Nov 24, 2023 · 33m · news
The Ultimate AI Roundtable: What Happens Now in AI, Why Google are Vulnerable | E1085 · 20VC with Harry Stebbings
compilation · excluded from per-person scoring
Harry Stebbings · 8m spoken Yann LeCun · 5m spoken Des Traynor · 5m spoken Miles Grimshaw · 3m spoken Richard Socher · 2m spoken Jeff Seibert · 2m spoken Emad Mostaque · 1m spoken Tomasz Tunguz · 1m spoken Christian Lanng · 1m spoken Douwe Kiela · 25s spoken Cris Valenzuela · 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
This special compiled episode of the TwentyVC podcast features a simulated roundtable with prominent AI leaders discussing the future of foundational models, shifts in value accrual from infrastructure to applications, and the strategic positioning of tech giants like Google and Apple.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 25.3% of the talking time here. How this is scored →
Harry as informed peer 4.0 Guest teaching 3.1 Guest disagreement 3.6 Harry pushing back 3.1
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 18:24 Christian Lang's Anti-Copilot Rant
Christian Lang forcefully rejects the industry's co-pilot narrative, calling current applications bad UX and mocking the concept as ending up with 10,000 Clippys talking to each other.
Hardest push from Harry ▶ 22:21 Harry Questioning Apple's Real ProgressHarry directly interrupts and challenges Des Traynor's optimistic prediction about Apple, pointing out that so far observers are left searching for any real AI evidence.
Biggest teaching moment ▶ 7:56 Douwe Kiela Exposing Open-Source NaivetyDouwe Kiela dismantles open-source optimism, educating the host and panel on OpenAI's massive scale economies and proprietary user intent data that open-source cannot easily replicate.
Harry holds his own ▶ 14:26 Harry Providing Concrete Business Model ExampleHarry demonstrates strong domain expertise by converting Miles Grimshaw's abstract idea of selling work into a clear, concrete marketing spend automation scenario.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Foundational Model Layer & Commoditization | 4 | 2 | 2 | 3 | Harry sets up the core discussion around foundational model commoditization and prompts Des Traynor to share specific internal benchmark results comparing OpenAI to competitors. The guests offer balanced, empirical views on model quality and narrowing gaps without aggressive pushback. | |
| Importance of Model Size and Longevity | 4 | 3 | 4 | 3 | Harry guides a panel debate on whether model size translates directly into defensibility, contrasting guest perspectives. Guests engage in moderate disagreement, with Cristóbal Valenzuela arguing models are not a moat while Jeff Seibert emphasizes data parameter correlation. | |
| The Open vs. Closed Source Debate | 4 | 4 | 5 | 3 | Harry frames the open versus closed source debate, prompting Yann LeCun and Douwe Kiela to detail opposing philosophies. Douwe forcefully challenges open-source optimism, labeling expectations that open models can match OpenAI's scale economies as incredibly naive. | |
| Value Accrual: Infrastructure vs. Application Layer | 3 | 3 | 2 | 2 | Harry introduces the value accrual debate between infrastructure and application layers. Tom Tunguz and Des Traynor deliver structured, analytical breakdowns comparing Web2 cloud market caps to AI value capture in a collaborative manner. | |
| AI Business Models: Selling Work vs. Software | 5 | 2 | 4 | 3 | Harry actively demonstrates domain knowledge by articulating a concrete example of selling outcomes rather than software seats. Jeff Seibert pushes back against panel consensus, arguing AI is merely an underlying tech tool like Memcached that won't disrupt standard pricing models. | |
| The Co-pilot Phenomenon and Intelligent Assistants | 4 | 4 | 5 | 3 | Harry probes whether co-pilots are fundamentally an incumbent strategy. Christian Lang colorfully rejects the co-pilot paradigm, comparing it to an annoying future of 10,000 Clippys talking to each other. | |
| Tech Incumbents: Apple's Silent Strategy | 4 | 2 | 3 | 5 | Harry directly challenges Des Traynor's optimistic stance on Apple by highlighting that Apple has shown very little public evidence of AI leadership so far. Des and Jeff explain Apple's long-term on-device and privacy strategy in response. | |
| Google's Innovator's Dilemma and Vulnerability | 5 | 3 | 3 | 4 | Harry pushes Des Traynor to move beyond easy theoretical criticism and state what concrete strategic choices he would make as CEO of Google facing Wall Street pressure. Des responds with a proposal for sponsored LLM ad injections. | |
| Amazon's Engineering Strength and Strategic Options | 3 | 5 | 4 | 2 | Harry asks Yann LeCun about existential AI risks and job displacement. Yann delivers an authoritative reframe, dismissing doomer narratives as nonsense and drawing historical parallels to early aviation safety fears. |
Statements from this episode (22)
Mostaque: Only 5 or 6 foundation model companies will exist in 3-5 years
“I think that there's only going to be five or six foundation model companies in the world in three years, five years. I think it's going to be us, Nvidia, Google, Microsoft OpenAI, and Meta and Apple probably are the ones that train these models.”
Seibert: Foundational AI models will inevitably be commoditized
“I certainly think we will, and this may not be a popular position, Obviously, opening eyes, charging ahead, sort of leading the way right now. I think the market forces at work mean there's just immense energy to have an open source equivalent. Meta appears to…”
Mostaque: No AI models available today will be used in a year
“The reality is no models that are out today will be used in a year.”
Emad Mostaque: There is no such thing as an unbiased AI model
“There is no such thing as an unbiased model.”
Cristóbal Valenzuela: AI models are not a competitive moat
“I think models are not a mode. Models eventually don't matter. What matters most is the people building those models, and how fast can you change and learn from those models.”
Kiela: Believing open-source AI can match OpenAI is incredibly naive
“I would like it to be true that with open source, we could just keep up with all of that. But I think that's just incredibly naive. Open AI has this very deep understanding of how people want to use language models. Basically nobody else has, and they have thi…”
Socher: An open-source GPT-4 equivalent will arrive by end of 2023
“I predicted that we'll have a GPT-IV equivalent model Before the end of the year, that's open source.”
Tunguz: Top 3 cloud providers equal top 100 apps in market cap
“In web two, if you take the top three clouds and you look at their market cap, so AWS, GCP, and Azure, it's about a 2.1 trillion dollar market Just for the cloud businesses. And then if you take the top 100 publicly traded cloud companies, both on B to C and B…”
Traynor: AI infrastructure will consolidate into 3-4 providers in direct price competition
“There'll probably be three or four big providers is my guest, probably GCP, AWS, maybe open AI directly. And I'm sure Azure, that's how I think that that layer will play out. And they'll end up in direct price competition with each other.”
Grimshaw: AI software will sell work and outcome SLAs rather than seats
“I think the way to encapsulate would be this idea of selling the work, not the software, and that we'll move from a paradigm where you might think of it as moving from what we see right now as a copilot and moving to what I think about as a control center, whe…”
Lang: Software pricing will shift to consumption-based models to retain value
“I think we're going to absolutely move to consumption based pricing. It's the only way I think it's going to be very hard to hold the moat on, on recurring revenue as it is today, because customers will want to see more value, and they'll want to see a more so…”
Seibert: AI will be commoditized quickly without disrupting existing software pricing
“So this may be just me, but I very much see AI as a tool, not a product. And so it's a technology. It's like your database. It's like Memcash back in the day, and so because of that, I don't think it'll change how people price in specific industries. Like, if …”
Grimshaw: AI copilots are an incumbent strategy, not a startup opportunity
“I think Copilot is an incumbent's strategy. Incumbents own distribution, they own data, they own the UX, and they own a business model that all aligns to a Copilot.”
LeCun: Users will talk to AI assistants instead of searching Google or Wikipedia
“So people are probably going to use this almost exclusively in the future for their interaction with the digital world. You're not gonna go to Google or Wikipedia. You're just gonna talk to your assistant.”
LeCun: AI assistant infrastructure will be forced to be open source
“They would be so pervasive, so much will ride on those systems that I don't think anyone will accept that those assistants be behind the event horizon in a private company. They will insist that the infrastructure is open.”
Traynor: Action-capable Siri will transform desktop and iOS interaction models
“I think what they'll win is this idea of Siri might finally become useful. Siri is currently not useful because it doesn't really have enough smarts, but I think when Siri can be as conversational as ChatGPT and can take actions on the device, it'll change the…”
Seibert: On-device iPhone LLMs will make OpenAI obsolete within five years
“I bet Apple can and will, and so if you project forward five years, if they get to the point where they can run a sufficiently large LLM on your iPhone, then OpenAI is out of the picture. Don't even need to hit their servers. It's just on your phone.”
Seibert: Google delayed Gemini to Q1 due to poor performance
“They've just punted Gemini into Q one, which tells me it's not doing very well.”
Socher: Google generates $500M per day from search advertisements
“They make five hundred million dollars a day With privacy invading advertisements on that page.”
Traynor: Amazon could outright acquire Anthropic to integrate into AWS EC2
“I'd be wary of Amazon. So like I could see Amazon just like Flat up like buying Anthropic and being like, let's just make this part of the EC two cluster.”