Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Singapore startup Sapient enters global enterprise AI race with new model architectures

Dec 10, 2024 - venturebeat.com
Sapient Intelligence, Singapore's first foundation model AI startup, has raised $22 million in seed funding, aiming to address limitations in GPT-style models by developing new architectures inspired by neuroscience and mathematics. These models blend transformer components with recurrent neural network structures, enabling them to tackle complex, multi-step reasoning tasks with greater precision and reliability. Sapient's innovations have shown promising results in benchmarks like Sudoku, where their model achieved 95% accuracy without intermediate tools, and in tasks such as two-dimensional navigation and complex mathematical problem-solving.

The company focuses on practical applications, starting with enterprise coding and robotics, deploying autonomous AI coding agents in environments like Sumitomo's to maintain and contribute to codebases. Sapient aims to offer similar services to other enterprises, positioning its AI agents as autonomous, smart employees. The startup also advances embodied AI for real-time interaction and adaptation in robots. With a global vision, Sapient is building an AI research lab in Singapore and planning for the Bay Area, emphasizing diversity and strong partnerships with Japanese investors. Its long-term goal is to create a generalized agent for personal and enterprise use, with future public-facing products like autonomous coding agents and personal assistants.

Key takeaways:

```html
  • Sapient Intelligence, a Singapore-based AI startup, raised $22 million in seed funding to develop new foundational model architectures for complex reasoning tasks.
  • The company is creating a novel model architecture that combines transformer components with recurrent neural networks, inspired by neuroscience and mathematics.
  • Sapient's models excel in benchmarks like Sudoku and complex problem-solving, requiring only question-and-answer pairs for training, reducing dependency on curated datasets.
  • The startup aims to deploy autonomous coding agents and embodied AI for real-world applications, with a focus on enterprise coding and robotics.
```
View Full Article

Comments (0)

Be the first to comment!