Reality came knocking on Mark Zuckerberg’s ambitions for artificial intelligence this past April. Just a few weeks before that, the Meta CEO, who is 41 years old, had made confident remarks about how powerful the company’s new AI system would be.
He had spoken about wanting the system to stand on the same level as those built by companies like OpenAI. Internally, he had mentioned his desire for the AI to support voice-enabled chat and other advanced tools, according to people who discussed with him.
When Meta organized its AI gathering that same month, things did not go as planned. The latest AI system failed to meet the performance of its rivals. Some of the advertised features, such as voice conversation tools, were not yet available.
Delays During Meta’s AI Event
Many of the developers who had arrived with high expectations ended up feeling let down by what they saw. Those who were close to Zuckerberg said he recognized that Meta was behind in this area, which he found unacceptable.
Right after that, he started exchanging ideas on WhatsApp with top figures in the company, including Chris Cox, who oversees product, and Andrew Bosworth, who is the chief technology officer, to figure out their next steps.
That group chat marked the beginning of a fast-paced series of actions that reached across Silicon Valley. The vice president who was leading Meta’s generative AI efforts was removed from the role.
Zuckerberg then approved a huge investment of $14.3 billion in a start-up called Scale AI and brought on board its 28-year-old founder, Alexandr Wang. Meta also got in touch with other companies working on AI, including Perplexity, which runs an AI search platform, to discuss possible partnerships.
Discussions Around Llama and Outside AI Options
A different set of conversations took place within Meta’s AI unit. Two people familiar with those talks said there were discussions about cutting back support for Meta’s own AI model, Llama. Llama is designed to be “open source,” which means its structure is publicly available for others to work with.
During these talks, company leaders also considered shifting toward using systems from other companies like OpenAI and Anthropic, both of which keep their code private.
A spokesperson from Meta made it clear that the company still plans to continue developing Llama, and that they are aiming to put out multiple new versions later this year.
