #hype #doomer