How Meta Won Open Source AI, Then Threw It Away

In 2023, Meta did something no other AI lab had the guts to do. They gave a frontier model away for free. Two years and a billion downloads later, they destroyed the trust they'd built in a single weekend. This is the story of Llama. 🔗 LINKS / SOURCES Yann LeCun's Financial Times interview: https://www.ft.com/content/e3c4c2f6-4ea7-4adf-b945-e58495f836c2 Free summary of the FT interview: https://the-decoder.com/you-certainly-dont-tell-a-researcher-like-me-what-to-do-says-lecun-as-he-exits-meta-for-his-own-startup/ Nathan Lambert's breakdown of the Llama 4 release: https://www.interconnects.ai/p/llama-4 DeepSeek V3 Technical Report: https://arxiv.org/abs/2412.19437 Original LLaMA paper (Touvron et al., Feb 2023): https://arxiv.org/abs/2302.13971 Aider polyglot benchmark leaderboard: https://aider.chat/docs/leaderboards/ Aider polyglot benchmark writeup: https://aider.chat/2024/12/21/polyglot.html OpenAI gpt-oss announcement: https://openai.com/index/introducing-gpt-oss/ TechCrunch on the unmodified Maverick ranking #32 on LM Arena: https://techcrunch.com/2025/04/11/metas-vanilla-maverick-ai-model-ranks-below-rivals-on-a-popular-chat-benchmark/ Meta's official Llama 4 launch post: https://ai.meta.com/blog/llama-4-multimodal-intelligence/ 🔗 Links: MY 12K+ DISCORD 💬 https://discord.gg/GkrFX4zT2C CONNECT WITH ME ON SOCIAL 📸 Instagram: https://instagram.com/lewismenelaws 🎚TikTok:https://tiktok.com/@lewismenelaws 🐣 Twitter: https://twitter.com/LewisMenelaws My gear 💻 https://liinks.co/lewismenelaws ⏱️ CHAPTERS 00:00 The model that broke the rules 00:59 Early 2023: the closed-AI consensus 02:17 LLaMA 1 leaks to 4chan 03:18 Llama 2 goes properly open 04:17 Llama 3 and 3.1 become the open-source default 05:50 DeepSeek changes the math 07:22 The pressure to ship Llama 4 08:21 Enter Yann LeCun (and Alex Wang) 09:24 Saturday, April 5, 2025 — Llama 4 drops 10:38 The community finds the cracks 12:10 The LM Arena submission 13:02 The benchmark-blending allegations 13:42