The proprietary training solution builds on Sama’s commitment to excellence and an industry-leading 99% client acceptance rate by reducing project ramp times by up to 50% while increasing individual annotators’ tag accuracy by 16% and shape accuracy by 15% on average. For Sama’s enterprise clients, this results in higher-quality models going into production faster, saving both time and capital. For Sama employees, this new platform improves the training experience, offers greater understanding of data annotation and AI development principles, and builds their skills for successful long-term careers in the digital economy.
.gif)
“Basic data annotation training is just that: basic. It encourages rote memorization instead of truly learning the ins and outs of correct annotation. At Sama, we have always believed in the power of project-specific training to increase quality and reduce rework. This new iteration of our training platform takes that a step further — it’s more scalable and well-suited for even the most complex tasks, including automotive and Generative AI data annotation, even when client parameters may change mid-project,” said Duncan Curtis, SVP of AI product and technology at Sama. “Our annotators can now actively learn at their preferred pace and receive useful feedback for fuller comprehension. These sessions will help them not only excel on current AI projects, but build and master new skills, which will prepare them for future AI innovations and development needs.”
Training for complex tasks, such as annotating LiDAR data or complex sensor fusion data, previously required lengthy courses and, consequently, a significant amount of time to provide detailed feedback for a trainee to master the skills. Sama’s productized training ties into its responsible AI framework by emphasizing data annotation work’s role as a stepping stone. By building a talent pipeline that is actively learning and mastering concepts, Sama is investing in its own workforce. That same talent pipeline, primarily consisting of women and underrepresented communities, allows AI developers to more easily access a broader range of perspectives about how AI should be developed and what needs to be corrected, promoting more responsible and ethical models overall.
“We envision Kenya as a growing hub for new AI innovations and talent, able to reap the economic benefits of AI while ensuring that models are developed with diverse perspectives. For this vision to become reality, digital skilling is a must. This latest development is a clear signal of Sama’s deep commitment and investment in local talent,” said Maxwell Okello, CEO of America Chamber of Commerce (AmCham) Kenya.
The new training platform begins with annotation tasks that have gold answers, which the customer or trainer has verified. During training, Sama’s AutoQA™ platform autonomously compares an annotator’s answers to these ground truth responses and can offer specific instruction on where to improve. If an annotator feels stuck, they also have access to hints, such as briefly showing the correct shapes. They can track their progress and others’ to see their advances in real time. Early results have yielded a 15% increase in shape accuracy and a 16% increase in tag accuracy compared to previous Sama training modules, reducing the odds of delays caused by rework. Project ramp time (the time from when a contract is signed to when work on a project begins) has been reduced by up to 50% with these new features.
In addition, the platform has built-in flexibility to adjust to changing client needs. When instructions or criteria change during the middle of a project, Sama can update training instructions and easily deploy re-training modules to the entire workforce. This allows for a smooth transition to follow the new criteria and can reduce rework.
This new solution joins a suite of products designed to scale to all project sizes, including some of the largest open-source models in the world. Sama employs a human in the loop (HITL) approach to constantly and consistently provide models with feedback from expert annotators, validating a model’s behavior and ensuring it is performing to standards. This feedback occurs during the entire model development process, including data creation, supervised fine-tuning, LLM optimization and ongoing model evaluation, ensuring clients can develop models in a more responsible way.
Sama’s work is backed by SamaAssure™, the industry’s highest quality guarantee, which routinely delivers a 98% first batch acceptance rate.
Sama has been named the Best AI Model Validation Solution by the annual AI Breakthrough Awards, which recognize the breakthrough companies, technologies, products and services in the field of AI around the world.
Sama provides ML Professionals and AI team Leads with an indispensable solution for Computer Vision data labeling.
