French AI Startup Raises $5.5M to Boost Team Productivity With Large Language Models

Dust, a new AI startup based in France, has successfully raised $5.5 million in a seed funding round with the goal of enhancing team productivity through the utilization of large language models (LLMs).

Founded by Gabriel Hubert and Stanislas Polu, Dust is dedicated to leveraging LLMs on internal company data to empower team members with new capabilities and overcome internal barriers, according to a report by TechCrunch.

How Dust Came to Be

The entrepreneurial journey of Hubert and Polu began with their previous startup, Totems, which was later acquired by Stripe in 2015. Following the acquisition, both co-founders spent several years working for Stripe before eventually embarking on separate paths.

Polu joined OpenAI, where he focused on enhancing the reasoning capabilities of LLMs, while Hubert assumed the role of head of product at Alan.

Reuniting once again, Hubert and Polu founded Dust with a distinctive approach. It is worth noting that Dust does not primarily aim to develop new large language models.

Instead, the company intends to build applications on top of existing LLMs developed by prominent entities such as OpenAI, Cohere, and AI21.

Initially, the Dust team focused on creating a versatile platform for designing and deploying large language model apps.

However, their efforts soon changed towards a specific use case: centralizing and indexing internal data to enable LLMs to effectively utilize it. To achieve this, Dust employs connectors that continuously retrieve data from various internal platforms like Notion, Slack, Github, and Google Drive.

The retrieved data is then indexed, facilitating semantic search queries. When a user interacts with a Dust-powered app, the relevant internal data is retrieved and utilized as context for LLMs to generate informative responses.

Dust's capabilities go beyond traditional internal search tools. It not only retrieves search results but also extracts information from various data sources and presents it in a more user-friendly format.

Beyond Serving as an Internal ChatGPT

The potential applications of Dust are not limited to serving as an internal ChatGPT; it also serves as the foundation for developing customized internal tools.

To bring their vision to life, Dust is collaborating with design partners to explore various implementation and packaging strategies for their platform. Stanislas Polu expressed their belief that numerous products could be created in the enterprise data and knowledge worker domain with the support of models like Dust.

While Dust is still in its early stages, the startup is actively addressing challenges related to data retention, LLM-related issues like hallucination, and other associated complexities.

As LLM technology evolves, the issue of hallucination may diminish. Additionally, Dust may eventually develop its own LLMs to ensure data privacy and security.

In the recently concluded seed funding round, Dust secured $5.5 million (€5 million), with Sequoia leading the investment. Other notable participants included XYZ, GG1, Seedcamp, Connect, Motier Ventures, Tiny Supercomputer, AI Grant, and a group of esteemed business angels.

The roster of business angels includes industry figures such as Olivier Pomel from Datadog, Julien Codorniou and Julien Chaumond from Hugging Face, Mathilde Collin from Front, Charles Gorintin and Jean-Charles Samuelian-Werve from Alan, Eléonore Crespo and Romain Niccoli from Pigment, Nicolas Brusson from BlaBlaCar, Howie Liu from Airtable, Matthieu Rouif from PhotoRoom, and Igor Babuschkin and Irwan Bello, according to TechCrunch's report.

Dust firmly believes that LLMs have the potential to fundamentally transform how companies operate. The product's effectiveness is further amplified in organizations that prioritize transparency, written communication, and autonomy rather than information retention, endless meetings, and top-down management.