Feb 1, 2024 · 41m · mad
How Nomic AI Is Driving The Open Source Revolution
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The MAD Podcast, host Matt Turck interviews Nomic AI founders Brandon Duderstadt and Zach Nussbaum as they unveil Nomic Embed and discuss their mission to democratize artificial intelligence through open-source models, data visualization, and fully auditable datasets.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 23.3% of the talking time here. How this is scored →
Matt as informed peer 4.1 Guest teaching 3.1 Guest disagreement 1.7 Matt pushing back 1.9
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 41:15 Brandon rejects host's AI talent categorization
Brandon explicitly rejects Matt's distinction between classic AI engineers and generative AI native talent, declaring the premise a false dichotomy.
Hardest push from Matt ▶ 38:56 Matt pushes back against SF vs NYC rivalry narrativeMatt refuses Brandon's critical framing of San Francisco's tech scene, asserting that startup ecosystems do not need to be viewed as a zero-sum contest.
Biggest teaching moment ▶ 6:33 Brandon corrects the overnight success mythBrandon educates Matt on the reality of model building, explaining that GPT4All's rapid success was actually supported by a year of foundational work on Nomic Atlas.
Matt holds his own ▶ 23:45 Matt references past vector database founder interviewsMatt demonstrates deep industry domain knowledge by recalling previous podcast conversations with vector database CEOs who framed text embeddings as a solved issue.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Founding Nomic AI and Academic Dropout Stories | 3 | 2 | 1 | 1 | Matt shares a VC heuristic about PhD dropouts, while Brandon gently clarifies a factual mix-up regarding who dropped out of NYU versus Johns Hopkins. | |
| Academia to Industry Flight and GPT4All Origins | 2 | 4 | 2 | 1 | Brandon pushes back on Matt's assumption that GPT4All was merely a weekend success, detailing how a year of underlying Atlas development laid the groundwork. | |
| Defining GPT4All and the Open-Source Ecosystem | 4 | 3 | 1 | 1 | Matt asks for a layman definition of quantization and accurately summarizes it as model weight compression, demonstrating good intuitive understanding. | |
| Health and Vibrancy of Open-Source AI | 2 | 2 | 1 | 1 | Matt jokes about VC funding actually helping open source development and asks Brandon about their hands-on contribution to llama.cpp. | |
| Exploring Nomic Atlas and Business Strategy | 5 | 3 | 1 | 1 | Matt poses a sharp go-to-market question distinguishing open-source community traction from enterprise monetization strategies. | |
| Announcement and Details of Nomic Embed | 6 | 3 | 1 | 2 | Matt demonstrates industry knowledge by citing past interviews with vector database CEOs who previously treated embeddings as a solved problem. | |
| Demystifying RAG and the Importance of Open Data | 6 | 3 | 1 | 4 | Matt directly challenges prevailing industry habits by asking if withholding training datasets is open-washing marketing rather than true open source. | |
| Benchmarking Standards and Product Ecosystem Vision | 4 | 3 | 1 | 1 | Matt asks about benchmark gaming practices and probes how Nomic integrates three distinct offerings into a coherent long-term strategy. | |
| Building an AI Startup in New York City | 5 | 5 | 6 | 5 | Matt rejects Brandon's anti-San Francisco framing as unnecessary zero-sum thinking, while Brandon explicitly shuts down Matt's generative versus classical AI question as a false dichotomy. |
Statements from this episode (16)
Duderstadt: Dataset curation unlocks AI model utility in high-stakes domains
“When you're working in these domains that are, you know, like defense and medicine and finance, where it's like super high impact and you really have a vested interest in your models being right, you start to learn very quickly that a lot of the A lot of the u…”
Turck: PhD dropouts make strong startup founders due to intellectual impatience
“I think that's a really interesting it's a really interesting pattern that I've seen across many founders to be PhD dropout. So meaning having the fundamental intellectual caliber and intellectual curiosity of doing your PhD, but not the patience of lasting th…”
Duderstadt: GPU scarcity in academia is driving talent flight to industry
“And even nowadays, like, you know, it's increasingly hard to get those GPUs in academic positions. And so it's making increasing amounts of sense for, I think the flight that we're seeing from academia to industry.”
Nussbaum: Accessibility drove GPT4All's success more than model quality
“I think the access part of it was the bigger part versus the actual model itself. Like the model itself at the end of the day, like there would suddenly trade a better model like the next week or the next month or something. And there's been many iterations of…”
Duderstadt: Mistral 7B 4-bit fine-tuned on OpenOrca is GPT4All's top model
“I think the most popular model right now is Mistral-Seven-B quantized to four-bit precision fine-tuned on the open ORCA dataset set..”
Duderstadt: Nomic AI hires full-time staff to work on llama.cpp
“We've been able to actually, like, bring staff on full time to work on things like improving lawless CPP.”
Duderstadt: Spatial scatter plots excel at AI debugging because errors cluster spatially
“What we found is this sort of layout is particularly useful for debugging AI models because their errors tend to pop up in like localized regions of it.”
Duderstadt: Nomic's enterprise sales funnel is literally Twitter, Discord, then sales
“Our sales funnel is literally like Twitter to discord to enterprise sale.”
Nussbaum: Nomic Embed is first open long-context embedder beating OpenAI Ada
“So we're launching Gnomec Embed is the first open source reproducible long context text embedder that beats OpenAI ADA as well.”
Duderstadt: Nomic releases Nomic Embed model weights and curated training dataset
“Is we're not just releasing the model weights. We're releasing the data set that Zach spent all of his time curating.”
Duderstadt: No top benchmark AI embedding models release their datasets
“And as far as we're aware, none of the other, you know, top models on any of these benchmarks have gone out and released their data sets.”
Duderstadt: Training data is almost always the secret sauce of AI models
“The data is often the secret sauce. Like when people talk about like, oh, what is the secret sauce of your AI model? It's almost always the data.”
Duderstadt: Unequal access from market consolidation is the top AI risk
“What I would say the number one risk of this technology is, which is the sort of unequal access element to it. Like the version of the world that I fear most is like, there's two or three mega companies that like lock everything down in like the early days.”
Duderstadt: New York is a strong candidate for global AI nexus
“If I were to bet on where like the global nexus necessarily would be, like if you're accounting for the fact that like, you know, there are things happening outside of the US, I think New York is a very strong candidate.”
Duderstadt: Nomic AI recruits heavily directly from its open-source Discord community
“And also we hire a lot out of open source. And so if people want to get involved, like, you know, we hired Zach out of open source you know, a couple of people that we picked up during the big hiring spree after we raised our A. You know Adam, Aaron Jared. The…”