Themis AI announces the public launch of its Capsa platform. A much-needed addition to the AI landscape, Capsa allows any machine learning engineer to de-risk their AI model with just a few lines of code. After rigorous and extensive pilot testing the company looks forward to sharing their technology with a broader range of enterprises and teams pushing the boundaries of innovation.
“With Capsa, we are building on years of MIT research to address pressing concerns about AI safety via an innovative, unique, and extremely efficient approach. Our leading team of machine learning scientists and engineers has developed a powerful product platform that can provide meaningful value to any business using an AI model,” said Dr. Stewart C. Jamieson, Head of Technology at Themis AI. “We are now ready to put Capsa in the hands of every machine learning professional and every company with the goal of responsibly-deployed AI.”
Themis AI aims to tackle some of the most critical safety concerns facing the AI industry: hallucinations and unreliable outputs. Driven by a team of machine learning experts, Themis AI created Capsa to allow software developers across the globe to address these issues and de-risk AI model outputs. Capsa is able to surgically update the architecture of any AI model within seconds so that it can quantify its own uncertainty. With Capsa, any AI model can flag its own mistakes and hallucinations, unlocking AI’s potential across new industries while enabling quality assurance. Capsa makes developing safe AI models significantly faster and more efficient, and facilitates a meaningful reduction in compute costs in a world where AI chips and computing resources are scarce and expensive.
Also Read: GROQ Secures $640M to Boost Fast AI Inference Demand
“By automatically quantifying aleatoric and epistemic uncertainty, Capsa is a transformative technology that enables model errors to be caught before they become costly mistakes. It can be applied to any AI model in any ML framework, and expands the uses of AI systems in applications where safety and reliability are key, such as robotics and autonomous driving,” said Dr. Daniela Rus, Co-Founder of Themis AI and Director of MIT Computer Science and Artificial Intelligence Lab (CSAIL).
Capsa’s initial release supports all AI models developed in PyTorch, TensorFlow, and Keras, with JAX support forthcoming. It provides a simple Python interface for augmenting models to quantify the aleatoric and epistemic uncertainties of their own outputs. Aleatoric uncertainty measures ambiguity or risk inherent to the input data; for example, in classifying a blurry photo or generating a long-term weather forecast. Epistemic uncertainty instead measures model deficits resulting from incomplete or biased training datasets, which lead to issues like hallucinations in large language models. Capsa is the culmination of years of research into these kinds of uncertainties and developing effective, efficient methods to quantify them across every kind of AI model.
“Capsa brings together state-of-the-art uncertainty quantification abilities into an easy-to-use software tool—empowering enterprises of all sizes to develop uncertainty-aware AI systems. We look forward to a future where all AI systems leverage Capsa’s next-generation tools to deploy AI confidently,” said Dr. Alexander Amini, Co-Founder of Themis AI and MIT PhD.
Source: PRWeb
