TL;DR:
- Building large language models (LLMs) and generative AI systems is still a nascent field in the enterprise world.
- MosaicML offers a range of options for enterprises, including using existing models, fine-tuning tools, or building models from scratch.
- Blending models and customizing them for specific domains is an emerging concept in the industry.
- MosaicML emphasizes the importance of cost-effective model development, focusing on general capabilities and correctness.
- Enterprises are encouraged to incorporate their own data alongside existing datasets to enhance model performance.
- MosaicML is seeing early adopters putting models into production and continuously improving through feedback loops.
- The value of generative AI is recognized by MosaicML, and they strive to simplify model training through a stable interface.
- Traditional enterprises will take time to fully utilize generative AI, with fintech and healthcare being early adopters.
- Consumer experiences, personalized data utilization, and AI co-pilots are key use cases for generative AI.
- The pace of change in the field is rapid, but generative AI is expected to enhance, rather than replace, human jobs.
- The acquisition of MosaicML by Databricks highlights the growing demand and synergy in the enterprise market.
Main AI News:
In the ever-evolving landscape of enterprise technology, the realm of large language models (LLMs) and other generative AI systems is still shrouded in naivete. These cutting-edge technologies have only recently gained prominence in the mainstream, leaving many businesses to navigate uncharted territory. According to Naveen Rao, the visionary founder and CEO of MosaicML, enterprises have a wide array of options to consider. They can leverage existing models from OpenAI and other sources, fine-tune these tools to suit specific use cases, or even embark on the ambitious journey of building models from scratch. However, Rao points out that the concept of blending models or mixing and matching them is still relatively unexplored.
During a fireside chat with Matt Marshall, the esteemed founder and CEO of VentureBeat, Rao shed light on the current state of affairs. “Everyone is beginning to grasp the potential,” he stated. “We are witnessing an unprecedented shift in perception. Just nine months ago, the majority of individuals were unaware of the existence of large language models like GPT. This rapid transformation is truly remarkable, even in the context of my extensive career.”
Contrary to popular belief, building these models doesn’t necessitate exorbitant costs running into the millions. MosaicML, an organization dedicated to training and deploying LLMs and other generative AI technologies, recently made headlines with its acquisition by Databricks, a leading data lakehouse and AI company. The acquisition, valued at an astounding $1.3 billion, speaks volumes about the potential of this field. Notably, MosaicML developed its MPT-7B model at a relatively modest cost of $200,000.
“We need to dispel the notion that these models require exorbitant budgets,” emphasized Rao. “Organizations must understand that these models don’t need to delve into philosophical discussions about the rise and fall of ancient Rome. What matters is ensuring their general capabilities and correctness for their specific use cases. This is not necessarily the focus of OpenAI’s endeavors,” he explained. Rao also highlighted that many enterprises are still in the early stages of data accumulation, and the subsequent question becomes, “How can we harness this data’s potential through AI?“
Taking customization to the next level, Rao suggests that enterprises should consider pre-training and incorporating their proprietary data alongside existing datasets. He emphasizes that no single model provider can cater to every domain, and it is crucial to empower domain experts with the ability to build models tailored to their specific fields. MosaicML has observed early adopters putting models into production, actively seeking feedback from users, and iterating on their models to build a robust pipeline and a continuous feedback loop.
“It’s a perpetual cycle of innovation and improvement,” Rao affirmed, underscoring the dynamic nature of generative AI. MosaicML’s primary goal has been to establish a stable, cross-cloud interface that simplifies the training of large models. Remarkably, the company has achieved significant milestones with just $35 million in funding since its inception in 2023, boasting an impressive roster of 50 esteemed customers. Rao clarifies that MosaicML maintains a selective approach in choosing its clients, opting for organizations with robust teams and well-structured data.
From its very inception, MosaicML recognized the immense value of AI as a whole, particularly generative AI. “While ChatGPT may be a novelty to many, it was not new to us,” Rao remarked. He humorously referred to the chatbot as “entertaining,” confessing that he initially dismissed it as a passing trend until his teenage children began discussing its capabilities. Startups, he pointed out, possess a unique agility that allows them to take calculated risks, seize opportunities swiftly, and carve out their niches in the market.
Looking ahead, Rao predicts that traditional enterprises will require a few more years to fully embrace and exploit the potential of generative AI. Fintech has consistently been at the forefront of adopting new technologies, and healthcare is also witnessing an uptick in its usage, while big pharma holds significant promise. The most prevalent use cases will revolve around enhancing consumer experiences, exploring novel ways to leverage personalized data for tailored search results, and providing context and personalization. Support automation and the integration of AI co-pilots will serve as crucial tools in these endeavors.
“The pace of change is currently unprecedented, and it instills a sense of unease not only in me but in anyone observing these developments,” Rao admitted. He reassured that the rise of generative AI will not lead to job displacement but rather augment and enhance various professions. Lawyers, doctors, and professionals from diverse fields will have AI-powered co-pilots to support and amplify their capabilities.
Reflecting on the Databricks acquisition, Rao revealed that he had not actively sought a buyout for his company. However, the synergies between MosaicML and Databricks, a renowned enterprise software company serving over 10,000 customers, proved to be compelling. Rao expressed enthusiasm about how MosaicML can seamlessly complement Databricks’ existing infrastructure, creating a powerful amalgamation.
“There is an insatiable appetite for these advancements in the enterprise world,” Rao asserted confidently. “Our goal is to lead the charge and be at the forefront of this transformative journey.” With the relentless pursuit of innovation, MosaicML aims to establish itself as the trailblazer in the realm of generative AI.
Conclusion:
The insights from MosaicML shed light on the evolving landscape of generative AI. Enterprises have a range of options to leverage these technologies, and the concept of blending models is gaining traction. Cost-effective model development and customization for specific domains are key considerations. As generative AI continues to advance, it presents immense opportunities for diverse industries, including fintech, healthcare, and consumer experiences. The rise of generative AI will augment and enhance human capabilities, leading to the emergence of AI co-pilots across various professions. The acquisition of MosaicML by Databricks demonstrates the market’s appetite for these advancements and the potential for strategic partnerships in the enterprise software space.
