AI Top-of-Mind for 6.10.24 — Zombie Students
Today: Zombie students, AI for SD-WAN, thermodynamic computing, effective image prompting, more on agents, and movie marketing AI fails
Top-of-mind would have to be Apple’s WWDC and their ‘Apple Intelligence’ announcement, but that will need to wait a day. So, on to education, and some sobering analysis regarding on-line coursework and AI. It isn’t only students leveraging bots for coursework, but the possibility that a portion of the students are not real at all. More from Mary Rose.
Not many networking postings, but this one from Network World caught my eye, since so many vendors try to AI-wash their capabilities. From the article:
· As part of the AI platform rollout, the vendor introduced Marvis Minis, which simulate end user, device, and application traffic to learn the network configuration and continuously feed data back into the Mist AI engine. The Minis software works by setting up a digital twin of a customer’s network environment to simulate and test user connections, validate network configurations, and find/detect problems without requiring any additional hardware, according to Juniper.
· “Marvis Minis for SD-WAN adds and performs always-on, automated speed tests to ensure that the network is delivering as promised — even if users aren’t present,” wrote Jeff Aaron, vice president of product marketing, in a blog about the new features. “Now operators can resolve SD-WAN issues before users even connect.”
We read a lot about AI energy consumption, but is there a more efficient approach? Two rather technical articles, the first describing what is termed ‘thermodyamic computing,’ and the second on ‘neuromorphic’ architecture. Sam Vaseghi in ‘AI Advances’ writes:
· Thermodynamic computers are designed to utilize the natural behavior of physical systems towards energy minimization and entropy production as a means of computation.
· Thermodynamic computing is particularly suited for dynamic and complex systems where traditional approaches are inefficient or infeasible. This includes fields like neural networks, pattern recognition, and systems biology, optimization, ML tasks, and AI.
But the reality is that thermodynamic computing is still at the research level, but so was quantum a decade ago.
Then to neuromorphic and the following update as captured in ‘EE Times:’
A new neuromorphic supercomputer is claiming the title of world’s biggest. University of Dresden spinout SpiNNcloud, formed to commercialize technology based on a second generation of Steve Furber’s SpiNNaker neuromorphic architecture, is now offering a five-billion neuron supercomputer in the cloud, as well as smaller commercial systems for on-prem use. Among the startup’s first customers are Sandia National Labs, Technische Universität München and Universität Göttingen.
The diagram below helps place the three forms of compute in perspective.
As background, the Wiki entry on the technology.
Sticking to hardware, ‘EE Times’ coverage of Computex and how both Intel and Nvidia are evolving their offers. Pat Gelsinger announced Xeon 6, Lunar Lake, and described Ultra Accelerator Link, while Jensen Huang countered with NVLink.
Turning to models, are you using ChatGPT correctly? Andrew Best writing in ‘Artificial Intelligence in Plain English’ describes how less is more for prompting. His recommendation is to start with less, and then add detail until things go off the reservation.
In parallel, how do you properly build and deploy AI agents? Patrick Dougherty offers his thoughts. From the post:
How are telcos leveraging AI? We’ve read about deployment within basic customer support, but Verizon is taking the next step by integrating the technology into sales processes. Fierce Network details their plans that include personalized marketing via ‘Verizon Personal Shopper,’ a customized LLM with Verizon’s institutional knowledge that is based on Google’s Contact Center AI, and finally a tool called ‘Fast Pass’ that puts the customer in touch with a specialist depending upon their needs.
On the society front, Alberto Romero brings to the table ten questions about AI that we should all be asking, and potentially have a position. I’m not sure I agree with all his points, and a good exercise would be to look at each and offer your view.
Lastly, looking into creative, or lack thereof, I finally had the opportunity to watch the movie ‘Civil War.’ Very disturbing given current events. But there was a controversy with the marketing of the movie, where the creators used AI to depict various US cities in a state of destruction, with scenes that were not even in the film.
