Databricks' AI Ambitions Revive IPO Chatter
With the recent introduction of its DBRX large language model (LLM), Databricks has moved with the trend toward empowering enterprise users to leverage generative AI (GenAI) to create applications based on their own data. And the success of DBRX so far raises the question of how close Databricks may be to a potential IPO.
Announced March 27, DBRX is a highly efficient (and therefore cost effective) LLM for general use with enterprise data. It was built by the research team of MosaicML (aka Mosaic AI), a company for which Databricks spent $1.3 billion last year. The goal was to obtain the platform Databricks needed to help enterprises create their own LLMs using their own data. The vendor followed up with the purchase of Arcion, a startup partially funded by Databricks that specializes in replicating data across multiple data management platforms. Arcion was integrated with tech from Mosaic AI to port data from multiple enterprise sources into the Databricks products.
The release of DBRX hints that Databricks’ investments are paying off in a product that ticks many boxes for enterprises looking to adopt GenAI. For one thing, DBRX is a model of efficiency (pun intended). That’s because it is what is known as a “mixture of experts” model, a technically complex approach that results in streamlining the process of a model’s absorbing and acting on information. That means less time spent computing with expensive GPUs – a goal for any enterprise.
A Competitive Offering
Its efficiency allows DBRX to perform better than competing models, Databricks claims, in language understanding, programming, and math, as shown in the chart below:
Access CLOUD TRACKER PRO
|
|
Related Articles
Panzura Conducts Object Storage Symphony
Panzura has released Symphony, a platform that supports data from a variety of object storage systems, including ones from AWS, Azure, GCP, and Wasabi
Barclays Expands a Hybrid Cloud Deal with HPEThe UK’s Barclays bank has extended a ten-year contract with Hewlett Packard Enterprise (HPE) for management of its multicloud and hybrid cloud estate
Cloud Native Climbs at KubeConCNCF takes on AI -- and much more