Llama is struggling, Meta begins to seek cooperation on "third-party AI products"

Faced with the fatigue of its self-developed AI models in competing with industry leaders, Meta is adjusting its long-standing internal development strategy and turning to collaborate with external AI companies.

On Friday, August 22, Meta's new Chief AI Officer Alexandr Wang announced on social platform X that the company plans to engage in "technical cooperation" with AI image and video generation startup Midjourney, granting access to its "aesthetic technology," aimed at "bringing beauty to billions of people."

Wang stated that to ensure Meta can provide the best products, the company needs to adopt a "comprehensive" strategy, including collaboration with "the best players in the industry." This move marks a significant strategic shift for Meta in the AI field from closed self-development to open collaboration, directly impacting its future ability to integrate AI-generated content into social applications.

The backdrop of this collaboration is that despite CEO Mark Zuckerberg investing billions of dollars in pursuing "superintelligence," Meta's AI products seem to have fallen behind competitors, and confidence in the internally developed Llama series models has weakened.

Self-developed Models Lagging, Strategic Focus Shifts

The collaboration with Midjourney highlights the challenges Meta faces on its path of self-developed AI.

Although Meta launched the image generation tool Imagine in 2024 and plans to fully integrate the video generation model Movie Gen into Instagram by 2025, industry insiders believe that compared to models already released to consumers, such as Google's Veo 3 and OpenAI's Sora, Meta's products "already seem outdated."

A deeper challenge lies in its foundational large language model. According to insiders, due to waning confidence in its Llama model, Meta has begun using third-party models for internal tasks such as coding.

At the same time, the company has abandoned plans to publicly release its flagship large language model Behemoth, instead focusing on developing new models. This series of actions indicates that Meta is shifting from the philosophy of "doing everything by itself" to a more pragmatic strategy of combining internal and external resources.

Massive Investment and Internal Turmoil Coincide

Meta's strategic adjustment occurs against the backdrop of unprecedented investment in the AI field. In recent months, Zuckerberg has not only actively recruited top AI researchers from competitors but has also doubled down on investments in AI infrastructure, acquired AI voice company Play AI, and invested in data labeling company Scale AI.

However, the massive investment has not immediately translated into a stable organizational structure and leading products. Reports indicate that Meta's AI department, "Meta Superintelligence Lab," has undergone four reorganizations in the past six months, reflecting the company's difficulties in establishing an effective organizational structure. This contrast between investment and output may be a direct reason prompting Meta to seek external technological collaboration.

The "Aesthetic Technology" of AI Newcomers

The other party in this collaboration, Midjourney, is an AI image generation company founded in 2021. Its founder, David Holz, insists on not accepting venture capital and chooses to self-fund its development, but this has not hindered it from becoming one of the most popular image generation tools.

For Meta, the core value of Midjourney lies in its mature "aesthetic technology." In social media applications, the visual appeal of generated content is crucial. Through authorized collaboration, Meta can quickly enhance the quality of its AI-generated content, thereby catching up with competitors in user experience.

Midjourney founder David Holz also confirmed the collaboration on X and emphasized that the company remains "an independent, community-supported research lab with no investors," ensuring its independence in technological collaboration