#United States #San Jose #AI Models #Realset AI #Flatkey

Realset AI and Flatkey's Bold Move to Revolutionize Data for AI

In a significant stride for artificial intelligence research, Realset AI, a pioneering data lab, has announced that it has raised $10 million in Series A funding, alongside its partner, Flatkey. This influx of capital is destined for the enhancement of training data development aimed at large language models (LLMs), frontier models, and embodied agents.

The Importance of Real-World Data

With the digital landscape becoming increasingly saturated, the need for real-world data is paramount. The conventional repositories, mainly textual data from the internet, are reaching their limits. Realset AI aims to address this by focusing on capturing real human actions in authentic settings, such as homes and workplaces, offering valuable insights into practical tasks.

What Sets Realset Apart

Realset AI has devised a unique strategy that involves the collection of valuable data through expert demonstrators rather than generic crowd contributions. They opt for actual environments over simulations, ensuring that the data produced carries immense relevance and accuracy. Each data point collected is meticulously curated, with rigorous quality assessments at each stage of processing.

Innovative Data Production Techniques

Realset AI is focusing on three principal methods to generate reliable and actionable training data:

1. Realset Body: This method involves generating embodied data that supports physical policy models. It captures real-time actions of expert workers performing specific tasks in various environments—be it a kitchen or a warehouse. This data is enriched with extensive annotations tailored for vision-language-action training.

2. Realset Field: By simulating real-world workflows, this approach generates LLM and agent training data. It mirrors diverse operations in e-commerce, customer service, and logistics, ensuring that the data reflects authentic business outcomes and practices, complete with bilingual support for greater applicability.

3. Realset Judge: This component is focused on evaluating agents in practical settings, overseen by professionals who understand the intricacies of the tasks being automated. This evaluation is critical to continuously improve AI agents' performance, ensuring they can adapt to the dynamic nature of human labor.

A Commitment to Transparency and Quality

Understanding the importance of quality in data collection, Realset operates a comprehensive workspace where domain experts contribute their expertise. Each data record undergoes independent scrutiny to guarantee high standards, making every dataset verifiable and reliable for AI developers.

Open Benchmarks for Practical Applications

Realset AI is not only focused on collecting data but also on providing benchmarks that assess AI policies beyond theoretical parameters. They are set to introduce open benchmarks to evaluate the efficacy of AI models on real household tasks by Q4 2026, allowing researchers and developers to measure the practical impacts of their systems.

Future Vision

Hunter Guo, the founder of Realset AI, emphasizes the essence of capturing real-world actions, stating, "The easy data is gone. What is left is the physical world, and you cannot scrape it." The vision for Realset is cemented in building a company around the complexities of capturing, structuring, and quality-checking real human work.

About Realset AI and Flatkey

Realset AI is at the forefront of creating realistic training data for various applications, including robotics and AI solutions for businesses. Based in San Jose, California, they continue to innovate in the collection and evaluation of practical data.
Flatkey complements this mission as a robust AI infrastructure platform, providing easy access to numerous AI models and tools to developers.

As AI technologies evolve, initiatives like Realset's will be crucial in ensuring that future machines can effectively interact with and support human tasks in the real world.

For more information, visit Realset AI and Flatkey.