Dec 8, 2023 · 30m · saastr
The Where, When, and How of AI with Theory Ventures, Open AI, MotherDuck and Lamini
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
Hosted by Theory Ventures founder Tomas Tunguz, this panel featuring executives from MotherDuck, Lamini, and OpenAI explores how generative AI is transforming data architectures, enterprise model customization, and developer workflows. The leaders provide practical insights on deploying specialized enterprise models, leveraging natural language data tools, and adapting business operations for the future.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
Jason as informed peer 5.0 Guest teaching 3.3 Guest disagreement 0.3 Jason pushing back 0.4
speaking balance: gold is Jason, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 6:11 Sharon reframes model development from researchers to domain experts
Sharon directly counters the conventional assumption that top AI researchers dictate model value, asserting that domain practitioners will steer future model trajectories.
Hardest push from Jason ▶ 21:13 Tomas challenges sales org assumptionsTomas pushes past high-level productivity claims to ask whether the core structural division between SDRs and AEs is rendered obsolete.
Biggest teaching moment ▶ 15:45 Sharon clarifies production constraints over research benchmarksSharon educates the audience and panel on how ML engineers' benchmark accuracy metrics clash with production software engineering realities like API latency.
Jason holds their own ▶ 13:50 Tomas maps out the structural data stack evolutionTomas demonstrates deep industry domain expertise by laying out the 10-year architectural shift from disconnected data lakes to unified CI/CD machine learning pipelines.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Jason as informed peer | Guest teaching | Guest disagreement | Jason pushing back | Why |
|---|---|---|---|---|---|---|
| Panel Introductions and Setting the AI Agenda | 5 | 2 | 0 | 0 | Tomas establishes the agenda and introduces the panel, adding relevant context on how LLMs democratize data querying and citing Microsoft OpenAI customer growth statistics. | |
| Enterprise Fine-Tuning and Domain Expert Customization | 5 | 4 | 1 | 1 | Tomas frames the enterprise trade-off between vendor tuning and in-house subject matter experts. Sharon explains the depletion of public training data and clarifies why domain experts will drive enterprise models rather than pure AI researchers. | |
| Launching ChatGPT Enterprise and Commercial AI Enablement | 3 | 5 | 0 | 0 | Tomas yields the floor to Maggie to detail the launch of ChatGPT Enterprise. Maggie delivers a thorough breakdown of enterprise privacy requirements, cross-industry design partners, and tactical use cases like Code Interpreter. | |
| Merging Machine Learning Workflows with Production Software | 6 | 4 | 1 | 1 | Tomas articulates how the historical 10-year data stack is converging with CI/CD and ML production pathways. Sharon enriches this with an Uber deployment anecdote and the friction between benchmark accuracy and production latency. | |
| AI-Driven Sales Evolution and Strategic Value Selling | 5 | 4 | 0 | 1 | Maggie outlines the chronological evolution of B2B sales from mass outreach to value selling. Tomas probes whether AI will dismantle the traditional SDR and AE organizational structure. | |
| Daily Generative AI Workflows and Image Creation | 6 | 1 | 0 | 0 | The panel discusses tactical day-to-day tooling, where Tomas shares personal anecdotes regarding image copyright litigation and measurable click-through improvements from LLM-generated titles. | |
| Future Industry Predictions and Educational AI Transformation | 5 | 3 | 0 | 0 | Guests offer three-year forward predictions across data job reshuffling, scalable domain experts, and educational disruption, with Tomas reinforcing the educational power of patient AI tutors. |
Statements from this episode (15)
Tunguz: LLMs replace 3 to 12 months of SQL training for data analysis
“It's really much easier for someone who's maybe not be as sophisticated with SQL to suddenly start asking and answering questions of their data in ways that would have taken them three, six, 12 months to learn beforehand.”
Tunguz: Microsoft OpenAI customer count tripled in recent quarter
“You look at the total number of customers that are starting to use Microsoft's OpenAI technology has three X, I think, in the last quarter in the reporting.”
Zhou: Public training data for LLMs is running out
“Yeah, I think even just zooming back out at a technical level, I think actually public data is running out for all, all of what LLMs can take advantage of.”
Zhou: Best LLMs of the next wave will be enterprise models
“So actually the next frontier for LLMs is in enterprises, and I believe the best LLMs for this next, next wave essentially will be enterprise LLMs.”
Zhou: Domain experts, not AI researchers, will drive top models
“However, I believe that, and this is based on my experience training these models, it's actually the domain experts will be driving the best models out there. It won't be people like me who can actually do all the model training, et cetera.”
Hott: Customization is the top priority for ChatGPT Enterprise
“We've shared pretty broadly that the single biggest thing that we are working on right now is this idea of customization.”
Hott: OpenAI is building a self-serve ChatGPT version for SMBs
“The other thing that we are also building right now, which is really exciting is a business version. So more of a self-serve version for SMBs.”
Tigani: AI will require distinct tooling for developers, analysts, and data scientists
“Everybody uses ChatGPT in a different way, so the developer's gonna use it in a different way than an analyst is gonna use it, and which is different than a data scientist, and I think, yeah, there's gonna have to be different, different tooling for each one o…”
Zhou: ML benchmark accuracy fails in production without low API latency
“In machine learning, you know, a lot of AI people are like, yeah, we push the performance of this model, and we define performance as accuracy or, you know, accuracy along these, like, general benchmarks. Maybe it's sixth grade science questions or something. …”
Zhou: Adding images drove a 10x increase in marketing engagement
“I think one thing that, you know, we were analyzing our data when it came to all our marketing blog posts or like tweets, everything going out, and there was a 10 X increase in anything with an image, right?”
Zhou: Prompting Midjourney images matters more than writing blog posts
“And so we found that just spending some time on mid journey was worth more than spending time on the blog posts. In any single way, like hands down, just like spend a few extra minutes here, prompt engineering or really just like generating an image of your ch…”
Tunguz uses Midjourney for blog posts after being sued for image rights
“I was unfortunately sued for using an image where I didn't have the rights, and so I learned a very important lesson, and so now all, most of the images on the blog post that I write are mid-journey generated because now I can be confident that I'm not going t…”
Tunguz: Theory Ventures produced its first investment memo summary using an LLM
“We record the conversation so we don't have to ask them to repeat, and we're starting to save all that information, produce LLMs on top that summarize diligence, and we produced our first investment memo summary from an LLM.”
Tunguz: LLM-generated blog titles deliver 3x to 4x better click-through rates
“And so if I take like a blog post that I've written, and then I ask one of these models to produce like three or four suggested blog post titles, I actually get three to four times Better click-through rates on the titles, because I'm aggregating all that, tha…”
Tigani: Prompting tools will reshuffle data engineering and analytics roles
“I think the only thing that you know, I think is clear is that people will be doing things differently. Their jobs will be different. I think, for example, like The data space has sort of started to aggregate into these different job titles, analytics engineer…”