Tasq Technologies Ltd., a startup that provides “human guidance” for data collection, annotation and management tasks essential to building artificial intelligence models, today announced the availability of a new service that can aid in the building and fine-tuning of generative AI models at scale.

Tasq.ai says its platform reduces the time it takes to create AI models by centralizing the entire data process, including data collection, labeling and validation, and by providing companies with access to qualified experts that can help them fine-tune models to improve their performance.

Human guidance for AI refers to using human input to improve the performance, accuracy and fairness of AI systems. That can be done through labeled data, feedback, and evaluation, but it requires careful consideration of potential biases and fairness issues that only experts can perform.

Generative AI models such as OpenAI LP’s ChatGPT and Google LLC’s Bard are trained on massive amounts of data and can generate humanlike responses to almost any kind of question or prompt. The launch of ChatGPT last year ignited big interest in the technology, which has applications in areas such as search, coding, writing, drug discovery and more.

However, Tasq.ai says, most companies struggle to take advantage of generative AI because of the enormous size of these foundation models. They’re extremely difficult to create in the first place, hosting them is expensive, and even the seemingly straightforward task of tuning an off-the-shelf model like ChatGPT for a specific use case can be very complicated, with many firms struggling to get effective results.

Tasq.ai co-founder and Chief Revenue Officer Nati Catalan told SiliconANGLE that foundation models can be rather shallow when it comes to performing specialized tasks. “For example, they can draw up a very basic nondisclosure agreement, but they can’t assess the merit of a medical malpractice claim in a particular state with great accuracy,” he said. “For specific businesses, it can’t integrate their inventory data, CRM insights, or help to schedule local contractors.”

The difficulty with fine-tuning generative AI models is that companies must perform complex tasks such as data labeling, validation and ranking in an unbiased way, Catalan said. Traditionally, most companies will outsource this work, but doing so produces varying results, he said, because each task must be outsourced to a different provider.

Tasq.ai provides another option, providing access to thousands of highly qualified experts known as “tasqers” that can help companies to complete the various tasks required to build unbiased and accurate AI models. “Tasq.ai is the first solution that centralizes the entire process,” Catalan said. “All the data labeling, validation, ranking and refining happens in one place, saving weeks or months and money and ensuring that the results are more accurate and transparent.”

Catalan explained that Tasq.ai’s tasqers are incentivized, prequalified individuals from around the globe who have been filtered by their cognitive capabilities to excel in online tasks per demand.

“Our tasqers can be accessed flexibly when companies need them and they are from diverse backgrounds, which helps eliminate bias,” Catalan said. “The projects they are paired with are also based on their particular profiles, vis-a-vis performance and background. One of Tasq.ai platform’s unique features is to disassemble the complex labeling requirements into workflows of smaller simplified tasks that increase the chances for tasqers to succeed in.”

Tasq.ai Chief Executive Erez Moscovich said companies can benefit from a significant first-mover advantage if they embrace generative AI quickly. With human guidance, companies can adapt generative AI models to multiple use cases while ensuring the very best quality of those services. “It’s a game changer that finally allows companies to overcome the biggest bottleneck to unlocking the value that Generative AI creates,” Moscovich said.

Photo: Tasq.ai