@HPC Podcast Archives - OrionX.net - HPC News Bytes – 20240715

Episode Date: July 15, 2024

- AI survey by S&P Global Market Intelligence commissioned by Vultr - OpenAI proposes 5 levels of AI based on capability - SoftBank acquires Graphcore [audio mp3="https://orionx.net/wp-content/uploads/2024/07/HPCNB_20240715.mp3"][/audio] The post HPC News Bytes – 20240715 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 with Shaheen Khan. You know, we've talked several times lately about the bloom coming off the AI rose a bit with studies, statistics, and comments indicating that the vast amounts of money poured into AI aren't showing much return yet. My own feeling, Shaheen,

is that Gen AI and LLMs will, in the longer term, deliver major returns. And we're already seeing it in some sectors, as cited by Nestor Maslay of the Stanford Institute for Human-Centered AI and their annual AI Index Report. He appeared on our HBC podcast last month and discussed fields such as the legal industry, where AI is making workers measurably smarter and more productive. And now comes a study commissioned by cloud platform company Volter, which said the key to strong AI ROI is the relative maturity of organizations' AI deployments. There's some good data in the report. And of course, we all expect AI to be a steadily rising flood across the board. There's a big difference, however, between organizations that sell AI and those who buy it and the real battles on the user side. For consumers, we have personal AI assistance agents, AI

assisted search and AI assisted visual design and many of those will end up inside your phone and existing search engines and apps. The big prize is the enterprise and that is in its early stages at best. While there is universal interest in AI from boards of directors on down, finding the right use case and return on investment and finding a way to fund it is hard, which leads to a lot of proof of concept projects as organizations learn what works and how to integrate it. When a lofty goal like artificial general intelligence or super intelligence gets traction, it's useful to define a roadmap that at least tries to characterize relevant milestones. For driverless cars, six levels were defined by the Society of Automotive Engineers.

Level 0, no automation. Level 1, driver assistance. Level 2, partial automation. Level 3, conditional automation. level two, partial automation, level three, conditional automation, level four, high automation, and level five, full automation. It also has shown just how hard it is to get there. Well, OpenAI has come out with a new scale to track AI's progress in general. It probably would have been better if an industry group had done it so the community's perspective was included, but it's a good start. Now, we should mention that Google's Deep Mind had a paper last December that took a bigger picture and therefore more complex approach and proposed

multiple scales based on levels of performance, generality, and autonomy. Yeah, according to a report in Bloomberg, what OpenAI announced is five levels of AI. Level one is conversational AI, which is where they think we are now, despite common hallucinations. Level two is reasoners, human-level problem solving. Level three, agents, AI systems that can take action. Level four is innovators coming up with new ideas and inventions. And level five AI organizations. This is AI systems that can do the work of an organization. It's interesting, Shaheen, that after level four, they don't keep going towards AGI and super intelligence, but instead they go sideways across multiple functions of an organization,

what the cryptocurrency world calls decentralized autonomous organizations. We also have to wonder how long OpenAI thinks it will take to achieve these various levels of AI. Now, we've been watching especially AI chip companies for several years now to see how smaller entrants carve out a niche for themselves, besides the big players, NVIDIA, AMD, Intel, and then Qualcomm, Apple, and Samsung. There's been a group of up-and-coming chip vendors like Cerebras, Samba Nova, Graphcore, Grok, Mythic, Untethered, Etched, TenStore, and on and on. There was news last week that AI chip company Graphcore has been acquired by the Japan-based SoftBank investment term for an estimated $600

million. We have heard that Graphcore has struggled in recent years. According to Reuters, the company was valued at $2.8 billion in 2020, but a filing published last year revealed Graphcore needed more cash to break even after cutting its headcount by a fifth to about 500 employees and shutting down several operations around the world. Industry consolidation is bound to happen with companies that quietly disappear, probably selling their intellectual property at a discount or through mergers and acquisitions. Sometimes it's a capability buy, like when Grok acquired Maxeller, a strong FPGA solutions company. And sometimes acquisitions are primarily aimed at the target company's staff, the acqui-hire process. In any case, SoftBank seems like a great destination for GraphCore.

You've heard the saying, go big or stay home. Well, SoftBank pretty much invented the go really super big approach to investing, and it's always worth watching. 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 Inside HPC. Shaheen Khan and Doug Black host the show.

Every episode is featured on InsideHPC.com and posted on OrionX.net. Thank you for listening.