Jun 14, 2024 · 58m · more-or-less
#51: Vinod Khosla on What to Build in AI · More or Less Podcast
Vinod Khosla · 25m spoken Sam Lessin · 10m spoken Jessica Lessin · 9m spoken Dave Morin · 4m spoken Brit Morin · 2m 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
In this episode of the More or Less podcast, hosts Jessica Lessin, Sam Lessin, Brit Morin, and Dave Morin interview veteran venture capitalist Vinod Khosla to discuss OpenAI's surging revenues, Apple Intelligence, falling compute costs, disciplined seed investing, and the macroeconomic impacts of AI.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 49.5% of the talking time here. How this is scored →
The hosts as informed peer 5.7 Guest teaching 3.7 Guest disagreement 3.1 The hosts pushing back 3.7
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 7:36 Vinod dismisses the compute cost premise
Vinod directly rejects Sam's line of reasoning about balancing compute expenditure across different model tiers, bluntly advising all startups to ignore compute costs entirely.
Hardest push from the hosts ▶ 8:42 Sam challenges model cost deflation with training realitiesSam interrupts Vinod's deflationary thesis by highlighting that successive frontier models require an order of magnitude more capital to train, directly contesting the premise that compute can be ignored.
Biggest teaching moment ▶ 8:47 Vinod explains frontier model usage amortizationVinod breaks down the structural economics of foundation models, explaining why high upfront training costs compel providers to maximize inference volume and data capture rather than short-term revenue.
The host holds their own ▶ 20:55 Dave exposes Apple's restrictive developer sandboxDave demonstrates authoritative platform knowledge by exposing that Apple has wrapped local models inside Swift UI components without granting developers raw access to on-device weights.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Podcast Opening Title and Theme Song | 5 | 3 | 2 | 3 | Jessica opens by questioning Apple's AI platform strategy and whether OpenAI exclusivity will hold given mentions of Gemini and Anthropic. Vinod largely agrees on Apple's strategic optionality while reframing the milestone around computers learning human intent directly. | |
| Small vs. Large Models and Compute Economics | 6 | 6 | 5 | 6 | Sam probes the tension between expensive PhD-level big models and on-device small models, and Vinod counters by advising startups to completely ignore compute costs as market forces drive prices down. Sam sharply pushes back on the exponential rise in training cluster costs, prompting Vinod to detail the five-year data feedback loop. | |
| Vertical AI Disruption and Defensible Expertise | 6 | 6 | 6 | 6 | Vinod highlights specialized venture bets like synthetic structural engineers, while Sam challenges whether domain expertise is actually defensible once codified into commodity models. Vinod firmly disagrees with Sam's analysis of general practice medicine versus oncology, arguing specialized workflows create durable value. | |
| Dave Morin on Apple's Platform Strategy and Developer Ecosystem | 7 | 0 | 0 | 0 | Dave delivers an extended monologue analyzing Apple's launch, highlighting the rare execution trifecta of platform, product, and technology while critiquing the restrictive Swift wrapper that locks developers out of raw local model weights. Because this is a host-only platform breakdown without guest participation, guest and pushback scores are zero. | |
| Venture Capital Discipline and AI Seed Valuations | 6 | 4 | 2 | 4 | Sam lays out his skeptic case that seed-stage AI investing is broken by overfunded spinouts, finding rare alignment with Vinod who criticizes non-technical asset managers and ungrounded seed valuations. Vinod outlines his disciplined framework where 80 percent of his seed deals price under twenty-five million post. | |
| Everyday AI, ChatGPT Prompts, and Autonomous Strawberry Farming | 5 | 4 | 3 | 5 | The hosts and Vinod compare everyday ChatGPT workflows, transitioning into Vinod's portfolio company Zordi growing autonomous greenhouse strawberries. Sam repeatedly attacks the long-term margin and edge of boutique farming against agricultural giants like Monsanto, which Vinod brushes aside. | |
| Deflationary Economics, Content IP, and Cultural Value | 5 | 5 | 4 | 4 | Jessica asks about the legal standing and economics of training on copyrighted IP, which Vinod reframes into a broader macro thesis that AI will drive labor costs and expert fees to zero. When Sam suggests hiring humans as conversational data pumps, Vinod educates him on how value shifts entirely to experiential cultural phenomena like Taylor Swift. | |
| Silicon Valley Politics, Regulation, and the 2024 Election | 5 | 5 | 6 | 5 | Vinod articulates a sharp separation between traditional Republicans and MAGA extremists, openly condemning Trump's ethics while assessing Silicon Valley donors. When Jessica suggests the current administration has taken an overly heavy regulatory hand with tech, Vinod rejects the premise by praising bipartisan AI policy and lambasting sensationalist headline writers. | |
| Co-Host Post-Interview Debrief and Cultural Recommendations | 6 | 0 | 0 | 0 | In an executive post-interview debrief after Vinod logs off, the co-hosts unpack his assertions regarding compute deflation and automated greenhouse margins before trading theater recommendations. With no guest present in the segment, guest-side scores and host pushback remain at zero. |
Statements from this episode (22)
Khosla: Apple Partnership Validates OpenAI as the Best LLM
“Obviously from an open AI perspective, it clearly establishes where open AI is in the competition for the best LLMs. Obviously many people wanted that business. And many things were evaluated, and so it's really rewarding to see where Apple thinks the best AI …”
Khosla: Small models will never substitute for large, human-level AI models
“If you want human level intelligence from model, it'll be a big model, and there'll be a couple of them around. Some better than others, some a year or two ahead of others, but I don't think small models substitute for big models, but they do some things reall…”
Khosla: AI inference costs will drop up to 10x by June 2025
“I think what will happen is the expensive PhD will become dirt cheap. My bet is a year from now, it'll cost one-fifth to one-tenth of what it does today.”
Khosla: Startups should ignore compute costs; AI software optimizations become worthless
“And my advice to all our startups is ignore the cost of compute. Because any assumption you make, any dollar you spend on optimizing your software will be worthless within the year.”
Khosla: Open-source AI models will die due to massive training costs
“It is costing an order of magnitude more to train, and that's why I don't think open source models will be viable because of the training costs.”
Khosla: AI labs must prioritize usage and training data over near-term revenue
“So I'm pretty convinced The next few years, the revenue isn't the important metric. You don't want to, of course, lose so much money you can't afford it, but you're not trying to make tons of money. You're trying to get tons of usage and you're trying to get t…”
Khosla: OpenAI executes crisply, while Google has great technology but poor execution
“I think OpenAI has been very good in execution, but Google has great technology and technologies. They just haven't been as crisp about their execution.”
Khosla: Massive incumbent tech giants will dominate the large language model market
“Large LLMs will belong to the large players, players who can run on train on very large clusters, people who can pay for proprietary content, whether it's with Reddit or somebody else.”
Khosla: Investing in startups building independent GPU cloud infrastructure is silly
“Beyond the model itself, and I think it's silly to invest in the cloud, people who are buying GPUs to build clouds.”
Khosla: AI will eventually make almost all professional human expertise essentially free
“One of my big forecasts is almost all expertise will be free.”
Khosla: Cognition's Devin is building an autonomous AI programmer, not a Copilot
“Of course, we invest in something like Devon. That's very popular. Everybody knows about it. They're building an AI programmer. They're not building tools for programmers like Copilot. They're building a programmer.”
Khosla: AI primary care and therapy will become larger than specialized AI
“I think the specialties will become embedded in the generic. And the general, like primary care and mental health therapy are two examples that I think will be larger businesses because there's other aspects than just the knowledge aspect.”
Morin: Apple restricts developers from directly accessing its on-device AI models
“The one thing that I was disappointed about is that you cannot access these local models directly. It's something that a lot of folks aren't talking about. They've put a pretty intense wrapper around it, at least on the developer front. You know, you can use t…”
Khosla: Has never taken management fees in 20 years of venture investing
“Since I started the firm 20 years ago, I've never been I've taken a dollar in fees.”
Lessin: Seed-stage AI investing is a fool's errand for venture capitalists
“I think most seed investors getting into AI are playing a fool's errand because it's not a place where one to three million dollars Really can move the needle in terms of value, in terms of what the game is.”
Khosla: Blindly funding Google DeepMind spinouts with $2M seed checks is bad
“Anytime somebody leaves deep, deep mind, give them two million dollars, that's a bad strategy.”
Khosla: 80% of Khosla Ventures seed investments are under $25M post-money valuation
“I would guess That 80% of the investments we've made in C are below twenty-five million posts.”
Khosla: Passes on 80% to 90% of startups seeking billion-dollar valuations
“I'd say 80% or 90% of the billion dollar valuations, maybe more, we pass on.”
Khosla: Portfolio company Zordi operates fully automated, human-free strawberry greenhouses
“We have a greenhouse that's fully automated, including picking the strawberries, which would be hard to do by hand. And they sell at like five times the price in every New York restaurant prices them at like five bucks a strawberry or something. But there are …”
Khosla: Regulation ensures Europe will never participate in the tech industry again
“I think the Europeans, for example, have ensured they'll never participate in tech again. You know, they keep shooting themselves in the foot.”