Competition, Industry Adoption, and the Open-Source vs. Proprietary AI Debate

Key Insight

Initially, large enterprises like Goldman Sachs, Amazon, Apple, and Verizon exhibited apprehension, banning ChatGPT due to concerns over hallucinations, copyright infringement, and high costs, preferring to monitor the rapidly evolving technology. Despite this, generative AI steadily integrated into established online platforms: Spotify launched its AI-powered 'DJ,' Zoom introduced 'Zoom IQ,' Snapchat made its 'My AI' chatbot free to its 750 million monthly users, and Instacart unveiled a 'Shopping Assistant.' Microsoft aggressively integrated Copilot (formerly Sydney) across its ecosystem, incorporating DALL-E into Bing, adding Copilot to its Office 365 suite (Word, Excel, PowerPoint, Outlook, OneNote), and deploying GitHub Copilot X to boost coding speed by an average of 55 percent, eventually integrating Copilot into Windows 11. However, Copilot's cautious design led to it shutting down on simple questions and refusing to answer queries like 'Who won the 2020 U.S. presidential election?', yielding less than a 1 percent market share increase for Bing.

Google faced significant competitive pressure, including Samsung considering replacing Google as the default search engine on its phones with Bing, and Apple potentially following suit. In response, Google merged DeepMind and Google Brain into the unified Google DeepMind, led by Demis Hassabis, to accelerate AI progress. Google also lifted geographical restrictions on Bard, making it available in 160 countries, and integrated AI capabilities like 'Duet' into Gmail and Google Docs, emphasizing a 'bold and responsible' approach—a subtle dig at competitors' bots known for erroneous answers. While Apple and Amazon's voice assistants, Siri and Alexa, appeared stagnant, Meta made a strategic move by open-sourcing LLaMA, its large language model, after initially being caught off guard by ChatGPT's frenzy. This contrasted with the closed, proprietary nature of GPT-4, Bard, Copilot, and Pi, allowing users to customize and fine-tune LLaMA, leading to over 7000 spinoffs uploaded to Hugging Face by year-end.

The 'open source versus closed' debate emerged as a critical flash point in the AI community. Reid Hoffman, despite his background with open-source Mozilla, argued against open-sourcing foundation models, warning that once released, they become available to 'criminals and terrorists and rogue states.' Conversely, proponents like Yann LeCun championed open source for democratizing AI, accelerating progress, fostering a vibrant ecosystem, and serving as a 'competitive equalizer' for startups against tech giants, enabling them to leverage free, high-quality models without needing hundreds of millions in investment. A leaked Google researcher's memo in May warned that Google was underestimating the open-source threat, noting LLaMA's power—nearly matching GPT-4 while being smaller and efficiently runnable on personal devices—and concluding that open-source software was 'quietly eating our lunch.' This intense industry activity culminated in events like San Francisco's 'AI's Woodstock,' an AI meetup that unexpectedly drew 5000 attendees, symbolizing the hype and rapid community growth, though some veterans noted a prevalence of 'hustlers, hangers-on, and the curious,' reminiscent of the dot-com bubble.