MosaicML, co-founded by MIT researchers, is making waves in AI with its open-source tools. Discover how they're changing the game.
 |
| DBRX, a powerful open-source LLM, is challenging the dominance of closed AI models. This could be a game-changer for businesses and researchers. |
Imagine a world where powerful artificial intelligence (AI) models, once the exclusive domain of tech giants, are readily available to businesses and researchers of all sizes. This is the vision that MIT alumnus Jonathan Frankle and Professor Michael Carbin brought to life with MosaicML, a startup on a mission to democratize access to advanced AI.
The problem MosaicML tackled was clear: the most powerful AI models, often built with billions of parameters, were complex, expensive to train, and required significant expertise to use. This locked out smaller companies and researchers who lacked the resources to compete.
Frankle, who began his PhD studies at MIT in 2016, witnessed the nascent field of deep learning firsthand. "Deep learning was a promising area, but not the phenomenon it is today," he recalls. Recognizing its potential, Frankle focused his research on making deep learning models more efficient and accessible.
A turning point came in 2020 when Frankle and Carbin co-authored a paper outlining methods to significantly reduce the size of deep learning models without sacrificing performance. This paper caught the eye of Naveen Rao, a tech executive, who reached out with the idea of starting a company. MosaicML was born.
The MosaicML team focused on building an open-source platform that allowed users to train and improve complex AI models using their own data. This approach not only reduced costs but also empowered users to tailor models to their specific needs. Additionally, MosaicML developed open-source, efficient large language models (LLMs) – powerful AI systems capable of generating text, translating languages, and writing different kinds of creative content.
MosaicML's commitment to open-source resonated with MIT's E14 Fund, which provided crucial early-stage investment. "There was a democratization and an open-source angle to Mosaic's mission," says Frankle, reflecting on his own PhD experience where access to resources was limited. "Why can't everyone participate in this exciting field?"
In 2023, MosaicML's impact caught the attention of Databricks, a global leader in data storage and analytics. Recognizing the potential of MosaicML's technology and its alignment with Databricks' mission of democratizing access to AI tools, Databricks acquired MosaicML for a reported $1.3 billion.
The combined expertise of Databricks and MosaicML led to the creation of DBRX, an open-source LLM exceeding the capabilities of previous models. DBRX offers enterprises and the open-source community access to high-performance AI capabilities previously unavailable outside closed, proprietary systems.
"DBRX showed that you can build the best open-source LLM in the world with Databricks," says Frankle. "This empowers companies to solve specific problems with exceptional efficiency, without being limited by closed models."
Looking ahead, Frankle emphasizes the importance of open-source collaboration in driving AI advancements. "I'm a believer in science and progress, and I'm excited by the cutting-edge research happening in AI," he says. "Openness is critical. MosaicML and DBRX are testaments to how good science and good sharing can propel the entire field forward."
The story of MosaicML and DBRX highlights a significant shift in the world of AI. By making powerful tools accessible to a broader range of players, innovation is no longer limited to the labs of tech giants. This democratization of AI has the potential to unlock a new era of creativity and problem-solving across various industries.