It’s slowing becoming clear why Mark Zuckerberg was willing to spend hundreds of millions of dollars to poach AI researchers from rival labs.
Meta is in talks with Google about using its Gemini’s models to improve its ads business, The Information reports. The talks are still in relatively early stages and may not result in an agreement. Neither company has commented on the report.
Meta employees have reportedly proposed fine-tuning Gemini and Google’s open-source Gemma models on Meta’s ad data to improve its ad targeting capabilities. Interestingly, Google and Meta compete in the online ad market. Meta has also been considering partnerships with Google or OpenAI to improve its AI features, including conversational responses for queries on its chatbot Meta AI and to power AI features in Meta’s social media apps.
But this reliance on outside providers underscores just how far behind Meta’s own AI efforts currently are. Facebook had been an early pioneer in the AI space, setting up the Fundamental AI Research (FAIR) all the way back in 2013 in partnership with Yann LeCun, one of the godfathers of AI. Facebook then proceeded to deeply embed AI into its core products and services — AI models powered better ad targeting, news feed ranking, content moderation, and personalization. The company developed FB Learner Flow, a platform enabling engineers to run hundreds of thousands of machine learning experiments monthly using Facebook’s vast data resources. One of Facebookâs most influential AI projects was the development of PyTorch, an open-source machine learning framework first released in 2016. Pytorch is widely used in the AI and ML community even today, and is one of the tools powering the AI revolution.
Meta had also appeared to start off strongly in the generative AI era by releasing powerful Llama models that were the best open-source AI models for a while. But Llama 4 proved to be a disappointment, and didn’t perform as per expectations. Meta never seemed to recover from the blow, and saw its open-source advantage completely overshadowed by open-source alternatives coming out of China.
Meta then attempted to overhaul its AI efforts. It acquired a large stake in Scale AI, a data labeling company, and made its founder Alexandr Wang its Chief AI Officer, with even Yann LeCun reporting to him. Meta has also been splurging large sums of money to acquire the best AI talent, and offering pay packages exceeding $100 million to poach researchers away from rival labs.
But it appears that Meta still needs time to set its AI plans into motion. It recently partnered with Midjourney for image generation models, and released Vibes, an AI video feed powered by Midjourney and Black Forest in its Meta AI app. It now seems to be partnering with Google to power its ad models. And these partnerships indicate that Meta doesn’t only have these AI capabilities in-house, but also doesn’t believe it’ll develop them anytime soon. And while these partnerships will help Meta build and grow its substantial social media and ads business, they do indicate that the company — at least as of now — it some way away from the frontier in AI.