If you’ve ever wondered what happens inside Silicon Valley’s conference rooms, a new report provides some direct insights.
💰💸 Don’t miss the move: SIGN UP for TheStreet’s FREE Daily newsletter 💰💸
On January 14, 2025, a U.S. court unsealed several pages of leaked emails between two prominent members of Meta Platforms (META) . From October 2023, these messages between VP of Generative AI Ahmad Al-Dahle and AI researcher Hugo Touvron illustrate the company’s priorities regarding artificial intelligence (AI).
The messages are heavy with industry jargon, but they show that Meta has been heavily focused for years on creating an AI model that can compete with its rivals. They were sent in October 2023, just under a year after the launch of ChatGPT kicked off the current AI revolution.
It’s been over a year since these messages were sent, and the question remains: Can Meta accomplish what it has been trying to do?
A published email chain from a court filing shows that Meta’s AI team has had ambitious model building plans since 2023 (Photo by Artur Widak/NurPhoto via Getty Images)
Can Meta Platforms build the next great AI model?
The world of AI is moving quickly, and keeping track of its developments can be challenging. However, the current boom is often attributed to the launch of OpenAI’s ChatGPT in November 2022.
One thing that the leaked messages between Al-Dahle and Touvron make absolutely clear is that the Meta AI leader sees OpenAI’s models as the gold standard that his much larger company should be striving for.
Related: Mark Zuckerberg makes bold accusations against big tech rival
“We need to learn how to build frontier and win this race,” he stated in one message. “Honestly…our goal needs to be GPT4.”
GPT-4 is a version of OpenAI’s large language model (LLM) system. At the time of Al-Dahle and Touvron’s email conversation, it had only been released roughly eight months prior. As such, it makes sense that it would be the model that Meta’s AI research team would strive to beat.
Since this conversation, though, OpenAI has rolled out several new AI models, including GPT4o, an updated and significantly advanced version of GPT-4. Released in May 2024, OpenAI states that its newer model can “reason across audio, vision, and text in real-time.”
As such, it is likely that Meta is currently aiming to beat the newer GPT-4o. That’s according to HeraHaven.AI founder Komninos Chatzipapas, who notes that Meta’s LLaMa 3.3 AI model is already beating GPT-4.
Chatzipapas thinks Meta has some of SIlicon Valley’s best AI talent, highlighting the recent hiring of tech founder and investor Clara Shih as its head of Business AI. He rates them as a serious contender among AI rivals such as Google (GOOGL) and Anthropic.
More Tech Stocks:
Mark Zuckerberg makes bold accusations against big tech rivalA fierce technology battle erupts within Donald Trump’s fanbase5 quantum computing stocks investors are targeting in 2025
“I definitely think the largest model of the next LLaMa release (likely 70 billion or 405 billion parameters) will beat GPT-4o in the benchmarks,” Chatzipapas predicts.
However, he adds that the comparison is difficult because OpenAI does not publicly release the size of its models.
Chatzipapas flags that Meta’s choice to release fully open-source AI models sets it apart from other tech companies. In his words, “This gives other developers the ability to continue fine-tuning and improve the model after its release.”
What does this mean for the future of AI?
Dev Nag, CEO of QueryPal, describes this decision as a strategic paradox. “Meta is simultaneously trying to build a superior model to OpenAI’s models while actively working to devalue such models through open-sourcing,” he states.
Nag also notes that the 64,000 graphics processing units (GPUs) that Al-Dahle references in one message indicate that Meta believes raw computational power can help it overcome the “first mover advantage” that has helped OpenAI rise to the top of its field.
Related: OpenAI rival startup may be about to blow past its valuation
As he sees it, Meta’s internal focus on beating models like GPT-4 reflects an overall goal of reshaping industry dynamics.
“Al-Dahle’s obsessive focus on GPT-4 as a benchmark reflects Meta’s understanding that they need a clear, measurable goal to rally their organization, with a strategic endgame for controlling the platforms and infrastructure where AI models will ultimately be deployed and monetized,” he explains.
However, Nag also believes that Meta’s willingness to deploy an open-source approach is part of a more aggressive strategy, one that shows how the company really views AI models as a means to an end rather than as products to be directly monetized. But he theorizes that this is part of a much bigger future growth plan.
“The goal appears to be (eventually) establishing Meta as the dominant platform for AI development and deployment,” he states, “making their dominant social networking properties and computing infrastructure the essential substrate for the next generation of social AI applications.”
Related: Veteran fund manager issues dire S&P 500 warning for 2025
