Ex-Google engineers’ probabilistic, AI startup raises seed funding
Cette publication existe aussi en Français
Probabilistic AI computation startup Normal Computing Inc. (New York, NY) has said it has raised US$8.5 million in a seed round of funding.
The company was founded by former members of the Google Brain Team and X Engineers who built generative AI production systems for Alphabet. Normal Computing was founded in 2022 and funding has come from Celesta Capital and First Spark Ventures, with participation from Micron Ventures.
The company intends to build a “full-stack” probabilistic compute infrastructure enabling artificial intelligence (AI) for critical and complex applications.
At the heart of the approach is AI hardware based on thermodynamic processes. The fundamental building blocks are inherently stochastic — which means that noise becomes a crucial resource for computation. Normal reckons it has way to use this property to transcend limitations of general-purpose generative-AI, analog AI and quantum AI.
‘Hallucinations’
Those limitations, such as unpredictable factual errors, pose challenges for professional and enterprise adoption of generative-AI and are leading to broader discussion in society of how generative-AI should be deployed and possibly curtailed.
Probabilistic AI is a paradigm that may solve these and other roadblocks by giving control over reliability, adaptivity, and accountability to AI models powered by customers’ private data. Normal Computing is supporting applications such as automated insurance underwriting and workflows for generating and validating software code that adheres to mission-critical constraints.
Normal Computing asserts that probabilistic AI models can detect when they synthesize inaccurately by also generating probable, auditable explanations of how they reached a conclusion. They can even revise themselves by adaptively making an additional query to a datastore or human-in-the-loop.
“Artificial Intelligence has the potential to address some of the greatest human challenges of our time. But in order to do so, it must be reliable, transparent, and able to comprehend the limits of its own reasoning so that it knows how best to engage and explain to humans in the loop,” said Nicholas Brathwaite, founding managing partner at Celesta Capital.
Probabilistic AI can enhance large language models (LLMs) and diffusion models, as well as enable new architectures, Normal Computing said.
Related links and articles:
News articles:
AI boom could drive 25 percent of Broadcom chip sales in 2024
AI server shipments to leap up 40 percent in 2023
Trillion-dollar Nvidia subsidized to create Taiwan AI R&D center
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
