Nov 1, 2024 · 33m · y-combinator
The 10 Trillion Parameter AI Model With 300 IQ · Y Combinator
Garry Tan · 10m spoken Harj Taggar · 9m spoken Diana Hu · 7m spoken Jared Friedman · 3m spoken Sarah Friar · 22s spoken
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
Y Combinator partners analyze the implications of OpenAI's o1 model and future 10 trillion parameter superintelligent systems, examining how shifting developer preferences, near-perfect reasoning accuracy, and voice APIs are redefining startup strategy and execution.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The partners hold 98.7% of the talking time here. How this is scored →
The partners as informed peer 6.4 Guest teaching 1.8 Guest disagreement 1.1 The partners pushing back 1.9
speaking balance: gold is the partners, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 22:30 Challenging OpenAI Moat Durability
Harj directly challenges the premise that OpenAI has built lasting defensibility, pointing out they repeatedly lose their leads to Anthropic and Meta within months.
Hardest push from the partners ▶ 31:07 Garry's AltaVista Counter-AnalogyGarry pushes back on over-indexing on Cursor's rapid market dominance by reminding the table that AltaVista was also once dominant before Google emerged.
Biggest teaching moment ▶ 5:31 Diana's Fourier Transform LectureDiana educates the room on the mathematical history of Fourier transforms, explaining how 19th-century periodic equations took 150 years to enable modern telecom and image compression.
The partners hold their own ▶ 12:15 Jared and Harj Dropping Hard Batch MetricsHarj and Jared leverage proprietary YC internal survey data to demonstrate Claude jumping from 5% to 25% share while OpenAI's monopoly dissipated.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The partners as informed peer | Guest teaching | Guest disagreement | The partners pushing back | Why |
|---|---|---|---|---|---|---|
| 10 Trillion Parameters and Historical Model Scaling Parallels | 6 | 2 | 1 | 2 | Garry and Diana discuss scaling laws, comparing the jump to 10 trillion parameters to the historical shift from GPT-2 (1B) to GPT-3 (175B). Diana articulates technical scaling principles and logarithmic scaling curves smoothly in a collaborative roundtable setup. | |
| Artificial Superintelligence and the 200 IQ Benchmark | 7 | 3 | 1 | 2 | Diana provides an in-depth breakdown of Fourier transforms from 1800 to 1950s digital applications to illustrate how fundamental breakthroughs take decades to hit consumer tangibility. Garry and Harj build on this by debating whether AI's clock started decades ago or at ChatGPT. | |
| Consumer AI Tangibility, Meta Ray-Bans, and Model Distillation | 6 | 2 | 1 | 1 | Harj highlights consumer hardware tangibility like Meta Ray-Bans while Garry discusses internal model distillation (using giant teacher models to train efficient student models like GPT-4o-mini). The dialogue is highly collaborative and analytical. | |
| YC Batch Market Share Shifts: Claude, Llama, and OpenAI | 7 | 2 | 1 | 2 | Harj and Jared share empirical data from the YC Summer 24 batch, showing Claude's market share surge from 5% to 25% and Llama reaching 8%. They analyze how developer choice rapidly eroded OpenAI's monopoly before o1. | |
| Inside the YC OpenAI Hackathon: Hands-On with o1 | 6 | 2 | 1 | 3 | Diana shares direct observations from the ongoing YC o1 hackathon with Freestyle and Replit agents. Harj probes the tension between OpenAI capturing all value vs. lowering the deterministic barrier for founders. | |
| Enterprise Efficiency Gains and the Rise of Vertical AI Agents | 7 | 1 | 1 | 1 | Garry details financial metrics of a 2017 YC company cutting 60% of customer support tickets to achieve profitability without raising capital, and breaks down vertical AI wedges like TaxGPT. | |
| OpenAI's Defensibility vs. Rapid Model Convergence | 6 | 2 | 2 | 3 | Harj questions whether OpenAI's o1 breakthrough will provide a defensible moat or if Claude, Llama, and Gemini will close the reasoning gap within six months like past release cycles. | |
| The Realtime Voice API and Call Center Transformation | 6 | 1 | 1 | 2 | Garry analyzes the macroeconomic impact of OpenAI's $9/hr Realtime Voice API on call-center economies, while Jared and Diana cite batch successes like Happy Robot and Domu. | |
| The Rise of Cursor and the New Paradigm of AI Coding Assistants | 7 | 2 | 1 | 2 | Jared cites batch statistics showing Cursor taking 50% share over GitHub Copilot's 12%. Garry pushes a cautious historical analogy to AltaVista, noting early lead advantages don't guarantee permanence. | |
| The 10 Trillion Parameter Future: Accelerating Scientific Progress | 6 | 1 | 1 | 1 | Jared lays out the steel man thesis for 10T parameter models: applying infinite compute and reasoning power to unread scientific literature to unlock breakthroughs like room-temperature superconductors. |
Statements from this episode (23)
Friar: OpenAI's capital spending prioritizes compute first, then top talent
“It's compute first and it's not cheap. It's great talent second. And then of course it's all the normal operating expenses of a more traditional company.”
Friar: OpenAI expects each successive frontier model to scale 10x
“But I think there is no denying that you are, we're on a scaling law right now where orders of magnitude matter. The next model is going to be an order of magnitude bigger and the next one on and on. And so that does make it very capital intensive.”
Hu: Leading frontier AI models operate around 500 billion parameters
“Yeah, for a bit of context right now, the frontier models, I mean, they're not public, exactly how many parameters they have. But they're roughly in the 500 of billions-ish, like Lama-III, four or five billion. Anthropic is speculated to be five hundred billio…”
Hu: 10T parameter models will spark a GPT-3 level innovation leap
“I think the type of level of potential innovation could be the same leap we saw from GPT-II, which was around one billion Parameters that was released with the paper of a scaling loss, which was one of these seminal papers that people figure out, okay, this is…”
Tan: AGI is effectively already here for most knowledge worker tasks
“Like you could make a strong case that AGI is basically already here. The majority of the tasks that, you know, 98% of knowledge workers do day-to-day it is now possible for a software engineer probably sitting in front of cursor, to write something that gets …”
Taggar: AI smart glasses with human-level voice will be world-changing
“Like, I think consumers, once this becomes something that's, like, visual in your, like, smart glasses, plus, like, a voice app that you can talk to, and it, like, is indistinguishable from a human being, like, that's gonna be a real change the world moment fo…”
Tan: OpenAI enabled internal model distillation as a developer lock-in strategy
“OpenAI itself has now enabled distillation internal to its own API. So you can use O-one, you can use even GPT-IV or IV-O to distill it down into a much cheaper model that's internal to them, like GPT-IV, IV-O-Mini. And that's sort of their, you know, lock-in …”
Taggar: Claude jumped from 5% to 25% market share in YC startups
“And some of the stuff that really stood out is Claude has, even just in six months, from the winter batch to the summer batch, has gone from, like, five percent developer market share to, like, 25%.”
Taggar: Llama market share in YC startups grew from 0% to 8%
“LARM has gone from zero percent to eight percent.”
Friedman: Software adopted by YC startups predicts future global tech winners
“One thing that we know from running YC for a long time is that whatever the companies in the batch use is a very good predictor of what Like the best companies in the world are using, and therefore what products will be most successful. A lot of YC's most succ…”
Taggar: 15% of the YC batch adopted OpenAI's o1 before public release
“It seems like, or, 15% of the batch are already using O-one, even though it's not, like, fully available yet.”
Hu: A YC startup built a Replit Agent clone in hours using o1
“They basically got a version of Replit Agent working with the product. All they had to prompt O-one with was all their developer, actually, some of their developer documentation and some of their code, and they could just prompt it, build me a web app that wri…”
Taggar: 100% AI accuracy will create winner-take-all software markets
“If they just on day one, you could be guaranteed a hundred percent accuracy, like as though you're just building a web app on top of a database, the barrier to entry to build these things goes way down. It's gonna be more competition than ever. And then it wil…”
Hu: YC startup DryMerge reached near-100% accuracy by adopting OpenAI's o1
“There's a company Dry Merch. That you work with, and they went from 80% accuracy to pretty much a 99 or for intents and purposes, a hundred percent using a one and they unlocked a bunch of things.”
Tan: $50M ARR YC company automated 60% of support to reach profitability
“They knew that they needed to raise more money, but in the year since, they automated about 60% of their customer support tickets. And they went from something that needed to raise another round imminently to something that was totally cash flow break even whi…”
Tan: Foundation model giants won't build vertical applications due to inefficiency
“If you treat OpenAI as the Google of the next 20 years, you want to invest in Google and all the things that Google enabled, like Airbnb. Google could do Airbnb, it probably won't. Just from, like, I don't know, Coase's theorem of the firm, probably. It's just…”
Taggar: OpenAI consistently pioneers breakthroughs but never maintains its competitive lead
“OpenAI seems like it is continually the one pushing the envelope, but they always seem to be The first ones to make major breakthroughs, but they have never been able to maintain the lead so far.”
Hu: OpenAI's o1 massively increases GPU compute requirements for inference
“I think the other thing that's interesting about one is that it makes a lot of the GPU needs even bigger because it's moving a lot of the computation needs a lot higher for inference because it's taking a lot more time to do a lot of the inference. So I think …”
Tan: OpenAI's $9/hour real-time voice API threatens call center-reliant national economies
“The Ongoing usage-based pricing is nine dollars per hour, and it sort of points to a sort of powerful thing. Like, if I were a macro trader, I would be very, very bearish on countries that have, that rely very heavily on call centers right now because, you kno…”
Hu: AI has effectively passed the Turing test for phone calls
“At this point, AI has passed Turing tests and is solving all of these very menial problems over the phone.”
Friedman: AI is the fastest-improving technology in human history
“It's the fastest any tech has ever improved. I think. Yeah. Certainly faster than processors, certainly faster than the cloud.”