Meta is broadly considered the world’s largest social media firm, but it’s struggling to retain high expertise in its AI division. The corporate has misplaced 11 of the 14 unique authors credited within the whitepaper that launched its AI mannequin Llama in 2023, as reported by Enterprise Insider.
Attending to know the staff behind Llama
Meta formally launched its Llama AI mannequin in February 2023 in a 27-page whitepaper outlining the mannequin’s technical basis. The Meta paper credited 14 researchers for his or her contributions to the AI mannequin’s improvement, together with:
- Naman Goyal: Labored with Meta for six+ years earlier than becoming a member of Considering Machines Lab, an organization centered on creating modular AI architectures for particular functions.
- Eric Hambro: Labored with Meta for 3+ years earlier than becoming a member of Anthropic, the staff behind the Claude household of LLMs.
- Marie-Anne Lauchaux: Labored with Meta for five years earlier than turning into a founding member of Mistral, an organization centered on pure language processing, coding, and textual content era.
- Aurélien Rodriguez: Labored with Meta for almost 3 years earlier than becoming a member of Cohere, an organization specializing in generative and retrieval AI.
All 4 — together with a number of different contributions — have since left Meta to work with competing AI firms. In keeping with Enterprise Insider, solely three of the unique 14 authors stay at Meta: Hugo Touvron, Xavier Martinet, and Faisal Azhar.
SEE: Meta, Google, Amazon, Microsoft Workers Reveal How Tech’s Dream Jobs Turned Cutthroat
Exploring potential authorized and moral points
Meta’s AI ambitions are additional sophisticated by a rising variety of lawsuits alleging using copyright-protected supplies within the coaching of its LLMs. Whereas Meta has denied the claims, critics proceed to query the transparency of its knowledge practices.
In keeping with the unique Llama paper: “We prepare our fashions on trillions of tokens, and present that it’s doable to coach state-of-the-art fashions utilizing publicly out there datasets solely.”
Meta can also be underneath hearth for allegedly sourcing the content material from unlawful, pirated torrents. Regardless of Meta’s reassurance that the corporate solely used publicly out there datasets to coach its AI fashions, the staff has but to resolve these points in court docket.
Paving the way in which for future AI innovation
Whereas engaged on Meta’s Llama AI mannequin appears to learn these builders’ careers — even when they ultimately work for the competitors — these retention points and authorized challenges are main some individuals to be skeptical about the way forward for Llama AI.