“We need to learn how to build frontier, and win this race…”
Ahmad Al-Dahle, Meta’s VP of GenAI instructions by private message to Meta AI researcher Hugo Touvron could not have been more clear on the intense focus of the mission.
The internal messages and emails were revealed as part of the discovery process in the the case of Kadrey v Meta before the District Court of California.
The until now private communications reveal a desperate internal scramble at Meta to gazump OpenAI’s leading large language model, GPT4.
The communications, date from late 2023, and lay bare the extreme focus for Meta on winning the AI arms race.
The strategic importance of winning the AI race for the magnificent 7 is hardly a well kept secret, and in early 2024, shortly after these communications occurred Meta confirmed it had placed an order for 350,000 H100 GPU’s from Nvidia, with the purchase price totalling just shy of $100 billion. Such is the importance of placing well in this ‘winner takes all’ affair that is AI development.
Addressing France’s Mistral LLM, Al-Dahle was dismissive, “mistral is peanuts for us”.
The chat turns to the MMLU score of GPT4 (which at that date scored 86.4%), with Meta confident it could better this.
These internal messages from Meta have come to light through litigation involving a number of disgruntled author suing Meta for breach of their copyright works (in this instance, books).
The backdrop of these proceedings is a larger strategic issue faced by Meta: training of LLMs on datasets riddled with copyright content (potentially without permission from the owners) may have occurred on a far broader scale, and the resulting legal issues that they are now brushing with could prove to be problematic downstream as foundational model owners (such as Meta) look to monetise widespread usage of their products.
The issue of usage of copyright works in datasets allegedly without permission is certainly not an issue isolated to Meta,
Meta CEO Mark Zuckerberg’s considerable and very public recent efforts to cosey up with the incoming Trump Administration suggests that he anticipates significant regulatory headwinds for the company, many of which may require regulatory framework change.
HARNIS specialises in the safe, efficient and ethical implementation and usage of AI by business.
If you’d like to discuss the strategic implementation of AI in your business, please reach out to us at hello@harnis.ai
