Sama, the leader in providing data annotation and model validation solutions announced Sama GenAI, its specialized solution that has already helped power some of the most well-known Generative AI (GenAI) and foundation models in the world. Unlike other solutions on the market, Sama GenAI leverages SamaIQ™, a combination of proprietary algorithms and an expert workforce, to provide ongoing insights into data and model predictions, enabling faster development of high-performance models.
Recommended: NovelVox Introduces Intuitive Visual IVR for Cisco, Avaya, and Genesys Contact Centers
“GenAI is extremely confident in what it is creating — even when it shouldn’t be. Model hallucination, where it creates text or images that don’t make sense or are simply factually incorrect, is a real and documented problem across all GenAI models”
Sama’s proprietary approach is designed to keep foundation models, including GenAI models, constantly learning, beginning in proof-of-concept and training to post-deployment. SamaIQ can also help identify biases in a model’s original dataset or better understand how they are being created by data. In addition, SamaIQ can prioritize and annotate data that will have the most impact on performance to speed up the development process. Finally, the insights surfaced by SamaIQ can then be used to create more training data to directly address identified issues, ultimately building a more accurate model. Each time data is processed by Sama, the model improves — creating a flywheel effect that significantly reduces the chances of failure, including catastrophic failure. To date, Sama GenAI has already surfaced potential biases with its customers, including major foundation models that have been globally adopted.
“GenAI is extremely confident in what it is creating — even when it shouldn’t be. Model hallucination, where it creates text or images that don’t make sense or are simply factually incorrect, is a real and documented problem across all GenAI models,” said Duncan Curtis, SVP of Product at Sama. “Sama’s approach, using both technological and human insights, allows us to identify when models are too confident, then provide additional training data to adjust that confidence level accordingly and ultimately help them perform better. We are proud to be supporting this quickly-growing, innovative field of AI now and in the future.”
Sama is also working with companies to build GenAI models specifically designed to generate synthetic data, which can be used in a variety of applications ranging from autonomous driving and precision agriculture to manufacturing and robotics. For example, synthetically-generated images and videos of humans can be used to reduce bias from imbalanced data sets in models while protecting privacy. This synthetic data can augment model training data to ensure there is adequate representation across age, gender, skin tone and more to improve model accuracy and precision.
AiThority Interview Insights : Balancing AI and Human Insight in Content Marketing
Sama is currently working with Synthesis AI, a company that builds synthetic images and video for computer vision and perception AI applications. They use a unique combination of generative AI, procedural generation and cinematic VFX rendering systems to build a diverse set of photorealistic images and video.
“We have engaged Sama to help validate our generative AI models by verifying the accuracy of ethnicity, age and gender of our digital humans,” said Yashar Behzadi, CEO of Synthesis AI. “These insights will help increase model performance and enable us to generate synthetic images and videos faster and more accurately.”
Foundation models have become useful across multiple enterprise use cases aside from synthetic data generation. These include customer support and experience, automation of repetitive tasks, rapid generation of content or code and more. However, as more use cases emerge, different datasets are required to properly introduce models to the use case and ensure that it performs correctly. Sama’s Platform 2.0 and Human in-the Loop (HITL) approach quickly surfaces model prediction gaps, then works with the client to find the correct data to train the model accordingly.
In its other work, the combination of Sama’s Platform 2.0 and in-house, fully trained workforce has led to reductions of up to 30% in the total cost of ownership for getting models into production and tripling or even quadrupling the speed at which a model can be brought to market. Platform 2.0 has also successfully achieved a 99% client acceptance rate for AI training data through SamaAssure, the industry’s highest quality guarantee, with an annotation delivery rate of 300+ million frames, 850+ million shapes and 10 billion annotation points a month. Sama achieves all of this while also committing to doing good for underserved communities and the planet. Its work has led the company to achieve B Corp certification, one of the first AI companies to do so.
Recommended: AiThority Interview with Keri Olson, VP at IBM IT Automation
[To share your insights with us, please write to sghosh@martechseries.com]
Comments are closed.