Zuckerberg's AI Bet Is A Total Disaster

In 2023, Meta released Llama and gave away a frontier AI model completely for free. Within two years, it had been downloaded over a billion times. Developers, startups, and governments around the world were building on it. Then, in April 2025, they launched Llama 4 and got caught submitting a secretly different model to a public leaderboard to fake their benchmark results. Now Zuckerberg is spending $600 billion trying to fix what Llama built for free. This is the full story of how Meta destroyed the most trusted brand in open source AI in a single weekend, from the benchmark fraud that ended Llama's dominance, to the $14.3 billion hire that immediately backfired, to the 6,500 engineers being recorded to train their own replacements, to what Yann LeCun, the man who built Meta's AI lab, said on his way out the door.