Where's the AI?
New meet-up group, Sonoma AI, shows there's more real tech going on in Sebastopol than you might think
After Christy Bergman gave her presentation at the first Sonoma AI meetup, held at Punchdown in the Barlow, a man sitting up front asked the question: “Where’s the AI?” Bergman, a freelance AI engineer, had explained the technical process of grabbing two different sets of conversations (from Discord servers) and then processing them and producing a graph that clustered a common set of topics that people had been talking about. She talked about how she used to do it before AI and how she was doing it now using AI, and how AI made several different steps in the process much easier for what she wanted to do.
The man who asked the first question commented that he could see what she was doing but where was the AI? If there was AI, he wondered, why would she have to do anything at all? Bergman offered an answer but the man wasn’t buying it. He wanted see where the magic was.
The Sonoma AI group came out of a discussion among a group of people having dinner at Handline a while back. A few were former O’Reilly employees like Paco Nathan and others like Christy Bergman, who work remotely on AI projects. Nathan told me the small group saw a need to help connect people working on AI in Sonoma County. Their idea was to organize a meetup like one might find in San Francisco, New York or Washington, D.C. About 30 people (10 women and 20 men, according to one person’s count) showed up to sip natural wine and talk about AI — along with many other acronyms.
On Tuesday, I attended a MYN meeting at the Sebastopol fire department. One person (actually my wife, Nancy) said that she was working on creating SOPs to describe basic MYN processes. Another woman interrupted her — “What’s an SOP?” acting like she’d been hit by a blast of bad air. “Standard Operating Procedure” was the reply from several people in the room. The woman was grateful for the answer but she shook her head, mumbling that she didn’t care much for acronyms, even though the meeting was convened under the auspices of an acronym—MYN, which once meant “Map Your Neighborhood” but now means “Meet Your Neighbors.”
I thought of her at this AI meeting because she might have really lost it after hearing a steady barrage of acronyms — LLM, NP, NER, NLP, UBO, ICIJ, GPU, and RAG, to name a few. Most of the people in the audience nodded knowingly through the technical jargon. For many, acronyms are like a sampler box of See’s chocolates; you wonder what’s inside each chocolate and find out by taking a bite.
I’ll try to explain the first of those acronyms. LLM stands for Large Language Model, a neural network that is a “type of Artificial Intelligence that can process, understand and generate human language.” That’s how AI explains itself anyway. It’s why you can ask ChatGPT a question and get back several paragraphs of text as an answer, which leaves you either amazed or upset that computers can think like humans.
AI takes a lot of math and computation, and that kind of processing power requires larger data centers and demands more energy. That’s why Sam Altman, the head of OpenAI, in recent interviews has said that the US needs to be better at building infrastructure, which is needed to keep up with AI.
A small but important piece of that infrastructure is being built here in Sebastopol at Luma Optics. Eric Litvin is president and co-founder of the company that has two warehouses in town. (Who knew?) His company sponsored the Sonoma AI meetup, and he spoke at the beginning, saying that such meetings can help people find jobs. He said that Luma Optics produces “optical AI fabrics” that are used to interconnect computers in 37 GPU-based data centers around the world. GPUs are Graphical Processing Units — the brains of computers. They are produced by Nvidia who originally developed them for graphics processing, but their ability to do fast mathematical calculations has made them the best choice for training LLMs (large language models — got it?). Nvidia surpassed Apple last year with a $3 trillion market cap.
You might be curious to know what does all this AI do? Paco Nathan gave a fairly compelling answer — it can help track down bad actors; it can help with fraud detection; it can track where dark money comes from and where it goes. Nathan, who was formerly the Director of Learning at O’Reilly, still lives in Sebastopol but works for a company called Senzing. Almost nobody has heard of the company, he said, but when you register to vote, for instance, they are involved in the process. Senzing’s website says that they work “with the largest government, financial services & healthcare organizations.” Nathan said, introducing his talk, “Money crosses borders; laws don’t.”
Nathan talked about “entity resolution,” which is what Senzing specializes in. You could think of an entity as a person or an organization or a particular location. Given all kinds of data, mined from all kinds of sources, how can you tell if two people with similar names are the same person or not? Or whether two addresses expressed in two different ways are actually the same address. Presumably, the government wants to know that the name and address that you gave them to register to vote is the same name and address that you used on your driver’s license or your tax form. The AI that Nathan describes might answer that question by calculating the probability that two entities are the same or not. “Fraudsters are trying to hide behind small data errors,” he said, explaining that mistakes in names might be intentional.
If you were interested in learning more about tracking down dark money, Nathan recommended a few books and the Netflix movie “Laundromat.” He showed a fascinating slide with an example of the “Azerbaijani Laundromat,” where $2.9bn dollars was funneled through four UK shell companies using the Danske Bank in Estonia and used to bribe politicians and pay for private schools.
What AI is doing — my explanation, not Nathan’s — is helping to follow the money and connect the people and institutions that are involved in a web of corruption. Nathan’s acronym was RAG, a new one for me, “retrieval augmented generation,” and he referred to a visualization called Graph RAG (a nice big chocolate caramel). I can’t explain it to you. However, I could follow the slide which showed three separate clusters (think different shell companies) who seem disconnected from one another.
Nathan then showed a slide with the connections among each of the different nodes. In other words, criminals are good at covering their tracks and hiding connections. and AI is being used to produce a map of how they are connected, a kind of MYN for criminal associates.
When Nathan was done, the man in the front row asked his same question: “Where’s the AI?” Nathan smiled kindly, explaining that he first got into AI in the 1980s at Stanford and that he preferred a definition of AI from that era. “AI is teams of people and machines working together to do things that they couldn’t do by themselves,” said Nathan. It was a nice, thoughtful reply, but the man wasn’t buying that answer either. He came to see the magic, and he just doesn’t see it demonstrated.
The last presenter was Allison Ding, who works as a Senior Developer Advocate at Nvidia, the AI hardware superstar company. She came up from Silicon Valley to explain the developer ecosystem that Nvidia provides and said that “data processing for LLMs needs accelerated computing.” She used the acronym SOTA as in “SOTA models.” It confused me but I saw later on that it simply meant “State Of The Art” (a chocolate-covered cherry).
As you might expect, and the audience did expect this by now, the man in the front row asked the Nvidia representative: “Where was the AI? I don’t see it.” It reminded me of the Wendy’s catchphrase that Walter Mondale asked Gary Hart in a debate: “Where’s the beef?”
Allison responded shyly, saying that it can be difficult to explain what AI is. She still struggles with it, even as she is working on her second Masters. “Well, AI is combined with a lot of things,” she said. “Developers are using automated tools to build a pipeline.” The man shook his head. Nope, that’s not what he wanted to hear. The meeting and patience with the question was running out. Christy Bergman intervened, closing the session: “We can debate during happy hour what AI really means.”
On the way out, I met Steven and Elenni, a pair of young developers who are living in Sebastopol. These “good actors” were really excited to talk AI and meet other technical people who live and work in the area. Making connections whether with AI or face-to-face seemed to satisfy most of the people there.
You can join a LinkedIn group - https://www.linkedin.com/company/sonoma-ai-with-wine
Great article. As a self described Luddite Sebastopol I am always amazed and grateful to learn about