Scale AI Inc., a startup that helps OpenAI and Nvidia Corp. train their artificial intelligence models, has received a $13.8 billion valuation in a late-stage $1 billion funding round announced today.

Accel was the lead investor in the Series F round. Nvidia took part as well along with Intel Corp. and Advanced Micro Devices Inc., which contributed through their venture capital arms. The chipmakers were joined by more than a dozen other backers including Amazon.com Inc., Meta Platforms Inc., Cisco Investments and ServiceNow Ventures.

Advanced AI models are trained on large datasets that can be prohibitively time-consuming for machine learning teams to assemble on their own. Scale AI provides a cloud service, the Generative AI Data Engine, that promises to speed up the process. It allows developers to specify what training data they require for an AI project and have a team of experts assemble the necessary files on their behalf.

One of the tasks that the platform promises to speed up is data labeling. That’s the process of enriching the individual files in a training dataset with contextual details, such as a text snippet that describes what a given file contains. Such context enables AI models to more thoroughly understand the information on which they’re being trained, which in turn allows them to output higher-quality responses.

Scale AI’s platform can be used to annotate several types of data. It supports text, images and audio, as well as more specialized files such as sensor logs from self-driving cars’ lidar hardware. A graphical interface enables machine learning teams to provide the experts who annotate their data with instructions on how the process should be carried out.

“Our vision is one of data abundance, where we have the means of production to continue scaling frontier LLMs many more orders of magnitude,” said Scale AI founder and Chief Executive Officer Alexandr Wang. “We should not be data-constrained in getting to GPT-10.”

Scale AI also promises to ease a number of related tasks for AI developers. It provides a service that can be used to test newly developed neural networks for vulnerabilities, hallucinations and other issues. It says it uses a combination of automated and manual approaches to detect weak points in customers’ AI applications.

For public sector organizations, the company has developed a chatbot called Donovan. It’s capable of answering questions about sensitive files such as intelligence reports. Scale AI says the chatbot can double as a translation tool, as well as help developers with coding tasks.

Under the hood, Donovan is powered by large language models from commercial AI providers such as OpenAI and the open-source ecosystem. According to Scale AI, the chatbot configures those LLMs to include citations in prompt responses. Customers can optionally fine-tune the models by supplying them with information annotated using Scale AI’s flagship training data platform.

Scale AI said that it will use the $1 billion round announced today to make training data more accessible for machine learning teams. Additionally, a portion of the capital will go towards initiatives focused on evaluating AI risks. 

Image: Unsplash