Meta Delays AI Model Release
Meta Inc., led by Mark Zuckerberg, is postponing the launch of its upcoming artificial intelligence model by about two months. The decision comes after internal tests indicated that the new technology didn’t measure up to competitors from Google, OpenAI, and Anthropic.
In light of their findings, Meta’s AI leaders are contemplating temporarily licensing Google’s Gemini for their project, as detailed by The New York Times.
Sources have disclosed that the internal testing of the new Avocado AI model yielded disappointing results, despite extensive effort from the engineering team.
Reports suggest that Avocado lacked proficiency in reasoning, coding, and writing, which has pushed its release to May instead of this month.
This setback follows Zuckerberg’s significant push into AI development. The company has invested billions in recruiting premier AI talent and building data centers, with an anticipated spending of up to $135 billion this year. This amount is almost double the $72 billion spent on AI initiatives last year.
Zuckerberg remarked in July that new AI models from Meta would “move to the forefront” in the coming year. According to The Times, while Avocado scored better than Meta’s previous AI attempts, it still lagged behind Google’s latest model.
In an investor call, Zuckerberg expressed optimism about the first model, stating, “We expect it to be good and to reflect our rapid growth in this area.”
A spokesperson for Meta maintained that the company is continuously making strides in AI, looking forward to showcasing their developments soon.
In the competitive landscape of AI, Meta’s endeavors, particularly through its ownership of Facebook, Instagram, and Threads, are becoming increasingly notable. However, it’s widely acknowledged that competitors like Google, OpenAI, and Anthropic are currently leading in foundational AI model development.
Following the underwhelming performance of its predecessor, Llama 4, last year, Zuckerberg intensified AI initiatives by investing $14.3 billion in Scale AI and appointing Alexandr Wang, 29, as the new chief AI officer.
Wang has played a crucial role in establishing TBD Lab, which has been working on Avocado. The lab completed the “pre-training” phase late last year, with “post-training” beginning in January and the final aim originally set for mid-March.
So far, Wang’s division has launched only one product, a video app called Vibes, similar to OpenAI’s Sora.
Despite some employee turnover in TBD Lab, the team, comprising around 100 members, has been focused on the model’s launch.
There has also been internal discussion about whether the new AI model will be open source, a move that has been a hallmark of Meta’s strategy, contrasting with the cautious approach of rivals like OpenAI.
Reports have highlighted emerging tensions between Wang and other key figures at Meta regarding how the new model could enhance the company’s advertising operations.
Recently, Meta informed staff of plans to form a new AI team reporting to CTO Andrew Bosworth, which will collaborate with Wang, as initially reported by the Wall Street Journal.
