The Twenty Minute VC

Alex Lebrun: Why the EU's AI Regulation is a Disaster; How Zuck Prepares for Meetings | E1027

Alex Lebrun is the Co-Founder and CEO of Nabla, an AI assistant for doctors. Prior to Nabla, he led engineering at Facebook AI Research. Alex founded Wit.ai, an AI platform that makes it easy to build apps that understand natural human language. Wit.ai was acquired by Facebook in 2015. Prior to Wit, Alex was the Founder and CEO of VirtuOz, the world pioneer in customer service chatbots, acquired by Nuance Communications in 2013. ------------------------------------------- Timestamps: 0:00 Intro 0:20 Who is Alex Lebrun? 3:27 How Mark Zuckerberg Prepares for Meetings 6:56 Does founding a startup get easier over time? 10:05 The AI Hype Cycle 15:48 Evaluating Emad Mostaque’s Predictions 17:14 AI Startups vs Incumbents 18:17 AI Models: Pre-Trained vs Fine-Tuned 27:03 Open-Source vs Closed-Source: Which will dominate AI? 28:27 National Data Sets 36:14 Will AI replace doctors? 59:43 Do French startups sell too soon? 1:05:25 AI in China ------------------------------------------- In Today’s Episode with Alex Lebrun We Discuss: 1. Third Time Lucky and Lessons from Zuckerberg: How did Alex make his way into the world of startups with the founding of his first company? What worked with Alex’s prior companies that he has taken with him to Nabla? What did not work that he has left behind? What were the single biggest takeaways for Alex from working with Mark Zuckerberg? How does Mark prepare for meetings? How does Mark negotiate so well? 2. Open vs Closed: Why does Alex believe the winning AI models will always be open? Why are open models not as transparent as people think they are? What are the biggest downsides to both open and closed models? Does Alex agree with Emad @ Stability that we will have “national data sets”? 3. Incumbent vs Startup: Who wins in the AI race; startups or incumbents? How important is access to proprietary data in winning in AI today? How does Alex respond to many VCs who suggest so many AI startups are merely “a thin layer on top of a foundational model”? Is that a fair critique? Which startups are best placed to challenge incumbents? Which incumbents have been most impressive in adopting AI into existing product suites? 4. Models 101: Size, Quality, Switching Costs: Why will the best companies switch the models that they use often? Will any models in action today be used in a year? How important is the size of the model? How will this change with time? In what way is new EU regulation around models going to harm European AI companies? 5. Location Matters: Who Wins: When looking at China, US and Europe, who is best placed to win the AI war? What are the biggest challenges Europe and China face? Why is the US best placed to win the AI race? What does it have to overcome first? If Alex were a politician, what would he do to ensure his country were best positioned? -------------------------------------------- #AlexLebrun #Nabla #HarryStebbings #20vc #venturecapital #artificialintelligence

Harry StebbingshostAlex Lebrunguest

Jun 18, 20231h 16mWatch on YouTube ↗

WHAT IT’S REALLY ABOUT

AI Pioneer Alex Lebrun Slams EU Rules, Reinvents Doctors With Assistants

Alex Lebrun, three-time AI founder and CEO of Nabla, reflects on two decades building chatbots and how large language models have finally reached viable market timing. He argues that generative AI applications are far from a thin layer on foundation models, stressing the deep technical and product advantages of understanding and orchestrating LLMs. A major focus is healthcare: he believes AI won’t replace doctors, but doctors using AI will outperform those who don’t, primarily through automated clinical documentation and ambient assistants. Lebrun also criticizes the EU’s proposed AI Act as disastrously out of touch, predicts open models will dominate, and warns Europe will fall further behind the US and China unless regulation becomes more realistic and better informed.

IDEAS WORTH REMEMBERING

5 ideas

Deep expertise with LLMs is a durable edge even on shared infrastructure.

Lebrun insists that knowing how models are built, where they fail, when to swap them, and how to wrap them with additional ML is a major competitive advantage; applications are not just a “thin layer” on top of public models.

Proprietary data matters less than before; small, high‑quality datasets can be enough.

He notes modern pre-trained models can be fine-tuned effectively with surprisingly small but well‑curated datasets (e.g., ~1,000 Q&A pairs in the LIMA paper), reducing the historic advantage of owning massive data troves.

AI will radically augment clinicians before it replaces them.

Nabla focuses on ambient AI assistants that listen to consultations, auto‑generate documentation, and integrate with EHRs, freeing doctors from admin work that currently consumes ~49% of their time and drives burnout.

Healthcare startups must start from “who pays,” not just “what problem.”

Because payers, providers, and patients are misaligned, he argues founders must begin by identifying the paying stakeholder and designing a solution and problem framing that fit existing reimbursement structures.

Regulating AI too early and too strictly could cripple Europe’s competitiveness.

Lebrun calls the EU AI Act a disaster, saying requirements like full licensing for all training data would make nearly all current LLMs illegal, pushing serious AI work out of the EU and delaying Europe by decades.

WORDS WORTH SAVING

5 quotes

AI will not replace doctors, but doctors who use AI will replace doctors who don't.

— Alex Lebrun

As an entrepreneur, the only thing you need to know is when to add or remove AI from your deck, and this will change about every three or four years.

— Alex Lebrun

Having a lot of proprietary data was very, very important for the last cycle five years ago. Maybe it's less and less true.

— Alex Lebrun

The new [EU] regulation is a disaster... In practice it means that 100% of the LLMs that were trained these last three years would be illegal in Europe.

— Alex Lebrun

For the first 50 years of computerization, computers have been a bad news for doctors.

— Alex Lebrun

Lebrun’s startup journey, lessons from Facebook/Meta, and working with Mark ZuckerbergMarket timing and evolution of chatbots and large language modelsStartups vs incumbents, open vs closed models, and where AI value accruesMyths about proprietary data, hallucinations, AGI fear, and AI safety narrativesNabla’s strategy: AI assistants for clinicians and the reality of healthcare workflowsGo-to-market, payment incentives, and structural challenges in healthcare systemsGlobal AI geopolitics and regulation, especially the EU AI Act’s impact on Europe

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.