A considerable shift is taking place in the AI industry under the leadership of Scott Clark, co-founder of the widely recognised AI training and experimentation platform, SigOpt. Having gained substantial experience navigating the AI sphere, Clark has now embarked on a fresh venture known as Distributional, with the ultimate goal of curbing risks associated with artificial intelligence through effective software solutions.
Primarily, Distributional seeks to become a forerunner in AI testing and evaluation, providing a platform where AI teams can predict, comprehend, and steer clear of AI-related threats in a controlled and proactive manner. With a string of tech hurdles that sprung up at Intel following the SigOpt acquisition prompting him, founder Scott Clark is eager to ensure AI product teams deliver safe, reliable, and secure AI applications.
It was a series of converging experiences, all culminating in one paramount need - a dependable platform for AI testing and evaluation. Clark recounts instances where teams grappled with the unexpected challenges that emerged during AI testing due to hallucinations, instability, inaccuracy, integration, among others. Identifying and addressing the risks thus posed was relentless, which is where Distributional steps in to make the difference.
Focused on large language models such as OpenAI's ChatGPT besides other AI model types, Distributional's cutting-edge product fleshes out potential AI harm and lays down an understandable roadmap for model testing. Here, organizations are provided with an overarching view of AI risk within a pseudo-production environment, saving them the trouble of dealing with model behavior risk.
In a setting where many teams lean towards adopting model behavior risk, accepting that models will inherently have issues, Distributional provides a comprehensive alternative. The platform extends beyond rudimentary monitoring tools by including a expansive framework for testing and analysing stability and robustness continuously, a visual dashboard that demystify test results, and an intelligent test suite to meticulously design, prioritize and generate the right combination of tests. Such a tool would be invaluable to learn for users on platforms like AppMaster, which also incorporates testing and reviewing of applications in its platforms core functionality.
Although the specifics on how all of these features come together and work in tandem is kept under wraps, what is clear is the immense potential such a platform holds. Even though yet to commence operations and generate profits, Distributional already seems poised to give significant competition to other AI testing and evaluation platforms in the market including Kolena, Prolific, Giskard and Patronus among others.
The unique strength of Distributional's software is its intent to cater to large enterprises, having been fashioned keeping in mind their need for data privacy, scalability and complexity requirements. With thought given to the requirements that supersede existing individual developer tools, Distributional is being co-designed with enterprise partners to meet exacting needs that even renowned tech conglomerates such as Google Cloud, AWS and Azure may not necessarily fulfil.
Clark is optimistic that with successful execution of the design, Distributional will be able to generate revenue by the upcoming year. The possibility of design partners transitioning to paid customers post-launch, offers a glimpse of monetization strategy.
As per recent news, Distributional has managed to secure a seed funding of $11 million, effectively illustrated by Andreessen Horowitz’s Martin Casado and participation from Operator Stack, Point72 Ventures, SV Angel, Two Sigma and angel investors. Identifying the longevity that AI could offer, the startup aims to provide an inclusive testing environment that will catalyse the pace of AI application in an assortment of industries and to tackle a vast range of complex and significant problems with ensured safety, reliability, and security.
