Artificial intelligence (AI) models today have become increasingly complex with billions of parameters. Existing AI models are largely inaccessible to many due to a lack of widespread knowledge of how to create and control them. MosaicML, a company co-founded by Jonathan Frankle PhD ’23 and MIT Associate Professor Michael Carbin, strives to overcome this issue. They provide a platform that allows users to customize, monitor, and enhance open-source models with their data, thus democratizing the AI technology.
MosaicML has already proved instrumental in making deep learning more accessible to a wider range of organizations, especially with the increasing interest in generative AI and large language models (LLMs). Moreover, they provide powerful tools for data management companies, enabling organizations to effectively utilize their data without necessitating any third-party involvement.
MosaicML’s promising approach led to its acquisition by Databricks, a renowned data storage, AI, and analytics company that serves some of the world’s largest companies. Post-acquisition, the companies have developed and launched DBRX, one of the most efficient general-purpose LLMs, renowned for its swift performance and utility in large enterprises. The model has shown considerable proficiency in tasks such as reading comprehension and solving logic puzzles and is freely available for customization.
Frankle, who previously earned his bachelor’s and master’s degrees in computer science from Princeton University, came to MIT to pursue his PhD in 2016. His interest in deep learning led to the establishment of MosaicML. With the release of a transformative deep learning architecture paper by Google in 2017 and an unsolicited email from eventual Mosaic co-founder Naveen Rao in 2020, Frankle’s journey in making AI accessible took off.
MosaicML’s progress has enabled a host of companies to develop their own AI models, democratizing AI technology. In line with their mission, MosaicML’s techniques and methods were open-sourced in 2023, giving developers more visibility and flexibility in training and running models. Following MosaicML’s acquisition by Databricks, the latter released DBRX, a breakthrough in providing open-source access to premium AI capabilities.
Frankle optimistically views the precedent set by DBRX, the high-performing open-sourced LLM, asserting its potential in nurturing a culture of openness and sharing in AI. This shift is crucial for frontier-pushing AI development, as it promotes collaboration over competition.
