The University of Alabama in Huntsville is installing a new computer cluster system containing IBM AIU chips for running advanced AI models developed by IBM and NASA: https://lnkd.in/gcHYY2si 🦾 - IBM Spyre is the first AIU (artificial intelligence unit) production accelerator born out of IBM Research, and is part of a long-term strategy of developing novel architectures and solutions for the emerging space of generative #AI. 🌎 - Researchers at #UAH will primarily use this new cluster for deploying the Prithvi geospatial and weather and #climate models from IBM and NASA - National Aeronautics and Space Administration, testing the operational workloads and running throughput-focused workloads. 🔴 - The cluster will run on Red Hat OpenShift AI, demonstrating the value of a full-stack solution that leverages heterogeneous accelerators. --- #IBM #Research #NASA #semiconductors

We’re exploring where quantum can make the largest impact in energy with E.ON. https://ibm.co/4fobSGB E.ON serves over 47 million customers across 17 countries everyday, powering infrastructure through electricity and gas—and that is no easy task. In years past, E.ON could reasonably predict costs and consumption to ensure these customers would always have the lights on. Now, with changes in technology, sudden weather, and the differing ways we use electricity each day, small variables to supply and demand of energy can drastically impact predictability of costs, resource allocation, and energy delivery. It is a complex problem E.ON faces, but that’s exactly what quantum computing aims to solve—problems with many variables that might take millennia to solve on classical supercomputers might have much more straightforward solutions using quantum computing algorithms. E.ON is exploring how quantum could help them plan for coming fluctuations and predict patterns years into the future, which ultimately lower costs for their customers and keep the lights on for all. Head over to the link above for the full case study and watch the film on https://lnkd.in/exCbDYQ6 for more.

In this week's newsletter, we are exploring the path to more powerful — and efficient — AI systems. We introduce the AIU family, explore IBM Spyre AIU at The University of Alabama in Huntsville, discuss a new toolkit for document conversion, and highlight key moments from the first-ever IBM Quantum Developer Conference. Read more for the latest updates and subscribe here:

We’re excited to share some upcoming innovations and roadmap updates needed to realize fully error-corrected quantum computing at scale. https://ibm.co/4etLCtg On the pathway to realizing full-scale quantum computing is developing couplers that run gates across multiple quantum chips. This year at the first-ever IBM Quantum Developer Conference (QDC), we reported the results of two kinds of couplers: l-couplers, which connect chips with cables, and m-couplers, which seam together adjacent chips. First is a proof-of-concept for l-couplers we’ve named IBM Quantum Flamingo, which connects two Heron r2 chips with 4 connectors measuring up to a meter long. The next is an m-coupler proof-of-concept called IBM Quantum Crossbill. This concept connects three Herons with 548 couplers and 8 interchip m-coupler connections. At the moment, we’ve benchmarked the best CNOTs with errors per gates of 3.5%, while state transfer takes around 235 nanoseconds on average, on Flamingo. We expect these metrics to improve, and hope to debut a production-ready Flamingo chip for use by our clients at our 2025 quantum state-of-the-union. We will soon begin development on c-couplers, or couplers that link distant qubits on the same chip, with hopes for demonstrating this in 2026. These innovations are necessary to for us to implement scalable quantum computing, as well as the error correction code we shared earlier this year (https://ibm.co/4ezbrrE). This code has the potential to store quantum information with a fraction of the overhead associated with other leading error-correcting codes, but needs higher qubit connectivity between multiple chips to reach its potential—which we're also demonstrating today. We are excited at the prospect of the proof-of-concept innovations we’ve unveiled at this year’s QDC to help us get to that point. More details at the IBM Quantum blog above.

"If you ever have to look at anything up to 500,000x, call me.” 🤙 IBM Research Advisory Engineer John Ott demonstrates how the scanning electron microscope (SEM) at the Thomas J. Watson Research Center in New York fires beams of electrons at an object through a magnetic lens - instead of a glass lens - to achieve 500,000x magnification. Join us on a look inside Ott’s lab as he uses the SEM to give you a view of a penny you’ve never seen before: https://lnkd.in/gGdkxNFk ---- #IBM #Research

Today at the first-ever IBM Quantum Developer Conference (QDC), IBM Researchers shared that they’ve successfully delivered a system capable of running accurate calculations employing circuits with 5,000 two-qubit gates. https://ibm.biz/Bda9Kz The second iteration of the IBM Quantum Heron quantum processing unit is what drives theese capabilities—powered by 156 qubits in a heavy-hex layout. The new design preserves the tunable coupler architecture we introduced last year to suppress crosstalk, and features new two-level system mitigation to reduce the impact of noise. This newer Heron QPU not only features a 16x improvement in performance, but a significant 25x speed-up in terms of quickness over previous generations. But these improvements also require the collective effort of the quantum computing community to develop algorithms that would be able to leverage the full power of a system like the one we’re sharing today if we hope to drive the field forward. We believed that 5,000 two-qubit gates was an ideal goal, being in the regime of circuits beyond classical simulation. Reliable results from quantum circuits with 5,000+ gates grants users the opportunity to perform real scientific discovery with quantum computers, and to push forward in the search for quantum advantage. And we’re thrilled to share that a number of our startup partners have also delivered utility-scale capabilities—directly integrated as part of the Qiskit Functions catalog—with many approaching that 5,000 gate threshold. We committed to delivering monumental improvements in both our hardware and software. We asked the community to help us in the push for quantum algorithms that could take advantage of those improvements when ready. Now, we’re fully ready for our developer community to start seeking quantum advantages to help us deliver useful quantum computing to the world. More at the IBM Quantum Blog linked above.

Large amounts of valuable enterprise data lie buried in PDFs, annual reports, and other business documents. #Docling, IBM's new open-source toolkit, is designed to extract and process this information so that large language models can digest it. How it works: Docling converts complex documents into #JSON and #Markdown files that can be used to fine-tune LLMs for enterprise tasks and to ground them on trusted data via retrieval-augmented generation. Docling can run on a standard laptop and takes just five lines of code to set up. 🦆 - Learn more on the IBM Research blog: https://lnkd.in/gwnkFQmV 🦾 - Try it and contribute via GitHub: https://lnkd.in/gUYgQByS

In this week's newsletter, we discuss surfing, customizing AI models, and really big microscopes. We examine customized AI models using LoRA, explore how Campbell Watson is applying AI to Earth science, and take a look inside an anechoic chamber at IBM Research in Yorktown. Read more for the latest updates and subscribe here:

The passion for catching the perfect wave set IBM Research scientist Campbell Watson on his professional journey to study the atmospheric conditions that influence his favorite hobby. Today, Watson works on advanced models of the atmosphere, including IBM Research and NASA - National Aeronautics and Space Administration’s newest Prithvi WxC model, that aim to help us better understand the changing world we live in: https://lnkd.in/gzrmTVfu

Fine-tuning large language models on specialized data the traditional way can involve updating billions to trillions of weights. Low-rank adaptation, or #LoRA, offers a shortcut. With LoRA, you change only a tiny subset of the base model’s weights, creating a plug-in module (also called a LoRA!) that gives your model domain-specific expertise at inference time. Like custom bits for a multi-head screwdriver, LoRAs can be swapped in and out of the base model to give it specialized capabilities. Here are several ways IBM Research is innovating with LoRA to make it easier to customize and serve #AI models at scale. https://lnkd.in/e_4G4rV8