What Big Tech Missed And How Startups Can Still Win

The Cost of Expectations [0:00]

• The real cost of a $1.2 billion funding round is not dilution but managing expectations.
• High expectations from investors and the public require tangible progress within a reasonable timeframe.

  "But the expectations you create in the outside world, and if you raise 1 billion and you do nothing for two years, or people don't see anything coming out of it for two years, then it's very, very hard to survive."

Early Ventures and the .ai Domain [1:09]

• Alex recalls co-founding Algolia in the same YC batch as Wit.ai, founded by Alex.
• Wit.ai was one of the first companies to use the .ai domain, which was difficult to acquire.

  "Yeah, it was the first .ai domain, actually. It was very hard back then to buy .ai. It's an island of Australia."

Selling Wit.ai to Facebook [2:32]

• Alex recounts receiving an email from Mark Zuckerberg, initially mistaking it for spam.
• Following YC playbook advice, he initially ignored offers but later engaged with Facebook.

  "And after like half a second, I deleted the email without opening it. It's a scam, you know, it's a spam."

Entrepreneurial Philosophy: Early Adoption [4:40]

• Alex's previous companies, including Virtuos (a chatbot company in 2002), were often ahead of their time.
• He believes his strength lies in working with early-stage technology and understanding nascent markets.

  "So I think my skills are more on the early side of the go-to-market. So that's why I prefer to take early ideas."

AmiLabs: The Ambitious Leap [7:10]

• AmiLabs is described as an ambitious project due to the immense cost of acquiring thousands of GPUs.
• The goal is to work upstream on AI models, potentially impacting millions of users.

  "So this is what is not normal in this project. After, so I've done several startups, always on the applied AI space."

World Models vs. LLMs [9:10]

• LLMs learn from text (what humans have written about the world), while world models learn directly from real-world sensory data (video, audio, touch).
• World models aim to provide machines with common sense and the ability to handle novel situations.

  "So LLM is like someone who has never gone outside, never left the room they were born in, but read all the books every day for centuries and centuries."

Applications and Advantages of World Models [13:00]

• Key applications include advanced robotics, enabling robots to operate safely and effectively in open environments.
• World models are proposed as a superior solution to current limitations of LLMs in robotics, such as VLA (Vision-Language-Action).

  "Current robots are very, very narrow, vertical robots. They do one thing in one position."

The Future of AI: Coexistence [17:50]

• LLMs will remain excellent for language, mathematics, and other symbol-based tasks.
• World models are expected to excel in high-dimensional, noisy, and long-horizon problems.

  "LLMs are designed for that. It works. We will never beat, I think, the LLM at that."

The Role of Startups in Innovation [21:00]

• Startups can take risks that large companies cannot, leading to groundbreaking innovations like GPT.
• Alex shares an anecdote about wanting to hire 10,000 concierges at Facebook, illustrating ambitious risk-taking.

  "But the one thing you have that they don't is the ability to make crazy things, to take some risks."

Bottlenecks and Ambition [25:00]

• The main bottlenecks for world models are talent, data, and compute (GPUs).
• Alex advises founders to be highly ambitious within a narrowly defined problem space.

  "No, if I could go back, you know, I think when you start, you should look at a very narrow problem, but be very, very ambitious in this narrow problem."

The Vision for AmiLabs [30:00]

• If AmiLabs succeeds, the future will involve helpful robots and machines with common sense.
• These advancements will make dangerous or difficult jobs safer and more manageable.

  "I think we'll have robots, you know, helpful robots, actually helpful robots."