Snowflake's Enterprise-Grade Leap in Generative AI

In the dynamic domain of artificial intelligence (AI), the dominance of versatile and broadly applicable generative models has long been recognized. Yet, with the rapid expansion of the generative AI landscape, fueled by the participation of cloud vendors of varying sizes, a new wave of models tailored specifically for enterprise clientele is making its mark.

Snowflake, a notable player in cloud computing, unveiled Arctic LLM, a generative AI model positioned as “enterprise-grade.” Released under the Apache 2.0 license, Arctic LLM is meticulously crafted to excel in “enterprise workloads,” including the intricate task of generating database code—a capability lauded by Snowflake as instrumental for driving enterprise innovation. Remarkably, this model is available for both research and commercial endeavors, underscoring Snowflake’s commitment to democratizing AI advancements.

Pioneering Enterprise Solution

Sridhar Ramaswamy, CEO of Snowflake, expressed fervent optimism about Arctic LLM’s potential during a press briefing, “I think this is going to be the foundation that’s going to let Snowflake — and our customers build enterprise-grade products and actually begin to realize the promise and value of AI. You should think of this very much as our first, but big, step in the world of generative AI, with lots more to come.” Said CEO Sridhar Ramaswamy.

Arctic LLM, the flagship model in Snowflake’s family of generative AI models called Arctic, stands as a testament to the company’s dedication to innovation. Developed over approximately three months, harnessing the power of 1,000 GPUs and a substantial investment of $2 million, Arctic LLM emerged in the wake of Databricks’ DBRX, another generative AI model targeting the enterprise space.

Arctic LLM Versus Industry Contenders

Snowflake boldly compares Arctic LLM with its counterparts, citing its superior performance in coding tasks and SQL generation compared to DBRX. Furthermore, Snowflake asserts Arctic LLM’s prowess over Meta’s Llama 2 70B (still not over Llama 3 70B) and Mistral’s Mixtral-8x7B, positioning it as a frontrunner in the enterprise AI landscape. Notably, Snowflake claims Arctic LLM achieves “leading performance” on the MMLU benchmark; it includes tests that can be solved through rote memorization.

Baris Gultekin, head of AI at Snowflake, emphasized Arctic LLM’s tailored focus on addressing specific enterprise challenges. He told TechCrunch in an interview that “Arctic LLM addresses specific needs within the enterprise sector, diverging from generic AI applications like composing poetry to focus on enterprise-oriented challenges, such as developing SQL co-pilots and high-quality chatbots.”

Arctic LLM, akin to its counterparts DBRX and Google’s Gemini 1.5 Pro, adopts a mixture of expert (MoE) architecture, incorporating efficiency and specialization in data processing tasks. Despite its colossal scale, with 480 billion parameters, Arctic LLM selectively activates 17 billion parameters at a time, leveraging 128 expert models—a testament to Snowflake’s commitment to efficiency and cost-effectiveness.

According to Snowflake’s claim, this structured design enabled it to train Arctic LLM on open public web data sets (including RefinedWeb, C4, RedPajama and StarCoder) at “roughly one-eighth the cost of similar models.”

Read More: India Abandons Proposal for Mandatory Approval of AI Model Launches

Deployment and Accessibility

Snowflake ensures accessibility and deployment ease for Arctic LLM by providing comprehensive resources such as coding templates and curated training sources. Acknowledging the potential complexity and costliness of deploying and fine-tuning the model—requiring around eight GPUs—Snowflake promises to broaden accessibility by making Arctic LLM available across various hosting platforms, including Hugging Face, Microsoft Azure, Together AI’s model-hosting service, and the enterprise-focused generative AI platform Lamini.

Arctic LLM will debut on Cortex, Snowflake’s platform tailored for building AI- and machine learning-powered applications and services. Snowflake positions Cortex as the premier choice for deploying Arctic LLM, emphasizing its robust features encompassing security, governance, and scalability. 

“Our dream here is, within a year, to have an API that our customers can use so that business users can directly talk to data, It would’ve been easy for us to say, ‘Oh, we’ll just wait for some open source model and we’ll use it. Instead, we’re making a foundational investment because we think it’s going to unlock more value for our customers.” said CEO Ramaswamy.

With Arctic LLM’s unveiling, a crucial question emerges: who stands to benefit from this enterprise-focused model besides Snowflake’s existing customers? 

Limitations and Hallucinations

In a landscape teeming with “open” generative models adaptable to diverse purposes, Arctic LLM’s unique value proposition may not be immediately apparent. While its architecture promises efficiency gains, it remains to be seen whether these advantages will suffice to sway enterprises from established alternatives like GPT-4 and others.

A critical aspect to consider is Arctic LLM’s relatively small context window, from 8,000 and 24,000 words, far lower than models like Anthropic’s Claude 3 Opus and Google’s Gemini 1.5 Pro, a factor influencing its ability to retain information from previous interactions. In the realm of generative AI, models with larger context windows tend to exhibit more robust performance by retaining the context of recent conversations. Arctic LLM’s context range, while substantial, falls short compared to competitors, potentially impacting its ability to generate coherent and contextually relevant outputs.

While Snowflake does not explicitly address it, Arctic LLM likely grapples with limitations inherent to generative AI models, including the risk of hallucinations—instances where the model confidently generates incorrect responses. These shortcomings, stemming from the model’s statistical nature and small context window, underscore the ongoing challenges facing the generative AI domain. Despite incremental improvements, substantial breakthroughs are necessary to address these fundamental limitations.

As we navigate the evolving landscape of generative AI, Snowflake’s unveiling of Arctic LLM represents a significant milestone, heralding a new era of enterprise-focused AI solutions. While the road ahead may be fraught with challenges and uncertainties, the relentless pursuit of innovation remains the driving force propelling the field forward.

Read More: Apple Acquires AI Startup Specializing in Supervising Manufacturing Components