AI Top-of-Mind for 7.8.24: NIM
Today: Nvidia NIM, five levels of AI agents, AI for mental health, more on power consumption, and 100,000 cats
Top-of-mind for today is Nvidia’s ‘NIM.’ Suhaib Arshad writing in ‘Towards AI’ offers a deep dive on its capabilities including text generation and multimodal, as well as how to deploy. From the post:
Nvidia Inference Microservice (NIM): In simple terms, NIM is a collection of cloud-native microservices that help in deployment of generative AI models on GPU-accelerated workstations, cloud environments, and data centers. They reduce the overall time taken for a generative model to reach the market, hence streamlining the entire process from development to production for enterprises.
Next is ‘McKinsey’ with a new podcast from Reid Hoffman. From the abstract:
The advent of steam power in the late 18th century utterly transformed manufacturing, transportation, and construction. A new kind of upheaval is already under way — one that will energize all language-based capabilities, including communication, reasoning, analysis, sales, and marketing. In this episode of the At the Edge podcast, Reid Hoffman, a partner at venture capital firm Greylock Partners, and cofounder of LinkedIn and Inflection AI, speaks with McKinsey’s Lareina Yee about the generative AI revolution and how it can teach users to understand and harness its power.
I’ve written about AI agents a few times in the past, so an update. Corbus Greyling looks at the levels of agents based on capabilities and use cases. From the post, pointing out in red where we are today:
Closer to home on healthcare, the ‘Mercury News’ reports on a company called ‘Dysolve’ and its impact on dyslexia. From the article:
· Dysolve began testing patented technology in 2014 to reprocess students’ brains and transition them out of special education, and by 2023 expanded nationwide to individual subscribers and schools. The technology tailors its program to each individual’s learning needs and abilities, making it the only AI system for diagnosing and correcting dyslexia.
· Q: How does Dysolve work?
· A: The whole program is game-based. We recommend that students use it five days a week and it’s 15 to 30 minutes a day. In the first month, you really get to know what the processing difficulties are and at what level. The major issues would be identified in the first two to three weeks of evaluation. The AI system is really trying to figure out how the child is processing language. For example, can the child pull out a particular sound from a word, even if the child doesn’t know how to spell that word? The system wants to find that out and uses very, very small pieces of processes to locate where the problem is.
· Q: What is the benefit of using an AI-generated program versus a traditional method for identifying learning disabilities?
· A: A traditional evaluation is between $5,000-$10,000 per child. This is AI doing the work. Every session that a child has with the AI system is evaluation, whereas with schools, they can only do it one time at the beginning. And that’s why there is a disconnect between where the child may be and what new issues may have come up. Dysolve can track all of that and that’s why it’s so much more efficient, that’s why it’s cost effective because you can get children out of special education.
An update on the AI power consumption front. A ‘CNN’ opinion piece looks not only at the expected demand, but what this could mean to all our power bills due to the increasing use of ‘peaker plants,’ as well as what solutions we require. From the article:
· Within the decade, we’re looking at a grid infrastructure that can’t keep pace with rapidly rising demand, even with all sources running around the clock. This pushes us toward a scenario unimaginable for most Americans — an unreliable energy grid, marked by blackouts and rolling brownouts. Recall the blackouts across Texas in winter 2021 and the 2003 East Coast blackout that impacted 50 million people. But this time, it won’t be a stroke of bad luck linked to extreme weather or a safety incident. Electricity rationing will become a standard part of American life while the grid races to add capacity.
· A truly national grid demands a singular federal oversight body. Such a consolidation of authority is proposed in the Streamlining Interstate Transmission of Electricity (SITE) Act. This, in concert with a nationwide push to build and finance new electricity generation and transmission — on the scale of New Deal era construction — is needed to meet the energy demand of an AI-powered, green economy.
Last year, the NY Times looked into this same issues, and published a great diagram of the task at hand:
Lastly, what happens when you use AI to generate 100,000 cat images? Wei Mao publishing in ‘Bootcamp’ used this to understand how humans and AI can collaborate creatively. From the post:
But what exactly is creativity? It’s about finding a point far outside your current thinking and then connecting it back to your existing knowledge. Thinking outside the box defines creativity, and AI excels at identifying distant points in its extensive models.
An example from the article showing the use of ‘-sref’ as applied to cats:
