Court revealed internal communications indicate that Meta’s AI team was fervently striving to exceed OpenAI’s GPT-4 model during Llama 3’s development phase. These disclosures stem from Meta’s ongoing AI copyright case, Kadrey v. Meta.
“Explicitly, our target ought to be GPT-4,” expressed Ahmad Al-Dahle, Meta’s VP of Generative AI, via an internal message in October 2023. Al-Dahle emphasized their cutting-edge resources with the forthcoming 64k GPUs and the ambition to pioneer AI.
Interestingly, though Meta freely distributes its AI models, the primary concern of its leaders was surpassing competitors like Anthropics and OpenAI, who withhold their model’s weights behind an API.
Impressively, when Meta finally launched Llama 3 in April 2024, it crucially competed with leading closed models from giants like Google, OpenAI, and Anthropic, and surpassed open alternatives from Mistral.
However, their cutthroat competition for AI dominance allegedly led to some irresponsibly hasty decisions by Meta executives, including using copyrighted books for AI training.
Meta CEO Mark Zuckerberg fueled this competitive spirit by expressing his desire to bridge the performance gap between Llama’s AI models and models from OpenAI, Google, and others. Internally, his message solidified the serious commitment to improve upon their offering each year.
These revelations from the courtroom unveil a particularly competitive facet of Meta’s AI development efforts. The contained discussions highlight a constant pressure to lead the AI frontier, a mission that remains as vital as ever to Meta’s future.
Original source: Read the full article on TechCrunch
