@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20240205

Episode Date: February 5, 2024

- Argonne National Lab's Nexus Integrated Research Infrastructure (IRI) - Delays in Intel's Fab in Ohio, CHIPS Act - Mitchel Institute report on Quantum Information Science and Technologies - Controlling Emergent AI, High Calory Data [audio mp3="https://orionx.net/wp-content/uploads/2024/02/HPCNB_20240205.mp3"][/audio] The post HPC News Bytes – 20240205 appeared first on OrionX.net.

Welcome to HPC News Bites, a weekly show about important news in the world of supercomputing, AI, and other advanced technologies. Hi, everyone. Welcome to HPC News Bites. I'm Doug Black. Hi, Shaheen. You know, we've been reporting on the U.S. Department of Energy's vision of an integrated research infrastructure, the IRI. And this week,

Argonne National Lab announced an effort it calls Nexus to drive this integration. IRI combines scientific equipment with supercomputers and data resources to accelerate research. An example is Argonne's advanced Photon source, which is getting a massive upgrade this year. Yeah, many scientific instruments are themselves becoming mini supercomputers. To put it in perspective, experiments using the X-ray light source at Argonne can generate 100 to 200 petabytes of data per year. The Nexus IRI streamlines capturing, reducing, storing, and processing that data. And it takes advantage of tools such as Globus, an early pioneer of grid computing well before there was a cloud, dating back to the mid-90s and developed at Argonne and at the University of Chicago.

Argonne's work here also includes collaborations with the National Fusion Facility in California and CERN's Large Hadron Collider in Switzerland. You may not be familiar with the Mitchell Institute for Aerospace Studies. It is an independent, nonpartisan think tank with close ties to the military community. It released three policy papers this week and associated recording of a panel discussion on the expected and potential impact of quantum technologies on national security. Among other recommendations, it makes the point that defense specific applications of quantum information science require that the Department of Defense

adopt a corresponding defense specific investment strategy that goes well beyond typical R&D-based investments. Yeah, the three papers describe the science behind quantum technologies with a focus on gaining what it calls valuable warfighter capabilities. They also emphasize something we've talked about here, that while quantum computing is an important goal, quantum sensing and communications are advancing rapidly and can, quote, fix known defense vulnerabilities and capability gaps within the next five years. Things like timing, navigation, sensing, and radio frequency reception. The papers and a YouTube video of a panel discussion are well done and highly recommended. The Wall Street Journal broke the news on Friday that Intel

is slowing down construction of its $20 billion semiconductor fab in Ohio due to, quote, market challenges and the slow rollout of U.S. government grant money to grow the domestic industry. Originally, Intel intended to have the fab up and running in 2025, but now the company says the facilities won't be completed for two years with actual chip production happening after that. So given the previous good news of Intel's 18A fab that moved to this year instead of next year and their New Mexico plant that went live, it's interesting to see this news. Any delay, of course, is disappointing, but the reasons they gave for this one were not technological reasons. We've applauded the Chips and Science Act, which expands US chip production capabilities and reduces American dependence on chip manufacturing in Asia. But Intel said from the beginning that the scope and pace of its expanded chip manufacturing would, quote,

depend heavily, end quote, on government funding. So if that funding is not flowing fast enough, then it matches where you see the delay. Their Ohio plant, which was a location that was prominent during the announcement of the Chips Act. We'll finish with the critical topic of AI ethics and regulation. Former Google CEO Eric Schmidt wrote a really interesting opinion piece in the Wall Street Journal this past week about a fascinating and troubling aspect of large language models, their what's called emergent capabilities. Some models, he said, appear to know more than just what is in their training data and can link concepts across fields, languages, and can link concepts across fields, languages, and geographies. At some point, they may be able to, for example, suggest recipes for

novel cyber attacks or biological attacks, all based on publicly available knowledge. For now, the response has been the use of so-called red teams, which provoke the LLM in order to uncover harmful output. But Schmidt said this approach may prove insufficient as models become more powerful. He suggests government regulators require AI models to be evaluated by government-certified private testing companies that would compete to scale their testing capabilities as fast as the models they're checking. This is such an important topic. On emergent capabilities of AI, we have to do a much better job of assessing the information

content of data that we feed AI. Complex data is information rich. It's high calorie data, so to speak. So if your AI is on a high calorie information diet and is getting more information than you think it is, then it's not the AI being emergent. It is humans not knowing what they're doing. And that's a problem in itself since that's how Darwinian mistakes can happen. It is impossible to slow down or pause the advance of the technology, and it may be undesirable to do so. But what we definitely need is policies that minimize any disruptive aspects of AI while allowing AI to solve all the problems that it can solve, from climate change to cures

for diseases and whatnot. We need policies that help facilitate a smooth transition to AI nirvana. All right, that's it for this episode. Thank you all for being with us. HPC Newsbytes is a production of OrionX in association with InsideHPC. Shaheen Khan and Doug Black host the show. Every episode is featured on InsideHPC.com and posted on OrionX.net. Thank you for listening.