The Man Who (Potentially) Cost Meta a Trillion Dollars

"Now I am become Death, the destroyer of worlds."

Based on leaked news reports, Stephen Colbert’s Late Night show lost CBS hundreds of millions of dollars over the past few years. Although this news story is popular at the moment as a political flashpoint, hundreds of millions in losses is peanuts compared to the value lost at Meta due to one individual: Yann LeCun.

First, let’s take a trip down memory lane to understand how single individuals can create or destroy economic value. Steve Ballmer, as CEO of Microsoft, led the company down unproductive paths. The Zune mp3 player was the central symbol of Microsoft’s lethargy. “Like Apple, but worse” was the general sentiment.

In August 2013 Steve Ballmer unexpectedly announced his retirement from Microsoft. How did markets react? Microsoft’s value increased by $18 BILLION dollars. Microsoft hadn’t even announced who would replace him. The market consensus on the difference between Steve Ballmer and an unknown CEO was $18 billion. Satya Nadella eventually stepped in, transformed Microsoft’s strategy, and now the company is worth many times its 2013 valuation.

(Image from Marginal Revolution)

Take any moment between five years ago and two months ago. Imagine you’re a cutting-edge LLM researcher looking to make a lasting impact in AI. Meta reaches out to you to join their AI team. There’s Meta, but what are your other options?

1. OpenAI – Post-November 2022, it’s an obviously attractive choice, but even before then it was a cutting-edge AI startup.

2. Anthropic – Solid, ambitious, and intelligent teams focused on the cutting edge.

3. Google – Slow, bureaucratic giant, but with unreal amounts of talent and resources.

4. xAI – Really smart people attacking problems at an almost maniacal pace.

5. Niche players (e.g., Midjourney, Mistral) – Cool if you’re into their niche and culture.

6. Apple – Jk. Apple doesn’t do AI. They’re praying on the sidelines that AI becomes commoditized.

Besides Apple, you have some interesting options.

How does Meta stack up? You use AI (not Google, you’re not a boomer) to review Meta’s public statements about AI and discover their Chief AI Scientist has some interesting opinions:

• “This AI Pioneer Thinks AI Is Dumber Than a Cat”

• “LLMs can do none of those [reason, plan, etc.] or they can only do them in a very primitive way and they don’t really understand the physical world. They don’t really have persistent memory. They can’t really reason and they certainly can’t plan.”

Meta’s Chief AI Scientist calls LLMs dumb parrots repeatedly and insists they are not the future. As a cutting-edge researcher… you don’t agree. LLMs are an incredible technology with limitations, but still amazing potential. Your own research has shown this.

You, along with most other teams, picked a different horse to bet on.

Meta’s Reaction

Meta eventually found themselves in last place (Apple has been DQ’d). They poured in money and turned to open source (not a bad strategy given their ecosystem alignment), but Meta still lacks a cutting-edge research team.

Eventually, Meta (likely Zuckerberg?) needed to stop losing or give up. They had not a single major model in the top 30 on arena leaderboards.

Meta began throwing amounts of money, once reserved for sports superstars, at everyone else’s top researchers. How did OpenAI feel about that?

“I feel a visceral feeling right now, as if someone has broken into our home and stolen something.” – Mark Chen

(I started with TBPN’s post and had some fun with AI and photoshop. 😆)

For the detailed breakdown on the Meta Superintelligence team, click the image below:

My Takeaways

I am impressed that Meta doubled down instead of giving up. It’s a gutsy move. Also, I don’t completely disagree with Yann LeCun! Ultimate superintelligence likely requires a larger reasoning and reinforcement framework beyond just an LLM. In my opinion, there are too many theoretical weaknesses to reach that point with only language token reasoning. However, I still believe LLMs will be a significant or even majority component of that larger intelligence framework. I believe LeCun got lost in ideal future purity over today’s practicality.

Imagine a world where the Chief AI Scientist is an LLM bull, and Meta attracts top-tier researchers early. Without exaggeration, this could have been worth a trillion dollars to Meta.

Who are the candidates for the greatest economic value destruction by an employee of a company?

Each step is a 1-2 orders of magnitude change:

Yann LeCun >> Steve Ballmer >> Stephen Colbert

Then again, they’re in the arena. Or at least, they’re in much bigger arenas than I am.