How Meta Went From Open Source Hero to AI's Biggest Villain

AI Summary

In 2023, Meta emerged as an unlikely hero in the AI community by releasing Llama, a frontier AI model, for free, allowing developers to download, modify, and build upon it without restrictions. This move, unprecedented in Silicon Valley, led to over a billion downloads within two years, establishing Meta as a beloved company in the AI space, especially when other major labs kept their models closed. However, this trust was severely damaged in early 2025 following the release of DeepSeek v3 by a Chinese AI lab, which matched or exceeded Llama's performance at a fraction of the cost, exposing Meta's inefficiencies. Under immense pressure, Meta rushed the launch of Llama 4 on April 5, 2025, making bold claims about its performance against rivals like GPT-4o and DeepSeek v3, and touting a 10 million token context window for its Scout model. However, community testing quickly revealed discrepancies: Maverick scored poorly on coding benchmarks, and Scout's context window degraded sharply past 256,000 tokens. The most significant controversy arose when it was discovered that the version of Maverick submitted to the LM Arena leaderboard, where it ranked highly, was an experimental, specially optimized chat version not available to the public. This led to accusations of 'fake' benchmark comparisons and allegations of training on test sets. Internally, this crisis led to a major shake-up, with Mark Zuckerberg sidelining the original Gen AI team, investing $14.3 billion in Scale AI, and hiring Alexander Wang as Meta's first Chief AI Officer to lead a new division, Meta Superintelligence Labs. This reorganization prompted the departure of Yann LeCun, a Turing Award winner and long-time head of FAIR, who criticized the company's shift towards short-term, benchmark-driven development over foundational research like his championed 'world models.' LeCun confirmed the fudged benchmarks and the internal turmoil in a Financial Times interview after his exit in November 2025. The presenter concludes that Meta destroyed the trust it had built by prioritizing optics over honesty, ultimately breaking its own 'move fast and break things' motto by breaking itself and its relationship with the open-source community.

Claims Extracted (13)