AI startup Humans& Inc., established just three months ago, has announced a $480 million seed funding round, valuing the company at $4.48 billion.
The investment was led by SV Angel and company co-founder Georges Harik, an early Google employee involved in developing several core services. Google's parent company, Alphabet Inc., participated through its GV fund. Other investors include Nvidia, Jeff Bezos, and over six additional backers.
Humans&'s founding team comprises approximately 20 AI experts from OpenAI Group PBC, Anthropic PBC, Meta Platforms Inc., and other key players in the AI market. The company states its researchers are developing neural networks aimed at boosting worker productivity.
According to The New York Times, the company's models are designed to accelerate collaboration-related tasks and online research. They are also expected to automate "other tasks suited for machines."
In a blog post, the company hinted that one of its goals is to equip its AI software with the capability to execute long-duration activities. These are complex tasks that would typically require large language models to operate for hours or longer.
Long-duration processing is a key focus for machine learning researchers. For instance, Google recently disclosed an AI model architecture specifically optimized for long-running use cases. A core feature of this architecture is a component called a meta-controller, which generates software modules to optimize the model's inference workflow when assigned a lengthy task.
Humans& also plans to support multi-agent use cases for its software. This means its AI models will be able to collaborate with other neural networks to accomplish multi-step tasks. Furthermore, the company's models will proactively ask workers for the information needed to complete specific assignments.
The company intends to use reinforcement learning to train its algorithms—a common method for developing reasoning models. In such projects, an AI completes a set of training tasks and receives positive or negative feedback based on its performance, which the algorithm uses to improve its output quality.
Unlike some other development approaches, reinforcement learning does not require researchers to enrich training data with explanatory labels, thereby reducing costs. However, it can still be capital-intensive, as training reasoning models often demands significant GPU resources. This may be one reason Humans& sought such a substantial seed funding round.
The New York Times reports that the company plans to launch its first product earlier this year.
