Sam Altman sat comfortably between Satya Nadella and Sundar Pichai at a White House gathering of top AI CEOs in May — with a noticeable gap among them. With Alphabet, Microsoft, and OpenAI in attendance, it was impossible to miss Mark Zuckerberg’s absence. And that appeared to be no accident. The meeting, one administration official said, “was focused on companies currently leading in the space.” Political symbolism is overrated, but the words must’ve stung.
Ever since the gathering, Zuckerberg and Meta have rapidly shipped products that compete directly with Altman’s OpenAI, opening up one of the most intense — and overlooked — battles in tech today. Meta’s countered OpenAI’s GPT-4 with Llama 2, its own large language model that’s open-source and more customizable. It’s responded to ChatGPT with more than two dozen specialized chatbots in its messaging apps. It’s advocated for a more permissive AI research environment, opposing OpenAI’s call for controls. And today, nobody would leave Meta outside the room. “They probably have forced the issue,” one DC insider told me.
Zuckerberg’s offensive has baffled some looking for direct business results, and it’s easy to view it as an ego thing, but limiting Altman’s ascent is a clear business imperative. OpenAI is building popular consumer products, recruiting top AI talent, and pushing to restrict AI research. This all threatens Meta’s ability to grow and innovate, so the company’s going back to a familiar place: destroy mode.
Altman might not have known it then, but he began competing with Zuckerberg the moment he introduced ChatGPT. Messaging is Meta’s strongest category. The company owns two different billion+ user chat apps (WhatsApp and Messenger) and another counting Instagram. It tried to build chatbots (far too) early because it saw them as the next computing platform in 2016. Meta even built a ChatGPT precursor, called M, that responded to questions, booked flights, and drew pictures. M’s ‘technology’ was almost entirely human though, and Meta eventually shut it down. But the vision persisted.
When ChatGPT debuted in November 2022, Altman not only released one of Zuckerberg’s dream products, he built the fastest-growing consumer product ever (at the time). The technology caught up to the vision, and someone else was capitalizing. The risks to Meta were manifold. People spending time speaking to ChatGPT might ignore their friends on WhatsApp. OpenAI seemed interested in building that next computing platform — where everything from flight reservations to food ordering is available via chat — via its ChatGPT Plugins product. And Altman was good at making money from chat, something Meta’s long puzzled over.
Instead of building one generalized bot to limit ChatGPT’s growth — a clone strategy like Stories and Reels — Meta built more than two dozen specialized ones. The idea was that a multi-use bot was a nice demo, but the technology would eventually move to more specific, narrow use cases. AI bots were already specializing, including Harvey for law and Character.ai for entertainment. And ChatGPT lost users steadily over the summer. So in September, Meta released 28 different AIs in its messaging apps, including Coco (for dancing), Max (for cooking), and Victor (for training). It will take some time to evaluate the bots’ success, but Meta won’t be shy about pushing them to its 3.96 billion monthly active users. It has a track record in this area.
Meta is also going after the core of OpenAI’s business, taking on its GPT-4 model, which ChatGPT turned into a sensation. Companies are swiftly building on top of GPT-4, helping OpenAI’s reach $1.3 billion in annualized revenue. Just this week, Microsoft announced its OpenAI offering grew from 11,000 to 18,000 customers in one quarter. But Meta’s counterpunch might limit growth there as well.
In July, Meta released Llama2, an open-source large language model that gives developers more flexibility to customize vs. simply build on an API. Llama2 is already at the forefront of the open-source AI movement, the hottest area of AI research today, one that’s countering OpenAI’s proprietary approach. And in releasing it, Meta’s helping drive the breakthroughs OpenAI pioneered toward commoditization.