Galileo’s AI trust layer aims to enhance reliability of generative AI projects

As the integration of generative artificial intelligence (GenAI) continues to proliferate across various industries, recent studies indicate that only a small fraction of GenAI projects have successfully transitioned into production. Concerns over large language models (LLMs) generating unreliable or misleading responses—often referred to as “hallucinations”—are significantly impeding these advancements, prompting organizations to explore solutions that include implementing trust layers for AI.

Generative models, particularly LLMs, possess the capability to process vast amounts of unstructured data, allowing them to generate responses based on learned patterns. This potential has made them valuable tools for developing applications such as chatbots and semi-autonomous agents capable of performing complex language tasks. However, one of the primary challenges users face is the unpredictability of these models. Due to their non-deterministic nature, LLMs might produce inconsistent and often erroneous outputs, necessitating the establishment of monitoring and control mechanisms.

Organizations are turning to AI trust layers as a remedy for these failures. Take Salesforce, for instance, which has incorporated several methodologies within its Einstein AI models to mitigate the incidence of errors, including secure data retrieval and toxicity detection. Despite these efforts, other companies are seeking more versatile AI trust layers that can be integrated with a variety of GenAI platforms and LLM models. One such enterprise, Galileo, seeks to fill this niche.

Founded in 2021 by engineers Yash Sheth, Atindriyo Sanyal, and Vikram Chatterji, Galileo aims to address the challenges posed by non-deterministic models. With a background that includes a decade at Google developing LLMs for speech recognition, Sheth highlights the importance of effectively managing these models for successful enterprise application. “We saw that LLMs are going to unlock 80% of the world’s information, which is unstructured data,” Sheth remarked in an interview with BigDATAwire at the recent re:Invent conference. However, he noted that the inherent unpredictability of LLMs poses significant hurdles for their adoption in enterprise environments.

Recognizing the critical need for a robust trust framework, Galileo has focused on developing its AI trust layer designed specifically for generative models. The company aims to ensure that LLMs can be deployed in production settings while adhering to strict privacy and security standards. Sheth emphasized, “To actually mitigate the risk when it’s applied to mission critical tasks, you need to have a trust framework around it that can ensure that these models behave the way we want them to be, out there in the wild, in production.”

In developing its first product, Generative AI Studio, launched in August 2023, Galileo adopted a comprehensive approach over the past two years, conducting research and creating performance metrics for LLM behavior. The company has instituted mechanisms that activate guardrails to prevent undesirable outputs. These metrics function in real time to ensure the reliability of LLMs, blocking potential hallucinations and harmful interactions.

Galileo’s AI trust layer comprises three components: Evaluate, which allows for experimentation with a customer’s GenAI stack; Observe, which consistently monitors LLM performance to foster user satisfaction; and Protect, which safeguards against harmful requests. This framework enables enterprises to confidently deploy their GenAI applications akin to traditional software, with assurance that any issues will be promptly addressed.

The traction Galileo has gained since its inception is noteworthy. The company counts several Fortune 100 entities, including Comcast, Twilio, and ServiceNow, among its clients and established a partnership with HPE in July. Furthermore, Galileo successfully completed a $45 million Series B funding round in October, resulting in a total of $68.1 million raised.

As enterprises strive to commercialize their GenAI initiatives, the need for effective AI trust layers becomes increasingly paramount. Sheth underscores the potential transformational impact of GenAI on mission-critical software. With the right safeguards in place, he asserts that these innovative technologies can redefine operational capacities across various sectors.

In parallel to these developments, experts from Ernst & Young (EY) have recently advised on responsible GenAI development, stressing the importance of accounting for ethical implications in deploying these powerful technologies. This perspective aligns with a broader call for responsible practices as organizations navigate the complexities of integrating AI into existing systems. As organizations continue to assess their GenAI projects, the success of these initiatives may rely heavily on establishing the kind of robust trust layers exemplified by companies like Galileo.