Databricks’ decision to acquire MosaicML addresses a challenging concern: developing LLMs for artificial intelligence(AI) at a lower cost point and with fewer compute resources. Mosaic helps organizations develop their LLMs based on corporate data, not publicly accessing content. The move by Databricks also helps lower the cost for organizations to create and utilize their own LLMs versus licensing existing LLMs.
Databricks’ strategy incorporates extending the client’s ability to create 15 times lower-cost LLMs within the Lakehouse environment. With the announcement of LakeHouseIQ, this platform will bring together Databricks Lakehouse, MosaicML, and the Unity catalog as one offering. Like other LLM creators, Databricks will rent this platform to their clients.
The cost for licensing LLMs today could range in the millions of dollars. MosaicML helps drive down the cost of inference and training, and the process will also require fewer hardware platforms by optimizing GPU technology.
This hardware offload efficiency gives Databricks several advantages compared to open-source and closed platforms. With access to a lower LLM capability based on the organization’s data, the learning process becomes more efficient with fewer data management requirements at the process’s inception; Databricks extends the process of fine-tuning in their offering.
Organizations tapping their data into their LLM process could cut costs by 50% or more. More importantly, organizations moving ahead with AI and ML across several areas within their enterprise will have more confidence in security and privacy because the LLMs become based on their data.