6.S963 Final Paper 7/1/2024

Pablo Omeñaca Muro
Massachusetts Institute of Technology

Abstract

In recent years, the landscape of artificial intelligence has been significantly transformed by the evolution of large language models (LLMs). These models, capable of understanding and generating human-like text, have become central to a wide range of applications, from natural language processing to more complex problem-solving tasks. Among the various developments in this field, the emergence of open-source LLMs is transforming the access and use of this technology which raises questions that deserve a thorough analysis.

Historically, the development and deployment of LLMs were predominantly dominated by proprietary models, developed in closed ecosystems. Even though these models were partially released open-source to gain fast adoption, the intended approach was closed-source. The journey of open-source LLMs, from their early iterations to the near state-of-the-art models like LLaMA 3 and Falcon 40B, exemplifies a remarkable period of growth and development.

Despite the advantages of high-performing, open-source large language models (LLMs), their development raises ethical concerns. For instance, the accessibility of this technology allows malicious actors to exploit it for harmful purposes.

This short paper explores the impact of open-source LLMs on the AI landscape, examining their development, the challenges they address, and the opportunities they present for the future of AI research and application.

Full Paper

Disclaimer

This paper was written for Alfred Spector’s MIT Spring 2024 course 6.S963 Beyond Models – Applying Data Science/AI Effectively. It has not been peer-reviewed, and it may contain errors.