The heavily funded and previously somewhat secretive artificial intelligence startup SambaNova Systems Inc. announced today what it says is the first AI platform that’s purpose-built for the financial services industry.

The SambaNova GPT Banking service is designed to enable banks to jump start their AI capabilities in as little as a few days or weeks at the most.

With more than $1 billion in funding behind it and a massive $5 billion valuation, SambaNova has designed and built an integrated hardware and software platform that it says can run AI and data-intensive applications from the data center to the edge. Its DataScale hardware platform uses custom seven-nanometer chips that SambaNova says are more attuned to machine learning and deep learning processes than the general-purpose microprocessors and graphics processing units widely used for such tasks today.

The platform’s Reconfigurable Dataflow Architecture runs an integrated open-source software stack called SambaFlow that helps run each machine learning model optimally. It does that by minimizing the need to interface with memory, eliminating the main bottleneck in AI, which is the interconnect between chips and memory.

SambaNova GPT Banking is the first industry-specific offering to run on that stack. It’s being offered as a subscription service to help banks and other financial services firms deploy advanced machine learning models in a fraction of the time it would normally take to do so, the company said.

SambaNova said there’s a big need for such a solution, as many financial services firms struggle to deploy AI quickly enough to meet their needs. SambaNova calls this the “deep learning deployment gap,” explaining that it takes banks an average of 18 months to hire a skilled data science team, build the infrastructure it needs then train and deploy large language models. By that time, the models are already out of date, as the base model size will have increased by 10 times while the compute requirements will have grown dramatically.

“We built GPT Banking to improve bank’s competitiveness and efficiency while accelerating their digital transformation,” said SambaNova co-founder and Chief Executive Rodrigo Liang (pictured). “AI is the quickest and most cost-efficient tool to do that today and our service can be deployed and delivering value in weeks.”

SambaNova said GPT Banking has multiple uses, including sentiment analysis, scanning social media posts, traditional media and other sources to understand financial markets and investor sentiment. It can help with entity recognition, reducing human error and classifying documents to reduce manual, repetitive work. It also performs language generation tasks such as processing, transcribing and prioritizing claims and applications, creating documents and extracting information from them, and also translating documents into different languages.

“GPT Banking has the potential to transform nearly every aspect of banking, from improving operations to managing risk and compliance,” Liang added.

Analyst Andy Thurai of Constellation Research Inc. told SiliconANGLE that SambaNova’s GPT Banking model is similar to Open AI LLC’s Generative Pre-trained Transformer 3 but with a specific knowledge towards finance industry. GPT-3 is an auto-regressive language prediction model that uses deep learning to produce and analyze human-like text.

Thurai said that although GPT-3 has shown itself to be very capable in text analysis, it has many issues, including the fact it was trained on publicly available text that likely has a lot of bias and is not cleansed properly. As a result, it can end up producing biased models that deliver biased results, he said.

“The text, thought processes and assets from regulated industries such as finance and insurance never make it to the public domain, so models trained by OpenAI’s GPT-3 are not accurate when it comes to industry-specific topics in regulated industries,” he said. “Given the ties between OpenAI and Microsoft Azure, it is conceivable that GPT-3 is not completely open but somewhat controlled by Microsoft’s interests.”

With that in mind, Thurai said, it’s great to see SambaNova bringing a banking- and finance-specific GPT to market.

“While all the work associated in NLP areas such as sentiment analysis, entity recognition, language generation, and language translation has existed before from providers such as OpenAI and Google, this finance industry-specific solution set will hopefully add a domain-specific layer of knowledge to the original GPT,” he said. “This used to be a huge limitation in regulated industries adopting AI largely.”

The launch of GPT Banking suggests SambaNova is ready to start serving enterprises on a much bigger scale. Up until now it has always been very secretive about the companies using its platform, though it was previously rumored to be working on projects related to the U.S. Defense Advanced Research Projects Agency.

Photo: SambaNova Systems