Generative AI and large language models (LLMs) are transforming the way businesses use data and create value. But how can enterprises leverage these technologies without losing control of their data, spending millions of dollars, or relying on third-party models?
That’s the question that Naveen Rao, co-founder and CEO of MosaicML, a leading generative AI platform, will answer at the VB Transform event on July 12, 2023 in San Francisco. Rao will share his insights and best practices on how to build, own, and secure best-in-class generative AI models with your own data.
MosaicML is known for its state-of-the-art MPT LLMs, which are open-source and commercially licensed. With over 3.3 million downloads of MPT-7B and the recent release of MPT-30B, MosaicML has showcased how organizations can quickly build and train their own state-of-the-art models using their data in a cost-effective way. Customers such as AI2 (Allen Institute for AI), Generally Intelligent, Hippocratic AI, Replit and Scatter Labs leverage MosaicML for a wide variety of generative AI use cases.
In June 2023, MosaicML was acquired by Databricks, a data and AI analytics provider, for $1.3 billion. Rao says that the acquisition was a strategic decision that will enable his company to accelerate its mission of democratizing generative AI and making the lakehouse (the term used by Databricks to refer to its data offering) the best place to build generative AI and LLMs.
In a recent interview with VentureBeat, Rao explained why he is critical of fine tuning as a way to customize LLMs, why he prefers to train models from scratch with the right data mix, and why he thinks open source models will eventually overtake closed models.
“Fine tuning is a way to condition a model to behave in certain ways. It’s not a way to make something really domain specific. The right way to do it is actually to train a model from scratch,” Rao said.
He also said that using a vector DB or a prompt can work well for some cases, but not for others. “You always want to use a minimum effort to solve the problem at hand. In some cases, you can solve it with a vector DB, like pinecone. And you know, you can do that in a way that respects some privacy. You’re basically using a prompt with the vector db to modify the behavior of the large language model. That has its place,” he said.
He also made a bold prediction that open source models will eventually overtake closed models, just like Linux did with Solaris. “Open source is really just getting started. So in five years, I think the world will look a bit different,” he said.
If you want to learn more about generative AI and how to leverage it for your business, don’t miss Naveen Rao’s session at VB Transform on July 12, 2023 in San Francisco. Register now and get ready to join the generative AI revolution!
Join top executives in San Francisco on July 11-12 and learn how business leaders are getting ahead of the generative AI revolution. Learn More
Generative AI and large language models (LLMs) are transforming the way businesses use data and create value. But how can enterprises leverage these technologies without losing control of their data, spending millions of dollars, or relying on third-party models?
That’s the question that Naveen Rao, co-founder and CEO of MosaicML, a leading generative AI platform, will answer at the VB Transform event on July 12, 2023 in San Francisco. Rao will share his insights and best practices on how to build, own, and secure best-in-class generative AI models with your own data.
MosaicML is known for its state-of-the-art MPT LLMs, which are open-source and commercially licensed. With over 3.3 million downloads of MPT-7B and the recent release of MPT-30B, MosaicML has showcased how organizations can quickly build and train their own state-of-the-art models using their data in a cost-effective way. Customers such as AI2 (Allen Institute for AI), Generally Intelligent, Hippocratic AI, Replit and Scatter Labs leverage MosaicML for a wide variety of generative AI use cases.
In June 2023, MosaicML was acquired by Databricks, a data and AI analytics provider, for $1.3 billion. Rao says that the acquisition was a strategic decision that will enable his company to accelerate its mission of democratizing generative AI and making the lakehouse (the term used by Databricks to refer to its data offering) the best place to build generative AI and LLMs.
Event
Transform 2023
Join us in San Francisco on July 11-12, where top executives will share how they have integrated and optimized AI investments for success and avoided common pitfalls.
In a recent interview with VentureBeat, Rao explained why he is critical of fine tuning as a way to customize LLMs, why he prefers to train models from scratch with the right data mix, and why he thinks open source models will eventually overtake closed models.
“Fine tuning is a way to condition a model to behave in certain ways. It’s not a way to make something really domain specific. The right way to do it is actually to train a model from scratch,” Rao said.
He also said that using a vector DB or a prompt can work well for some cases, but not for others. “You always want to use a minimum effort to solve the problem at hand. In some cases, you can solve it with a vector DB, like pinecone. And you know, you can do that in a way that respects some privacy. You’re basically using a prompt with the vector db to modify the behavior of the large language model. That has its place,” he said.
He also made a bold prediction that open source models will eventually overtake closed models, just like Linux did with Solaris. “Open source is really just getting started. So in five years, I think the world will look a bit different,” he said.
If you want to learn more about generative AI and how to leverage it for your business, don’t miss Naveen Rao’s session at VB Transform on July 12, 2023 in San Francisco. Register now and get ready to join the generative AI revolution!
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.
Author: Matt Marshall
Source: Venturebeat
