Normal Computing startup was founded by former members of the Google Brain Team, Palantir, and X Engineers who have extensive experience in building core AI production systems for Alphabet. The startup has also collaborated with other leading AI companies in the United States and is initiating pilots with Fortune 500 companies across multiple sectors, including semiconductor manufacturing, supply chain management, banking, and government agencies.
Normal Computing is focused on developing full-stack probabilistic compute infrastructure for complex artificial intelligence (AI) applications, has announced that it has successfully raised $8.5 million in a Seed funding round. The funding was led by Celesta Capital and First Spark Ventures, with participation from Micron Ventures. The investment will enable Normal Computing to further its mission of empowering large companies to leverage technologies like Generative Artificial Intelligence (Generative AI) in real-world contexts that involve intricate and high-stakes scenarios. Additionally, the funds will support the research and development of Normal Computing’s application development platform and Probabilistic AI technology.
While large general-purpose AI models like OpenAI’s GPT-4 have garnered significant attention, they still face challenges related to reliability, such as unpredictable factual errors and “hallucinations.” These limitations may be acceptable for early consumer applications, but they present significant hurdles in advancing core enterprise workflows that have yet to fully unlock AI’s transformative value creation potential. Faris Sbahi, the CEO and Co-Founder of Normal Computing, highlights these challenges and emphasizes the need for solutions that address them effectively.
Normal Computing’s Probabilistic AI paradigm aims to overcome these obstacles by providing unprecedented control over the reliability, adaptivity, and auditability of AI models powered by the private data of their customers. Leveraging their experience in developing AI workflows at Alphabet, the company focuses on addressing use cases where risk has been a significant barrier to AI adoption. These use cases span a wide range of applications, including automating complex underwriting processes, generating and validating specialized code adhering to mission-critical constraints, and mitigating risks in dynamic airline supply networks.
One of the limitations of typical Large Language Models (LLMs) deployed in assisting financial advisors is their tendency to provide inaccurate or outdated information that may not be relevant to decision-making. Furthermore, these models often fail to offer transparent reasoning that would enable effective auditing. In contrast, Probabilistic AI allows models to detect instances of inaccurate synthesis and generate probable, auditable explanations of their decision-making process. These models can also adapt and revise themselves by making additional queries to datastores or involving humans in the loop.
Nicholas Brathwaite, Founding Managing Partner at Celesta Capital, emphasizes the importance of reliable and transparent AI systems that understand the limitations of their own reasoning. He expresses excitement about supporting Normal Computing’s development of cutting-edge Probabilistic AI, which plays a crucial role in building trustworthy AI solutions for critical public and private applications.
Normal Computing’s Probabilistic AI enhances existing models such as LLMs and Diffusion Models while enabling the development of new architectures. The company believes that integrating these large models into composed workflows with their Probabilistic AI technology, alongside specialized models, enterprise-specific plugins, and domain-specific processes, has the potential to solve complex real-world problems. Normal’s technology ensures the reliable deployment of these large AI systems, allowing them to detect and address failures like hallucinations, while also learning and adapting to private data and changing conditions in real-time.
Faris Sbahi emphasizes Normal’s commitment to collaborating with clients to enable applications that involve multiple stakeholders, complex data landscapes, and sophisticated security policies. While significant advancements have been made in scaling transformer models with GPUs, there remains a gap between these capabilities and the requirements of real-world production use cases, where data is often incomplete, noisy, and constantly changing. Faris states that the trend of doubling down on certain architectures and approaches is driven by their compatibility with conventional tools and infrastructure, rather than their trustworthiness or understandability.
To address this challenge, Faris emphasizes the need to redesign AI systems from the ground up, going beyond surface-level approaches like prompt engineering and retrieval-based methods. Normal Computing is enthusiastic about the support from investors in the Seed funding round, which will allow them to confront this challenge directly and develop principled systems that enable and advance AI solutions for their partners.
Normal Computing recognizes that transparency and openness are essential for the adoption of AI systems. As part of their commitment to transparency, the company provides AI systems backed by customizable open source models, similar to Stanford University’s Alpaca, ensuring full auditability. This approach stands in contrast to closed systems like OpenAI’s GPT-series, which keep their internals hidden. Normal’s system is designed to protect a company’s proprietary information, ensuring that there is no uncertainty about how their data is being used. This auditable approach enables businesses to maintain the integrity of their ground truth. Normal Computing is actively contributing to the open-source community by making some of its developer tools available for reliable Generative AI workflows.
Manish Kothari, Founding Managing Partner at First Spark Ventures, highlights the importance of ensuring responsible and reliable AI systems, particularly in domains like discovering new materials, nanotechnology, biology, and medicine. He commends Normal’s thoughtful approach to enabling this kind of high-potential technology.
As I conclude, Normal Computing’s Seed funding round will help propel the company’s efforts in developing full-stack probabilistic compute infrastructure for complex AI applications. By leveraging its Probabilistic AI technology, Normal Computing aims to overcome reliability challenges and enable the adoption of AI in critical and high-stakes real-world scenarios. The company’s focus on transparency, reliability, and adaptability positions it as a trusted partner for enterprises seeking to harness the transformative potential of AI.