Large language model (LLM) watermarks enable authentication of text provenance, curb misuse of machine-generated text, and promote trust in AI systems. Current watermarks operate by changing the next-token predictions output by an LLM. The updated (i.e., watermarked) predictions depend on random side information produced, for example, by hashing previously generated tokens. LLM watermarking is particularly challenging in low-entropy generation tasks -- such as coding -- where next-token predictions are near-deterministic. In this paper, we propose an optimization framework for watermark design. Our goal is to understand how to most effectively use random side information in order to maximize the likelihood of watermark detection and minimize the distortion of generated text. Our analysis informs the design of two new watermarks: HeavyWater and SimplexWater. Both watermarks are tunable, gracefully trading-off between detection accuracy and text distortion. They can also be applied to any LLM and are agnostic to side information generation. We examine the performance of HeavyWater and SimplexWater through several benchmarks, demonstrating that they can achieve high watermark detection accuracy with minimal compromise of text generation quality, particularly in the low-entropy regime. Our theoretical analysis also reveals surprising new connections between LLM watermarking and coding theory. The code implementation can be found in https://github.com/DorTsur/HeavyWater_SimplexWater

JACKSONVILLE, Fla., June 04, 2018 (GLOBE NEWSWIRE) -- Today, Black Knight, Inc. (NYSE:BKI) announced that it has acquired HeavyWater, a provider of artificial intelligence and machine learning (AI/ML) to the financial services industry. Black Knight will be integrating AIVA (pronounced Ay-vah) into its premier solutions, while also making the technology available to clients who seek to deploy AI/ML within other parts of their organizations to help enhance efficiency, effectiveness and accuracy. By taking on manual tasks that have diverted resources away from value creation projects, AIVA reads, comprehends and draws conclusions based on context to mimic cognitive thinking and build expertise over time. Located in Philadelphia, HeavyWater has been providing this solution to help lenders verify income, assets and insurance coverage, which are traditionally manual activities that take hours to complete and are prone to error. Clients benefit from accelerated processes and reduced expenses as AIVA gains experience and manual routines are automated.

With its acquisition of artificial intelligence and machine learning developer HeavyWater, Black Knight is turning to its Artificial Intelligence Virtual Assistant to streamline the mortgage process, with an immediate focus on the originations sector. The company, which acquired the Philadelphia-based startup late last week, aims to eventually get to a place where tech can guide and advise underwriters on potential opportunities and in decision making, according to Rich Gagliano, president of Black Knight's Origination Technologies division. "AIVA makes the decisions and recommends actually a conclusion, the operator can agree or disagree, and as a result of that two things happen: that activity or that job task gets done, and in certain situations, creates a reinforcement learning for AIVA," said Soofi Safavi, CEO of HeavyWater. "That activity ultimately will be shadowed by AIVA to be able to actually give her the same level of cognition that a human expert has," he added. Though AIVA's primary focus for the short term will be on the originations sector, it will also eventually support mortgage servicing.

Black Knight announced Monday its acquisition of HeavyWater, a provider of artificial intelligence and machine learning to the financial services industry. Black Knight explained it plans on integrating HeavyWater's AIVA solution, which leverages artificial intelligence and machine learning to perform operational functions more efficiently than traditional methods, into its premier solutions. It will also make the technology available to its clients looking to utilize AI within other parts of their organizations. "With the cost of origination and servicing at, or near, all-time highs, AIVA is poised to help increase efficiencies for Black Knight clients," Black Knight CEO Anthony Jabbour said. "AI, machine learning and neural network solutions are the future of delivering enhanced productivity and capabilities to our clients, and we are very excited about the potential HeavyWater has to offer."

Black Knight Inc., a Jacksonville-based firm with a $1 billion in annual sales analyzing mortgages for large banks, says it has agreed to purchase HeavyWater, a University City Science Center-based, two-year-old firm that is developing AIVA ("Ava,") an "artificial intelligence and machine learning" mortgage lending system to speed loan approvals, from founder Soofi Safavi. Safavi will remain in charge of the 14-person group, which he said plans to hire 5 more people by June 19. Black Knight said in a statement that it will use and improve AIVA -- the Artificial Intelligence Virtual Assistant -- to "read, comprehend and draw conclusions based on context to mimic cognitive thinking and build expertise," and that it will also make the technology available to clients as a way to speed mortgage approvals and boost accuracy in verifying income, assets and insurance coverage. "Artificial intelligence, machine learning and neural network solutions are the future," and Heavy Water will help make that happen for Black Knight clients, said that company's CEO, Anthony Jabbour, in a statement. Safavi served as chief technology officer at Radian, the Philadelphia mortgage insurer, before starting HeavyWater in September of 2015.