Meta CEO Mark Zuckerberg was just caught with not just his hand, but his entire arm, inside the cookie jar. <\/p>\n
In yet another sign that his company is seriously struggling to keep up in the AI race, Google informed Meta earlier this year that it was cutting Meta off from the Gemini AI capacity it needed, the Financial Times<\/a> reports<\/a>. <\/p>\n Ironically, the lack of access to Gemini \u2014 which is far superior to anything Meta has built in-house \u2014 slowed down Meta’s own AI projects, including data moderation and customer services, highlighting how heavily Zuckerberg’s company is leaning on its competitors as its own AI efforts languish. <\/p>\n The revelation is bound to sting. Despite committing over half a <\/a>trillion<\/a> dollars<\/a> over the next two years, the company’s flailing AI efforts have been plagued by rock bottom morale<\/a>, infighting<\/a>, high-level executive departures<\/a>, and related PR crises<\/a>. <\/p>\n Sources told the FT that Meta chose Gemini over its in-house Llama open-source models simply because it performed much better. <\/p>\n While Google had to tell other clients it couldn’t meet their AI needs as well, Meta stood out because of its “exceptionally high demand for Google’s models,” according to the financial newspaper. <\/p>\n Meanwhile, Meta has had to do a 180 after initially encouraging its employees to use as much AI as possible. The company is now telling them to be far more frugal with AI tokens, sources told the FT. <\/p>\n A similar story is playing out across the rest of the tech industry, with leadership suddenly getting major sticker shock as access to AI tools continues to ramp up in price. <\/p>\n Meanwhile, Meta is still trying to get its Muse Spark, a multimodal reasoning model, off the ground. According to the company<\/a>, it’s the “first step on our scaling ladder and the first product of a ground-up overhaul of our AI efforts.” <\/p>\n According to the FT’s sources, Spark could finally allow Meta workers to reduce their dependence on other models, like Gemini \u2014 although, embarrassingly, it sounds like they needed Google’s assistance to get there. <\/p>\n