Meta Unveils LLaMA AI Model to Challenge ChatGPT in Research Sector
Meta releases LLaMA, a collection of AI language models trained on public datasets that outperform GPT-3 on most benchmarks. The technology is available exclusively to researchers to advance academic and industry studies.
Meta introduces LLaMA, a new collection of foundation language models designed to compete in the rapidly growing generative AI space. The models range from 7 billion to 65 billion parameters and serve as powerful tools for generating text, summarizing documents, and solving complex problems like math theorems and protein structure predictions.
Unlike some competitors, Meta trains these models exclusively on publicly available datasets rather than relying on proprietary data. This approach yields impressive results, as the mid-sized LLaMA-13B model outperforms the massive GPT-3 on most benchmarks, while the largest LLaMA-65B model remains highly competitive with top industry standards like Chinchilla and PaLM.
Meta makes the weights for all LLaMA models openly available, but restricts access to academic researchers, government bodies, civil society organizations, and industry research labs. This targeted release strategy allows the scientific community to advance AI research safely while Meta establishes a strong position against popular rivals like ChatGPT.